Merge PR #1827: AI数字人v3 完整联调版（TTS直生+smart-cover+前后端冲突解决）(#1827)

4efa71ec 合并时 tests/unit/test_lipsync_routes.py 误取了 develop #1809 版本
（synthesize_speech/route._resolve_voice_id/断言 audio_url 移除），与合并后
v3 #1822 双模式实现不一致，导致 7 个单测失败。改用后端解冲突分支
backend/ai-avatar-merge-1827 (0e994072) 中已适配 v3 的版本：
mock submit_synthesize_task、emotion 兴奋→excited、双模式 audio_url 保留。
本地验证该文件 28 passed、lipsync/avatar/tts 相关 79 passed。

.gitea/workflows/ci-pipeline.yml

@@ -20,6 +20,7 @@ on:
         default: "手动触发 - CI漏触发补跑"
 permissions:
   contents: read
+  pull-requests: read
 concurrency:
   group: ci-pipeline-${{ gitea.ref }}
   cancel-in-progress: true
@@ -88,9 +89,22 @@ jobs:
         GITHUB_TOKEN: ${{ github.token }}
       run: |
         set -eu
+        # 优先用 git diff 判断 PR 改动范围（比 API 稳定）
         PR_NUMBER=$(echo "$GITHUB_REF" | sed 's|refs/pull/||; s|/.*||')
-        API_URL="${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}/files?limit=300"
-        FILES=$(curl -s -H "Authorization: token ${GITHUB_TOKEN}" "$API_URL" | python3 -c "import sys,json; [print(f['filename']) for f in json.load(sys.stdin)]")
+        if command -v git >/dev/null 2>&1 && [ -d .git ]; then
+          FILES=$(git diff --name-only origin/develop...HEAD 2>/dev/null || true)
+        fi
+        if [ -z "${FILES:-}" ]; then
+          # fallback 到 API
+          API_URL="${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}/files?limit=300"
+          FILES=$(curl -sf -H "Authorization: token ${GITHUB_TOKEN}" "$API_URL" | python3 -c "import sys,json; [print(f['filename']) for f in json.load(sys.stdin)]" 2>/dev/null || true)
+        fi
+        if [ -z "${FILES:-}" ]; then
+          echo "⚠️ 无法获取变更文件列表，保守运行完整 CI"
+          echo "skip_backend=false" >> $GITHUB_OUTPUT
+          echo "skip_frontend=false" >> $GITHUB_OUTPUT
+          exit 0
+        fi
         FRONTEND_COUNT=$(echo "$FILES" | grep -c '^apps/web/' || true)
         BACKEND_COUNT=$(echo "$FILES" | grep -cv '^apps/web/' || true)
         TOTAL=$(echo "$FILES" | grep -cv '^$' || true)
@@ -1187,6 +1201,8 @@ jobs:
         COSYVOICE_API_KEY: ${{ secrets.COSYVOICE_API_KEY }}
         DASHSCOPE_API_KEY: ${{ secrets.DASHSCOPE_API_KEY }}
         MEDIAKIT_API_KEY: ${{ secrets.MEDIAKIT_API_KEY }}
+        WECHAT_APP_ID: ${{ secrets.WECHAT_APP_ID }}
+        WECHAT_APP_SECRET: ${{ secrets.WECHAT_APP_SECRET }}
       run: |
         set -eu
         echo "Rendering .env from template + secrets..."

alembic/versions/066_asset_client_upload_id_ingest_job_asset_id.py

@@ -0,0 +1,34 @@
+"""add client_upload_id to assets and asset_id to ingest_jobs
+
+Issue #1714：上传 complete 幂等 + worker 转码回写关联。
+- assets.client_upload_id：客户端幂等 token（complete 去重）
+- ingest_jobs.asset_id：complete 阶段创建的占位 asset id（worker 回写关联，
+  防止 HEVC 转码改写 storage_key 后找不到占位而兜底新建 READY 记录）
+
+Revision ID: 066_upload_idempotency
+Revises: 065_dup_record_sim_match
+Create Date: 2026-09-05
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "066_upload_idempotency"
+down_revision = "065_dup_record_sim_match"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("assets", sa.Column("client_upload_id", sa.String(64), nullable=True))
+    op.create_index("ix_assets_client_upload_id", "assets", ["client_upload_id"])
+    op.add_column("ingest_jobs", sa.Column("asset_id", sa.String(36), nullable=False, server_default=""))
+    op.create_index("ix_ingest_jobs_asset_id", "ingest_jobs", ["asset_id"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_ingest_jobs_asset_id", table_name="ingest_jobs")
+    op.drop_column("ingest_jobs", "asset_id")
+    op.drop_index("ix_assets_client_upload_id", table_name="assets")
+    op.drop_column("assets", "client_upload_id")

alembic/versions/067_celery_task_id_revoke.py

@@ -0,0 +1,35 @@
+"""add celery_task_id to generation_tasks and ingest_jobs
+
+Issue #1714：孤儿恢复/超时清理撤销队列消息。
+- generation_tasks.celery_task_id：入队时记录的 Celery 消息 ID，清理时 revoke
+- ingest_jobs.celery_task_id：同上（素材转码任务）
+
+Revision ID: 067_celery_task_id
+Revises: 066_upload_idempotency
+Create Date: 2026-09-05
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "067_celery_task_id"
+down_revision = "066_upload_idempotency"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "generation_tasks",
+        sa.Column("celery_task_id", sa.String(64), nullable=False, server_default=""),
+    )
+    op.add_column(
+        "ingest_jobs",
+        sa.Column("celery_task_id", sa.String(64), nullable=False, server_default=""),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("ingest_jobs", "celery_task_id")
+    op.drop_column("generation_tasks", "celery_task_id")

alembic/versions/068_user_profile_completed.py

@@ -0,0 +1,26 @@
+"""add profile_completed to users
+
+Issue #1718：微信新用户首次登录需设置昵称（PATCH /auth/me）。
+- users.profile_completed：资料是否已完善；存量行默认 True（不触发引导），
+  微信新建用户在应用层置 False。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "068_user_profile_completed"
+down_revision = "067_celery_task_id"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "users",
+        sa.Column("profile_completed", sa.Boolean(), nullable=False, server_default=sa.text("true")),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("users", "profile_completed")

alembic/versions/069_project_is_default_idempotent.py

@@ -0,0 +1,72 @@
+"""Projects is_default + partial unique index for idempotent default project (Issue #1775)
+
+Revision ID: 069_project_is_default
+Revises: 068_user_profile_completed
+Create Date: 2026-09-08
+
+背景：
+小程序端 getOrCreateDefaultProject 在重试/并发/前端重复调用下，
+仅靠应用层"先查再插"不保证幂等，会给同一用户重复创建默认项目。
+
+改动：
+1. projects 表新增 is_default 布尔列（默认 false）
+2. 部分唯一索引 uq_projects_owner_default：(owner_user_id) WHERE is_default = true
+   —— 保证每个用户至多一个默认项目
+3. 存量数据回填：把名为"默认项目"的存量项目按创建时间最早者标记为 is_default=true
+   （只标记不删除；存量重复项目的清理另行确认后单独执行）
+
+注意：部分唯一索引依赖 PostgreSQL，不支持 downgrade 到其他方言。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "069_project_is_default"
+down_revision = "068_user_profile_completed"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 1. 新增 is_default 列
+    op.add_column(
+        "projects",
+        sa.Column(
+            "is_default",
+            sa.Boolean(),
+            nullable=False,
+            server_default=sa.text("false"),
+        ),
+    )
+
+    # 2. 存量回填：每个拥有"默认项目"的用户，只把最早创建的那一个标记为默认。
+    #    用 ROW_NUMBER() 取每组第一条；非"默认项目"命名的项目不标记（保守，不动用户自建项目）。
+    op.execute("""
+        UPDATE projects p
+        SET is_default = true
+        WHERE p.id IN (
+            SELECT id FROM (
+                SELECT id,
+                       ROW_NUMBER() OVER (
+                           PARTITION BY owner_user_id
+                           ORDER BY created_at ASC, id ASC
+                       ) AS rn
+                FROM projects
+                WHERE name = '默认项目'
+            ) t
+            WHERE t.rn = 1
+        )
+        """)
+
+    # 3. 部分唯一索引：每用户至多一个默认项目（只约束 is_default = true 的行）
+    op.execute("""
+        CREATE UNIQUE INDEX uq_projects_owner_default
+        ON projects (owner_user_id)
+        WHERE is_default = true
+        """)
+
+
+def downgrade() -> None:
+    op.execute("DROP INDEX IF EXISTS uq_projects_owner_default")
+    op.drop_column("projects", "is_default")

alembic/versions/070_add_scripts_table.py

@@ -0,0 +1,48 @@
+"""Add scripts table for oral broadcast script library (Issue #1795)
+
+Revision ID: 070_add_scripts
+Revises: 069_project_is_default
+Create Date: 2026-09-08
+
+新建 scripts 表，支持口播文案 CRUD + 分段存储。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "070_add_scripts"
+down_revision = "069_project_is_default"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "scripts",
+        sa.Column("id", sa.String(36), nullable=False),
+        sa.Column("user_id", sa.String(36), nullable=False),
+        sa.Column("title", sa.String(255), nullable=False),
+        sa.Column("content", sa.Text(), nullable=False, server_default=""),
+        sa.Column("segments", sa.JSON(), nullable=False, server_default="[]"),
+        sa.Column("tags", sa.JSON(), nullable=False, server_default="[]"),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            nullable=False,
+            server_default=sa.func.now(),
+        ),
+        sa.Column(
+            "updated_at",
+            sa.DateTime(timezone=True),
+            nullable=False,
+            server_default=sa.func.now(),
+        ),
+        sa.PrimaryKeyConstraint("id"),
+    )
+    op.create_index("ix_scripts_user_id", "scripts", ["user_id"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_scripts_user_id", table_name="scripts")
+    op.drop_table("scripts")

alembic/versions/071_add_lipsync_jobs_table.py

@@ -0,0 +1,47 @@
+"""add lipsync jobs table
+
+Revision ID: 071_add_lipsync_jobs
+Revises: 070_add_scripts
+Create Date: 2026-09-08
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "071_add_lipsync_jobs"
+down_revision = "070_add_scripts"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "lipsync_jobs",
+        sa.Column("id", sa.String(36), primary_key=True),
+        sa.Column("user_id", sa.String(36), nullable=False, index=True),
+        sa.Column("project_id", sa.String(36), nullable=False, server_default=""),
+        sa.Column("video_url", sa.Text(), nullable=False),
+        sa.Column("audio_url", sa.Text(), nullable=False),
+        sa.Column("enable_video_loop", sa.Boolean(), nullable=False, server_default=sa.text("false")),
+        sa.Column("mediakit_task_id", sa.String(200), nullable=False, server_default="", index=True),
+        sa.Column("status", sa.String(20), nullable=False, server_default="pending", index=True),
+        sa.Column("output_video_url", sa.Text(), nullable=False, server_default=""),
+        sa.Column("output_duration", sa.Float(), nullable=False, server_default=sa.text("0.0")),
+        sa.Column("error_message", sa.Text(), nullable=False, server_default=""),
+        sa.Column("error_code", sa.String(100), nullable=False, server_default=""),
+        sa.Column("submitted_at", sa.DateTime(), nullable=True),
+        sa.Column("completed_at", sa.DateTime(), nullable=True),
+        sa.Column("created_at", sa.DateTime(), nullable=False, server_default=sa.func.now()),
+        sa.Column("updated_at", sa.DateTime(), nullable=False, server_default=sa.func.now()),
+    )
+    # 复合索引：用户 + 状态（列表查询常用）
+    op.create_index("ix_lipsync_jobs_user_status", "lipsync_jobs", ["user_id", "status"])
+    # 项目 + 用户（项目维度查询）
+    op.create_index("ix_lipsync_jobs_project_user", "lipsync_jobs", ["project_id", "user_id"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_lipsync_jobs_project_user", table_name="lipsync_jobs")
+    op.drop_index("ix_lipsync_jobs_user_status", table_name="lipsync_jobs")
+    op.drop_table("lipsync_jobs")

alembic/versions/072_add_ai_avatar_render_jobs_table.py

@@ -0,0 +1,48 @@
+"""add ai avatar render jobs table
+
+Revision ID: 072_add_ai_avatar_render
+Revises: 071_add_lipsync_jobs
+Create Date: 2026-09-09
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "072_add_ai_avatar_render"
+down_revision = "071_add_lipsync_jobs"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "ai_avatar_render_jobs",
+        sa.Column("id", sa.String(36), primary_key=True),
+        sa.Column("user_id", sa.String(36), nullable=False, index=True),
+        sa.Column("project_id", sa.String(36), nullable=False, server_default=""),
+        sa.Column("lipsync_job_id", sa.String(36), nullable=False),
+        sa.Column("script_id", sa.String(36), nullable=False),
+        sa.Column("b_roll_segments", sa.JSON(), nullable=False, server_default="[]"),
+        sa.Column("title_config", sa.JSON(), nullable=False, server_default="{}"),
+        sa.Column("cover_config", sa.JSON(), nullable=False, server_default="{}"),
+        sa.Column("status", sa.String(20), nullable=False, server_default="pending", index=True),
+        sa.Column("progress", sa.Integer(), nullable=False, server_default=sa.text("0")),
+        sa.Column("output_video_url", sa.Text(), nullable=False, server_default=""),
+        sa.Column("output_cover_url", sa.Text(), nullable=False, server_default=""),
+        sa.Column("output_duration", sa.Float(), nullable=False, server_default=sa.text("0.0")),
+        sa.Column("error_message", sa.Text(), nullable=False, server_default=""),
+        sa.Column("submitted_at", sa.DateTime(), nullable=True),
+        sa.Column("started_at", sa.DateTime(), nullable=True),
+        sa.Column("completed_at", sa.DateTime(), nullable=True),
+        sa.Column("created_at", sa.DateTime(), nullable=False, server_default=sa.func.now()),
+        sa.Column("updated_at", sa.DateTime(), nullable=False, server_default=sa.func.now()),
+    )
+    op.create_index("ix_ai_avatar_render_user_status", "ai_avatar_render_jobs", ["user_id", "status"])
+    op.create_index("ix_ai_avatar_render_project_user", "ai_avatar_render_jobs", ["project_id", "user_id"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_ai_avatar_render_project_user", table_name="ai_avatar_render_jobs")
+    op.drop_index("ix_ai_avatar_render_user_status", table_name="ai_avatar_render_jobs")
+    op.drop_table("ai_avatar_render_jobs")

alembic/versions/073_add_lipsync_tts_fields.py

@@ -0,0 +1,45 @@
+"""lipsync_jobs 增加 TTS 直生字段（voice_id/script_text/speed/emotion）
+
+Revision ID: 073_add_lipsync_tts_fields
+Revises: 072_add_ai_avatar_render
+Create Date: 2026-09-09
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "073_add_lipsync_tts_fields"
+down_revision = "072_add_ai_avatar_render"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 对口型支持「传音色 + 文案直接生成」：后端内部先 TTS 合成音频再提交对口型
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("voice_id", sa.String(200), nullable=False, server_default=""),
+    )
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("script_text", sa.Text(), nullable=False, server_default=""),
+    )
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("speed", sa.Float(), nullable=False, server_default=sa.text("1.0")),
+    )
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("emotion", sa.String(20), nullable=False, server_default=""),
+    )
+    # audio_url 改为可空：直生模式下音频由后端 TTS 合成后回填
+    op.alter_column("lipsync_jobs", "audio_url", existing_type=sa.Text(), nullable=True)
+
+
+def downgrade() -> None:
+    op.alter_column("lipsync_jobs", "audio_url", existing_type=sa.Text(), nullable=False)
+    op.drop_column("lipsync_jobs", "emotion")
+    op.drop_column("lipsync_jobs", "speed")
+    op.drop_column("lipsync_jobs", "script_text")
+    op.drop_column("lipsync_jobs", "voice_id")

alembic/versions/074_ai_avatar_render_script_id_optional.py

@@ -0,0 +1,36 @@
+"""ai_avatar_render_jobs.script_id 放宽为可空串（手动文案直生场景不关联文案库）
+
+Revision ID: 074_render_script_id_optional
+Revises: 073_add_lipsync_tts_fields
+Create Date: 2026-09-09
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "074_render_script_id_optional"
+down_revision = "073_add_lipsync_tts_fields"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 列保持 NOT NULL（空串占位），仅应用层允许不传；这里显式补 server_default 防止历史约束歧义
+    with op.batch_alter_table("ai_avatar_render_jobs") as batch:
+        batch.alter_column(
+            "script_id",
+            existing_type=sa.String(length=36),
+            nullable=False,
+            server_default="",
+        )
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("ai_avatar_render_jobs") as batch:
+        batch.alter_column(
+            "script_id",
+            existing_type=sa.String(length=36),
+            nullable=False,
+            server_default=None,
+        )

apps/api/app/api/router.py

@@ -1,4 +1,5 @@
 from app.api.routes.ai import router as ai_router
+from app.api.routes.ai_avatar_render import router as ai_avatar_render_router
 from app.api.routes.asset_diagnosis import router as asset_diagnosis_router
 from app.api.routes.asset_libraries import router as asset_libraries_router
 from app.api.routes.assets import router as assets_router
@@ -11,10 +12,13 @@ from app.api.routes.feature_flags import router as feature_flags_router
 from app.api.routes.generation_cover import router as generation_cover_router
 from app.api.routes.generation_preview import router as generation_preview_router
 from app.api.routes.generation_tasks import router as generation_tasks_router
+from app.api.routes.generation_variant_plans import router as generation_variant_plans_router
 from app.api.routes.health import router as health_check_router
 from app.api.routes.ingest_jobs import router as ingest_jobs_router
 from app.api.routes.internal_render import router as internal_render_router
+from app.api.routes.lipsync import router as lipsync_router
 from app.api.routes.projects import router as projects_router
+from app.api.routes.scripts import router as scripts_router
 from app.api.routes.share import router as share_router
 from app.api.routes.subscription import router as subscription_router
 from app.api.routes.tags import router as tags_router
@@ -37,6 +41,11 @@ api_router.include_router(
     auth_router,
     tags=["Auth"],
 )
+api_router.include_router(
+    lipsync_router,
+    prefix="/lipsync",
+    tags=["Lipsync"],
+)
 api_router.include_router(
     projects_router,
     prefix="/projects",
@@ -99,6 +108,11 @@ api_router.include_router(
     prefix="/generation",
     tags=["Generation"],
 )
+api_router.include_router(
+    generation_variant_plans_router,
+    prefix="/generation",
+    tags=["Generation"],
+)
 api_router.include_router(
     generation_cover_router,
     prefix="/generation",
@@ -165,3 +179,13 @@ api_router.include_router(
     internal_render_router,
     tags=["Internal"],
 )
+api_router.include_router(
+    scripts_router,
+    prefix="/scripts",
+    tags=["ScriptLibrary"],
+)
+api_router.include_router(
+    ai_avatar_render_router,
+    prefix="/ai-avatar/render",
+    tags=["AI Avatar Render"],
+)

apps/api/app/api/routes/ai_avatar_render.py

@@ -0,0 +1,209 @@
+"""AI数字人渲染合成 API 路由 — #1798.
+
+接口：
+  POST   /api/v1/ai-avatar/render              提交渲染任务
+  GET    /api/v1/ai-avatar/render/jobs          任务列表
+  GET    /api/v1/ai-avatar/render/{job_id}      任务详情
+  POST   /api/v1/ai-avatar/render/{job_id}/cancel  取消任务
+  POST   /api/v1/ai-avatar/render/{job_id}/retry   重试失败任务
+"""
+
+from __future__ import annotations
+
+import logging
+
+from app.auth import AuthenticatedUser, get_current_user
+from app.dependencies import get_db_session
+from app.schemas.ai_avatar_render import (
+    AiAvatarRenderJobResponse,
+    CreateAiAvatarRenderRequest,
+    SmartCoverRequest,
+    SmartCoverResponse,
+)
+from app.services.ai_avatar_cover_service import generate_smart_cover
+from app.services.ai_avatar_render_service import (
+    AiAvatarRenderError,
+    AiAvatarRenderService,
+)
+from fastapi import APIRouter, Depends, HTTPException, Query
+from sqlalchemy.orm import Session
+
+logger = logging.getLogger(__name__)
+
+router = APIRouter()
+
+
+def _get_service(db: Session = Depends(get_db_session)) -> AiAvatarRenderService:
+    return AiAvatarRenderService(db)
+
+
+# ── POST / — 提交渲染任务 ────────────────────────────────────────────────
+
+
+@router.post("", response_model=AiAvatarRenderJobResponse, status_code=201)
+def create_render_job(
+    body: CreateAiAvatarRenderRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: AiAvatarRenderService = Depends(_get_service),
+):
+    """提交 AI 数字人渲染任务.
+
+    将对口型视频 + B-roll 素材 + 标题叠加 + 封面提取合成最终输出视频。
+    """
+    try:
+        job = svc.create_render_job(
+            user_id=current_user.user.id,
+            lipsync_job_id=body.lipsync_job_id,
+            script_id=body.script_id,
+            b_roll_segments=[s.model_dump() for s in body.b_roll_segments],
+            title_config=body.title_config,
+            cover_config=body.cover_config,
+            project_id=body.project_id,
+        )
+    except AiAvatarRenderError as exc:
+        status_map = {
+            "LipsyncJobNotFound": 404,
+            "LipsyncJobNotCompleted": 400,
+            "LipsyncJobNoOutput": 400,
+            "ScriptNotFound": 404,
+        }
+        raise HTTPException(
+            status_code=status_map.get(exc.code, 400),
+            detail={"code": exc.code, "message": str(exc)},
+        ) from exc
+
+    # 异步触发渲染
+    try:
+        from app.tasks.ai_avatar_render import execute_ai_avatar_render
+
+        execute_ai_avatar_render.delay(job.id)
+    except Exception:
+        logger.warning("Celery 任务提交失败，渲染任务已创建但未触发执行: %s", job.id)
+
+    return job
+
+
+# ── GET /jobs — 任务列表 ─────────────────────────────────────────────────
+
+
+@router.get("/jobs", response_model=dict)
+def list_render_jobs(
+    project_id: str = Query("", description="项目 ID 过滤"),
+    status: str = Query("", description="状态过滤"),
+    offset: int = Query(0, ge=0),
+    limit: int = Query(20, ge=1, le=100),
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: AiAvatarRenderService = Depends(_get_service),
+):
+    """获取 AI 数字人渲染任务列表."""
+    items, total = svc.list_render_jobs(
+        user_id=current_user.user.id,
+        project_id=project_id,
+        status=status,
+        offset=offset,
+        limit=limit,
+    )
+    return {
+        "items": [AiAvatarRenderJobResponse.model_validate(j) for j in items],
+        "total": total,
+        "offset": offset,
+        "limit": limit,
+    }
+
+
+# ── GET /{job_id} — 任务详情 ─────────────────────────────────────────────
+
+
+@router.get("/{job_id}", response_model=AiAvatarRenderJobResponse)
+def get_render_job(
+    job_id: str,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: AiAvatarRenderService = Depends(_get_service),
+):
+    """获取渲染任务详情."""
+    job = svc.get_render_job(job_id, current_user.user.id)
+    if job is None:
+        raise HTTPException(status_code=404, detail="渲染任务不存在")
+    return job
+
+
+# ── POST /{job_id}/cancel — 取消任务 ─────────────────────────────────────
+
+
+@router.post("/{job_id}/cancel", response_model=AiAvatarRenderJobResponse)
+def cancel_render_job(
+    job_id: str,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: AiAvatarRenderService = Depends(_get_service),
+):
+    """取消渲染任务（仅 pending 状态可取消）."""
+    job = svc.cancel_render_job(job_id, current_user.user.id)
+    if job is None:
+        raise HTTPException(status_code=404, detail="渲染任务不存在")
+    if job.status != "cancelled":
+        raise HTTPException(
+            status_code=400,
+            detail=f"任务状态 {job.status} 不可取消，仅 pending 可取消",
+        )
+    return job
+
+
+# ── POST /{job_id}/retry — 重试失败任务 ──────────────────────────────────
+
+
+@router.post("/{job_id}/retry", response_model=AiAvatarRenderJobResponse)
+def retry_render_job(
+    job_id: str,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: AiAvatarRenderService = Depends(_get_service),
+):
+    """重试失败的渲染任务."""
+    job = svc.retry_render_job(job_id, current_user.user.id)
+    if job is None:
+        raise HTTPException(status_code=404, detail="渲染任务不存在")
+    if job.status != "pending":
+        raise HTTPException(
+            status_code=400,
+            detail=f"仅 failed 状态的任务可重试，当前状态: {job.status}",
+        )
+
+    # 重新触发渲染
+    try:
+        from app.tasks.ai_avatar_render import execute_ai_avatar_render
+
+        execute_ai_avatar_render.delay(job.id)
+    except Exception:
+        logger.warning("Celery 任务提交失败，重试任务已重置但未触发执行: %s", job.id)
+
+    return job
+
+
+
+# ── POST /smart-cover — 智能获取封面（MediaKit 抽帧 + 评分选帧）────────
+
+
+@router.post("/smart-cover", response_model=SmartCoverResponse)
+def generate_avatar_smart_cover(
+    body: SmartCoverRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+) -> SmartCoverResponse:
+    """智能获取数字人视频封面.
+
+    复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧逻辑（非 FFmpeg 简单截帧），
+    并将选中帧转存到自家 OSS，返回非临时的封面公网 URL。
+
+    前端「智能获取封面」按钮可直接调用本接口；不依赖渲染任务完成。
+    """
+    video_url = (body.video_url or "").strip()
+    if not video_url.startswith(("http://", "https://")):
+        raise HTTPException(status_code=400, detail="video_url 必须是合法的 HTTP/HTTPS URL")
+
+    cover_url = generate_smart_cover(video_url, max_frames=body.max_frames)
+    if not cover_url:
+        return SmartCoverResponse(
+            cover_url="",
+            status="fallback_failed",
+            message="智能抽帧失败（MediaKit 不可用或抽帧异常），请稍后重试",
+        )
+    logger.info("智能封面生成成功: user=%s", current_user.user.id)
+    return SmartCoverResponse(cover_url=cover_url, status="completed")

apps/api/app/api/routes/asset_libraries.py

@@ -20,7 +20,7 @@ from packages.application import (
     GetProjectUseCase,
     ListAssetLibrariesUseCase,
 )
-from packages.domain import AssetLibrary, AssetLibraryKind
+from packages.domain import AssetLibraryKind
 
 from ._helpers import check_project_access
 
@@ -120,30 +120,11 @@ def ensure_default_library(
 
     kind = AssetLibraryKind(request.kind)
 
-    # 查找该项目下同 kind 的素材库，返回第一个
-    existing = asset_library_repository.find_by_project(request.project_id)
-    for lib in existing:
-        if lib.kind == kind:
-            return _to_asset_library_response(lib)
-
-    # 不存在 → 自动创建
-    import uuid
-    from datetime import datetime, timezone
-
-    now = datetime.now(timezone.utc)
+    # Issue #1775: 幂等获取/创建——依赖唯一约束 uq_asset_libraries_project_kind，
+    # 并发创建冲突时回滚重查返回已有记录，不再依赖应用层"先查后插"，也不会 500。
     default_name = _DEFAULT_LIBRARY_NAMES.get(request.kind, f"{request.kind}素材库")
-    library = AssetLibrary(
-        id=str(uuid.uuid4()),
-        project_id=request.project_id,
-        name=default_name,
-        kind=kind,
-        asset_count=0,
-        total_size=0,
-        created_at=now,
-        updated_at=now,
-    )
-    created = asset_library_repository.create(library)
-    return _to_asset_library_response(created)
+    library = asset_library_repository.get_or_create_default_library(request.project_id, kind, name=default_name)
+    return _to_asset_library_response(library)
 
 
 @router.delete("/{library_id}", status_code=status.HTTP_204_NO_CONTENT, response_class=Response)

apps/api/app/api/routes/auth.py

@@ -1,4 +1,5 @@
 """
+from __future__ import annotations
 Canonical authentication API routes.
 
 The route layer is intentionally thin: repository construction lives in
@@ -13,9 +14,9 @@ import jwt
 from app.auth import AuthenticatedUser, blacklist_token, get_current_user
 from app.config import settings
 from app.dependencies import get_auth_email_service, get_auth_session_store, get_user_repository
-from fastapi import APIRouter, Depends, Header, HTTPException, status
+from fastapi import APIRouter, Depends, Header, HTTPException, Request, status
 from fastapi.security import HTTPAuthorizationCredentials, HTTPBearer
-from pydantic import BaseModel, EmailStr
+from pydantic import BaseModel, EmailStr, field_validator
 
 from packages.adapters.redis import NoopSessionStore
 from packages.adapters.smtp import NoopEmailService
@@ -84,6 +85,23 @@ class CurrentUserResponse(BaseModel):
     phone: str = ""
     phone_verified: bool = False
     binding_complete: bool = False
+    wechat_bound: bool = False
+    profile_completed: bool = True
+
+
+class UserProfileResponse(BaseModel):
+    """用户资料负载（PATCH /me、绑定/解绑接口复用；字段与 GET /auth/me 一致，前端 normalizeUser 直接消费）"""
+
+    user_id: str
+    email: str
+    username: str
+    display_name: str
+    email_verified: bool
+    phone: str = ""
+    phone_verified: bool = False
+    binding_complete: bool = False
+    wechat_bound: bool = False
+    profile_completed: bool = True
 
 
 class PasswordResetRequestModel(BaseModel):
@@ -272,9 +290,52 @@ async def get_current_user_info(
         phone=user.phone or "",
         phone_verified=user.phone_verified,
         binding_complete=binding_complete,
+        wechat_bound=bool(user.wechat_openid),
+        profile_completed=user.profile_completed,
     )
 
 
+class UpdateProfileRequest(BaseModel):
+    """更新个人资料请求（当前仅支持昵称）"""
+
+    display_name: str
+
+    @field_validator("display_name")
+    @classmethod
+    def _validate_display_name(cls, v: str) -> str:
+        name = (v or "").strip()
+        if not name:
+            raise ValueError("昵称不能为空白")
+        if len(name) > 20:
+            raise ValueError("昵称长度需在 1-20 个字符之间")
+        return name
+
+
+class UpdateProfileResponse(BaseModel):
+    """更新资料响应：前端 normalizeUser(response.user) 直接消费"""
+
+    user: UserProfileResponse
+
+
+@router.patch("/me", response_model=UpdateProfileResponse)
+async def update_current_user_profile(
+    request: UpdateProfileRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    user_repository: UserRepository = Depends(get_user_repository),
+) -> UpdateProfileResponse:
+    """更新当前登录用户昵称（微信新用户首次设置昵称后置 profile_completed=True）。"""
+    user = current_user.user
+    user.display_name = request.display_name  # 已 strip（validator）
+    if not user.profile_completed:
+        user.profile_completed = True
+    user_repository.save(user)
+
+    logger.info("[资料更新] 用户 %s 更新昵称，profile_completed=%s", user.id, user.profile_completed)
+    # 重新读取，确保返回的是持久化后的最新状态
+    fresh = user_repository.find_by_id(user.id) or user
+    return UpdateProfileResponse(user=_user_profile(fresh))
+
+
 class _NoopSessionStore(NoopSessionStore):
     pass
 
@@ -426,6 +487,7 @@ async def get_wechat_auth_url() -> WechatAuthUrlResponse:
 @router.post("/wechat/callback", response_model=WechatLoginResponse)
 async def wechat_callback(
     request: WechatCallbackRequest,
+    http_request: Request,
     user_repository: UserRepository = Depends(get_user_repository),
 ) -> WechatLoginResponse:
     """微信登录回调处理"""
@@ -433,11 +495,30 @@ async def wechat_callback(
     from packages.application.auth.wechat_sync_use_case import WechatSyncRequest as SyncRequest
     from packages.application.auth.wechat_sync_use_case import WechatSyncUseCase
 
+    # 回调可观测性：记录 UA（区分微信内置浏览器 MicroMessenger）与 state，
+    # 便于排查"停留 open.weixin.qq.com / 回调失败"类问题（#1718）
+    user_agent = http_request.headers.get("User-Agent", "")
+    is_wechat_browser = "MicroMessenger" in user_agent
+    logger.info(
+        "[微信回调] 收到回调: state=%s code_len=%d UA=%r 微信内置浏览器=%s",
+        (request.state or "")[:8],
+        len(request.code or ""),
+        user_agent[:200],
+        is_wechat_browser,
+    )
+
     # 1. 用 code 换微信用户信息
     oauth_service = get_wechat_oauth_service()
     wechat_user, err = oauth_service.handle_callback(request.code, request.state)
     if err:
+        # state 校验失败 / 微信 errcode 等错误原文已在 service 内 log，这里带上 UA 上下文
+        logger.warning("[微信回调] 处理失败: err=%s 微信内置浏览器=%s", err, is_wechat_browser)
         raise HTTPException(status_code=400, detail=err)
+    logger.info(
+        "[微信回调] state 校验通过，微信用户信息获取成功: openid=%s unionid=%s",
+        wechat_user.openid[:8] if wechat_user.openid else "",
+        bool(wechat_user.unionid),
+    )
 
     # 2. 同步登录/注册（复用 wechat-sync 逻辑）
     use_case = WechatSyncUseCase(user_repository=user_repository)
@@ -472,6 +553,109 @@ async def wechat_callback(
     )
 
 
+# ==================== 微信账号绑定/解绑（已登录用户） ====================
+
+
+class WechatBindUrlResponse(BaseModel):
+    auth_url: str
+    state: str
+
+
+class WechatBindCompleteRequest(BaseModel):
+    code: str
+    state: str = ""
+
+
+class WechatBindCompleteResponse(BaseModel):
+    success: bool
+    user: UserProfileResponse
+
+
+class WechatUnbindResponse(BaseModel):
+    success: bool
+
+
+def _user_profile(user) -> UserProfileResponse:
+    binding_complete = bool(
+        user.phone_verified and user.email_verified and user.email and "@wechat.local" not in user.email
+    )
+    return UserProfileResponse(
+        user_id=user.id,
+        email=user.email,
+        username=user.username,
+        display_name=user.display_name,
+        email_verified=user.email_verified,
+        phone=user.phone or "",
+        phone_verified=user.phone_verified,
+        binding_complete=binding_complete,
+        wechat_bound=bool(user.wechat_openid),
+        profile_completed=user.profile_completed,
+    )
+
+
+@router.get("/wechat/bind/url", response_model=WechatBindUrlResponse)
+async def get_wechat_bind_url(
+    current_user: AuthenticatedUser = Depends(get_current_user),
+) -> WechatBindUrlResponse:
+    """获取微信绑定授权链接（已登录用户场景）。state 经 Redis 存储做 CSRF 校验。"""
+    from packages.application.auth.wechat_oauth_service import get_wechat_oauth_service
+
+    oauth_service = get_wechat_oauth_service()
+    auth_url, state = oauth_service.generate_auth_url()
+    logger.info("[微信绑定] 用户 %s 请求绑定授权链接", current_user.user.id)
+    return WechatBindUrlResponse(auth_url=auth_url, state=state)
+
+
+@router.post("/wechat/bind", response_model=WechatBindCompleteResponse)
+async def wechat_bind(
+    request: WechatBindCompleteRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    user_repository: UserRepository = Depends(get_user_repository),
+) -> WechatBindCompleteResponse:
+    """微信绑定完成：扫码回调后用 code 换 openid，绑定到当前登录账号（不创建新用户）。"""
+    from packages.application.auth.wechat_bind_use_case import WechatBindRequest, WechatBindUseCase
+    from packages.application.auth.wechat_oauth_service import get_wechat_oauth_service
+
+    oauth_service = get_wechat_oauth_service()
+    wechat_user, err = oauth_service.handle_callback(request.code, request.state)
+    if err:
+        logger.warning("[微信绑定] 用户 %s 换取微信信息失败: %s", current_user.user.id, err)
+        raise HTTPException(status_code=400, detail=err)
+
+    use_case = WechatBindUseCase(user_repository=user_repository)
+    result, error, http_status = use_case.bind(
+        WechatBindRequest(
+            user_id=current_user.user.id,
+            openid=wechat_user.openid,
+            unionid=wechat_user.unionid or "",
+        )
+    )
+    if error:
+        logger.warning("[微信绑定] 用户 %s 绑定失败: %s", current_user.user.id, error)
+        raise HTTPException(status_code=http_status, detail=error)
+
+    logger.info("[微信绑定] 用户 %s 绑定成功 openid=%s", current_user.user.id, wechat_user.openid[:8])
+    return WechatBindCompleteResponse(success=True, user=_user_profile(result.user))
+
+
+@router.delete("/wechat/bind", response_model=WechatUnbindResponse)
+async def wechat_unbind(
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    user_repository: UserRepository = Depends(get_user_repository),
+) -> WechatUnbindResponse:
+    """解绑微信：需账号仍有其他登录方式（密码/手机/真实邮箱），否则拒绝。"""
+    from packages.application.auth.wechat_bind_use_case import WechatUnbindUseCase
+
+    use_case = WechatUnbindUseCase(user_repository=user_repository)
+    result, error, http_status = use_case.unbind(current_user.user.id)
+    if error:
+        logger.warning("[微信解绑] 用户 %s 解绑失败: %s", current_user.user.id, error)
+        raise HTTPException(status_code=http_status, detail=error)
+
+    logger.info("[微信解绑] 用户 %s 解绑成功", current_user.user.id)
+    return WechatUnbindResponse(success=True)
+
+
 # ==================== 验证码 & 绑定 ====================
 
 

apps/api/app/api/routes/chunked_upload.py

@@ -14,6 +14,7 @@ from typing import Any
 from uuid import uuid4
 
 from app.api.routes._helpers import require_project_and_library
+from app.api.routes.upload import _persist_celery_task_id
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.celery_app import celery_app
 from app.core.storage import OSSStorageService, get_storage_service
@@ -381,7 +382,8 @@ async def complete_chunked_upload(
                 file_hash=request.file_hash,
             )
         )
-        celery_app.send_task("worker.ingest_asset", args=[job.id])
+        celery_result = celery_app.send_task("worker.ingest_asset", args=[job.id])
+        _persist_celery_task_id(ingest_job_repository, job, getattr(celery_result, "id", ""))
 
         # Update metadata status
         meta["status"] = "completed"

apps/api/app/api/routes/generation_cover.py

@@ -75,6 +75,86 @@ class GenerateCoverResponse(BaseModel):
 # ── Route ────────────────────────────────────────────────────────────────
 
 
+
+def _select_best_frame_from_snapshots(
+    snapshots: list[dict], plan_id: str
+) -> str:
+    """从 MediaKit 抽帧结果中，通过质量评分选出最佳帧。
+
+    降级策略：cv2 不可用或评分失败时，返回第一帧。
+
+    Args:
+        snapshots: MediaKit 返回的帧列表 [{"image_url": str, ...}, ...]
+        plan_id: 计划 ID（日志用）
+
+    Returns:
+        最佳帧的 image_url，或空字符串
+    """
+    if not snapshots:
+        return ""
+    if len(snapshots) == 1:
+        return snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+
+    try:
+        import tempfile
+
+        import httpx
+
+        from packages.shared.cover_frame_scorer import score_frames
+
+        scored_candidates = []
+        for snap in snapshots:
+            url = snap.get("image_url") or snap.get("url") or ""
+            if not url:
+                continue
+            # 下载帧到临时文件进行评分
+            try:
+                resp = httpx.get(url, timeout=15, follow_redirects=True)
+                resp.raise_for_status()
+                with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp:
+                    tmp.write(resp.content)
+                    tmp_path = tmp.name
+                scored_candidates.append({"image_path": tmp_path, "url": url})
+            except Exception:
+                # 下载失败的帧跳过，给默认低分
+                scored_candidates.append({"image_path": None, "url": url, "score": 0.0})
+
+        if not scored_candidates:
+            return snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+
+        scored = score_frames(scored_candidates)
+        best = scored[0] if scored else None
+        best_url = best.get("url", "") if best else ""
+        best_score = best.get("score", 0.0) if best else 0.0
+        logger.info(
+            "[封面生成] 帧质量评分完成: plan_id=%s candidates=%d best_score=%.1f",
+            plan_id,
+            len(scored_candidates),
+            best_score,
+        )
+
+        # 清理临时文件
+        for c in scored_candidates:
+            path = c.get("image_path")
+            if path:
+                try:
+                    from pathlib import Path
+
+                    Path(path).unlink(missing_ok=True)
+                except Exception:
+                    pass
+
+        return best_url
+
+    except Exception:
+        logger.warning(
+            "[封面生成] 帧质量评分失败，使用第一帧: plan_id=%s",
+            plan_id,
+            exc_info=True,
+        )
+        return snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+
+
 def _persist_cover_frame(
     frame_url: str,
     plan_id: str,
@@ -652,13 +732,13 @@ def generate_cover(
                     snapshots = mk_client.extract_frames(
                         video_url=primary_video_url,
                         strategy="SpecifiedFrames",
-                        max_frames=1,
+                        max_frames=5,  # 抽 5 帧，通过质量评分选最佳
                         poll_interval=2.0,
                         max_poll_attempts=5,
                         max_retries=0,
                     )
                     if snapshots:
-                        raw = snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+                        raw = _select_best_frame_from_snapshots(snapshots, plan_id)
                         if raw:
                             cover_url_from_task = _persist_cover_frame(raw, plan_id)
                             logger.info(
@@ -715,13 +795,13 @@ def generate_cover(
                         snapshots = mk_client.extract_frames(
                             video_url=src_url,
                             strategy="SpecifiedFrames",
-                            max_frames=1,
+                            max_frames=5,  # 抽 5 帧，通过质量评分选最佳
                             poll_interval=2.0,
                             max_poll_attempts=5,
                             max_retries=0,
                         )
                         if snapshots:
-                            raw = snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+                            raw = _select_best_frame_from_snapshots(snapshots, plan_id)
                             if raw:
                                 cover_url_from_task = _persist_cover_frame(
                                     raw,

apps/api/app/api/routes/generation_preview.py

@@ -14,6 +14,7 @@ from app.core.task_enqueue import (
     USER_PENDING_LIMIT,
     GlobalQueueFull,
     UserPendingLimitExceeded,
+    build_rate_limit_detail,
     safe_enqueue_generation_task,
 )
 from app.dependencies import (
@@ -23,6 +24,7 @@ from app.dependencies import (
     get_generation_task_repository,
 )
 from app.schemas.generation_task import (
+    BatchPreviewGenerationTaskResponse,
     CreatePreviewGenerationTaskRequest,
     PreviewGenerationTaskResponse,
 )
@@ -193,11 +195,19 @@ def _to_preview_response(task, generated_videos: list | None = None) -> PreviewG
     if started_at and completed_at:
         generate_duration = (completed_at - started_at).total_seconds()
 
+    title_cfg = getattr(task, "title_config", None)
+    title_cfg = title_cfg if isinstance(title_cfg, dict) else {}
+    extra_meta = getattr(task, "extra_meta", None)
+    extra_meta = extra_meta if isinstance(extra_meta, dict) else {}
+    voice_library_id = getattr(task, "voice_library_id", "") or ""
+    if not isinstance(voice_library_id, str):
+        voice_library_id = str(voice_library_id) if voice_library_id else ""
     return PreviewGenerationTaskResponse(
         task_id=task.id,
         status=task.status.value if hasattr(task.status, "value") else str(task.status),
         progress=float(task.progress or 0.0),
         is_preview=bool(getattr(task, "is_preview", True)),
+        variant_index=int(extra_meta.get("variant_index", 0) or 0),
         resolution=getattr(task, "resolution", "") or "",
         video_url=video_url,
         duration=duration,
@@ -206,6 +216,8 @@ def _to_preview_response(task, generated_videos: list | None = None) -> PreviewG
         transition_count=transition_count,
         material_usage=material_usage,
         error_message=task.error_message or "",
+        title_text=str(title_cfg.get("text", "") or ""),
+        voice_library_id=voice_library_id,
         created_at=task.created_at,
         started_at=started_at,
         finished_at=completed_at,
@@ -213,50 +225,100 @@ def _to_preview_response(task, generated_videos: list | None = None) -> PreviewG
     )
 
 
-@router.post("/preview", response_model=PreviewGenerationTaskResponse, status_code=201)
+def _resolve_preview_edit_plan_id(
+    *,
+    request: CreatePreviewGenerationTaskRequest,
+    task,
+    db: Session,
+    user_id: str,
+) -> str:
+    """确定任务关联的编辑计划ID：优先前端传入，否则按 template_id+user 兜底查找。"""
+    if task.source_edit_plan_id:
+        return task.source_edit_plan_id
+    if not request.template_id:
+        return ""
+    try:
+        from packages.adapters.sqlalchemy_impl.edit_plan_repository import (
+            SQLAlchemyEditPlanRepository,
+        )
+
+        _plan_repo = SQLAlchemyEditPlanRepository(db)
+        _plans = _plan_repo.list_by_template(request.template_id, limit=20)
+        for _p in _plans:
+            if (_p.created_by_user_id or "") == user_id:
+                logger.info(
+                    "[预览生成] 自动关联编辑计划: task_id=%s plan_id=%s",
+                    task.id,
+                    _p.id,
+                )
+                return _p.id
+    except Exception:
+        logger.warning(
+            "[预览生成] 查找关联编辑计划失败（不影响主流程）: task_id=%s",
+            task.id,
+            exc_info=True,
+        )
+    return ""
+
+
+def _variant_value(values: list[str], index: int, fallback: str = "") -> str:
+    """从变体数组中取值：长度1=共用，长度>N=按索引，空数组=回退 fallback。"""
+    if not values:
+        return fallback
+    if len(values) == 1:
+        return values[0]
+    return values[index] if index < len(values) else fallback
+
+
+@router.post("/preview", response_model=BatchPreviewGenerationTaskResponse, status_code=201)
 def create_preview_generation_task(
     request: CreatePreviewGenerationTaskRequest,
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     generation_task_repository=Depends(get_generation_task_repository),
     db: Session = Depends(get_db_session),
     asset_repo=Depends(get_asset_repository),
-) -> PreviewGenerationTaskResponse:
-    """创建预览生成任务。
+) -> BatchPreviewGenerationTaskResponse:
+    """创建预览生成任务（支持批量）。
 
-    预览渲染品质与正式生成一致（1080p, CRF 23, medium preset），确认生成时可直接复用预览产物。
-
-    Args:
-        request: 预览任务创建请求（template_id + asset_ids 等）
+    preview_count=1 时行为与旧版完全一致（创建 1 个任务）；
+    preview_count=N 时一次创建 N 个独立变体任务：
+    - 每个变体克隆独立编辑计划（独立 clips、独立随机素材起点），N 个预览内容互不相同
+    - 每个变体拥有独立 task_id / 状态 / 预览视频 URL，前端按 task_id 分别轮询
+    - 标题样式（font/color/position 等）全局共用；标题文字/配音/封面可按变体独立
+      （titles[] / voice_library_ids[] / cover_urls[]，长度1=共用，长度N=独立）
 
     Returns:
-        201 + 预览任务详情
+        201 + 变体任务数组 {items: [...], total: N}
     """
     user_id = authenticated_user.user.id
+    count = max(1, request.preview_count)
     logger.info(
         "[预览生成] 接收请求: user_id=%s, template_id=%s, asset_count=%d, preview_count=%d",
         user_id,
         request.template_id,
         len(request.asset_ids),
-        request.preview_count,
+        count,
     )
 
-    # 预检查队列限流
+    # 预检查队列限流（按变体总数计）
     try:
         user_pending = generation_task_repository.count_pending_by_user(user_id)
         global_pending = generation_task_repository.count_pending_total()
-        if user_pending + 1 > USER_PENDING_LIMIT:
-            raise UserPendingLimitExceeded(user_id=user_id, pending_count=user_pending + 1, limit=USER_PENDING_LIMIT)
-        if global_pending + 1 > GLOBAL_PENDING_LIMIT:
-            raise GlobalQueueFull(pending_count=global_pending + 1, limit=GLOBAL_PENDING_LIMIT)
+        if user_pending + count > USER_PENDING_LIMIT:
+            raise UserPendingLimitExceeded(
+                user_id=user_id, pending_count=user_pending + count, limit=USER_PENDING_LIMIT
+            )
+        if global_pending + count > GLOBAL_PENDING_LIMIT:
+            raise GlobalQueueFull(pending_count=global_pending + count, limit=GLOBAL_PENDING_LIMIT)
     except UserPendingLimitExceeded as e:
         raise HTTPException(
             status_code=429,
-            detail=f"您的待处理任务过多（当前 {e.pending_count - 1}/{e.limit}），请等待后再提交",
+            detail=build_rate_limit_detail(e, generation_task_repository, scope="user"),
         ) from e
     except GlobalQueueFull as e:
         raise HTTPException(
             status_code=503,
-            detail="系统繁忙，请稍后再试",
+            detail=build_rate_limit_detail(e, generation_task_repository, scope="global"),
         ) from e
 
     # 确定视频比例：优先前端传入，否则从模板 mode 推断
@@ -273,14 +335,11 @@ def create_preview_generation_task(
                 w, h = int(parts[0]), int(parts[1])
                 base = 1920
                 if w < h:
-                    # 竖屏
                     output_width = round(base * w / h)
                     output_height = base
                 else:
-                    # 横屏
                     output_width = base
                     output_height = round(base * h / w)
-                # 对齐到偶数
                 output_width = output_width - output_width % 2
                 output_height = output_height - output_height % 2
             except (ValueError, ZeroDivisionError):
@@ -289,42 +348,70 @@ def create_preview_generation_task(
 
     logger.info(
         "[预览生成] 分辨率: video_ratio=%s → %s (%dx%d)",
-        video_ratio, resolution, output_width, output_height,
+        video_ratio,
+        resolution,
+        output_width,
+        output_height,
     )
 
-    # 从模板读取 editing_mode / mode 作为 strategy_id（渲染 pipeline 的 mode 参数）
     strategy_id = _resolve_strategy_id_from_template(request.template_id, db, user_id)
-
-    title_config = request.title_config or {}
+    base_title_config = request.title_config or {}
 
     use_case = CreateGenerationTaskUseCase(generation_task_repository)
 
+    # ── 预创建第一个任务，仅用于解析源编辑计划（不落库为最终任务）──
+    # 先创建一个临时任务拿到 task 对象上下文，实际 N 个任务在循环中统一创建；
+    # 为保持与旧版一致的源 plan 解析逻辑，先创建任务0、解析源 plan，
+    # 再预克隆 N 个变体 plan，最后重建任务关联。
+    # 简化实现：直接创建全部任务，plan 关联在创建后、入队前完成。
+
+    created_tasks: list = []
+    variant_plan_ids: list[str] = []  # 每个变体最终关联的 plan_id（按变体顺序）
+
     try:
-        task = use_case.execute(
-            CreateGenerationTaskCommand(
-                project_id="",
-                asset_library_id="",
-                strategy_id=strategy_id,
-                voice_library_id=request.voice_library_id,
-                template_id=request.template_id,
-                asset_ids=list(request.asset_ids),
-                title_ids=list(request.title_ids),
-                voice_ids=list(request.voice_ids),
-                created_by_user_id=user_id,
-                source_edit_plan_id=request.source_edit_plan_id,
-                asset_select_mode="",
-                batch_id="",
-                video_title=request.video_title,
-                resolution=resolution,
-                bgm_config=request.bgm_config or {},
-                auto_retry_enabled=False,
-                auto_retry_max=0,
-                is_preview=True,
-                title_config=title_config,
-                output_width=output_width,
-                output_height=output_height,
+        for variant_index in range(count):
+            # 变体独立标题文字：titles[] 覆盖 title_config.text
+            variant_title_text = _variant_value(request.titles, variant_index, "")
+            variant_title_config = dict(base_title_config)
+            if variant_title_text.strip():
+                variant_title_config["text"] = variant_title_text.strip()
+
+            # 变体独立配音
+            variant_voice_library_id = _variant_value(
+                request.voice_library_ids, variant_index, request.voice_library_id
             )
-        )
+
+            task = use_case.execute(
+                CreateGenerationTaskCommand(
+                    project_id="",
+                    asset_library_id="",
+                    strategy_id=strategy_id,
+                    voice_library_id=variant_voice_library_id,
+                    template_id=request.template_id,
+                    asset_ids=list(request.asset_ids),
+                    title_ids=list(request.title_ids),
+                    created_by_user_id=user_id,
+                    source_edit_plan_id=request.source_edit_plan_id,
+                    asset_select_mode="",
+                    batch_id="",
+                    video_title=request.video_title,
+                    resolution=resolution,
+                    bgm_config=request.bgm_config or {},
+                    auto_retry_enabled=False,
+                    auto_retry_max=0,
+                    is_preview=True,
+                    title_config=variant_title_config,
+                    output_width=output_width,
+                    output_height=output_height,
+                )
+            )
+            task.extra_meta["variant_index"] = variant_index
+
+            # 解析源编辑计划（前端传入或按模板兜底查找）
+            source_plan_id = _resolve_preview_edit_plan_id(request=request, task=task, db=db, user_id=user_id)
+            task.source_edit_plan_id = source_plan_id
+            generation_task_repository.update(task)
+            created_tasks.append(task)
     except ValueError as e:
         logger.warning("[预览生成] 创建失败: %s", e)
         raise HTTPException(status_code=400, detail=str(e)) from e
@@ -332,93 +419,204 @@ def create_preview_generation_task(
         logger.error("[预览生成] 创建失败: %s", e, exc_info=True)
         raise HTTPException(status_code=500, detail="创建预览生成任务失败，请稍后再试") from e
 
-    # 关联编辑计划：如果前端未传 source_edit_plan_id，通过 template_id + user_id 查找
-    if not task.source_edit_plan_id and request.template_id:
+    # ── 独立变体 plan（#1743）──
+    # count=1：克隆源 plan（预览不污染源 plan，仅起点重算），行为与旧版一致；
+    # count>1：变体 0 保留源 plan，变体 1..N-1 用 reselect_plan_for_variant 完整
+    #   重跑单视频选片（素材洗牌+镜头洗牌+起点随机+跨变体避让+批次 20% 重叠重选），
+    #   所见即所得——预览变体差异即正式成片差异。
+    source_plan_id = created_tasks[0].source_edit_plan_id if created_tasks else ""
+
+    # #1749：各变体配音解析（严格守卫已在 schema；此处取每变体 voice 查时长）+ 时长分配
+    def _preview_voice_durations() -> list[float]:
         try:
-            from packages.adapters.sqlalchemy_impl.edit_plan_repository import (
-                SQLAlchemyEditPlanRepository,
-            )
+            from packages.domain.variant_voice_resolver import resolve_variant_voice_ids
 
-            _plan_repo = SQLAlchemyEditPlanRepository(db)
-            _plans = _plan_repo.list_by_template(request.template_id, limit=20)
-            for _p in _plans:
-                if (_p.created_by_user_id or "") == user_id:
-                    task.source_edit_plan_id = _p.id
-                    generation_task_repository.update(task)
-                    logger.info(
-                        "[预览生成] 自动关联编辑计划: task_id=%s plan_id=%s",
-                        task.id,
-                        _p.id,
-                    )
-                    break
+            voices = resolve_variant_voice_ids(
+                count=count,
+                voice_library_id=request.voice_library_id,
+                voice_library_ids=request.voice_library_ids or None,
+            )
         except Exception:
-            logger.warning(
-                "[预览生成] 查找关联编辑计划失败（不影响主流程）: task_id=%s",
-                task.id,
-                exc_info=True,
-            )
+            logger.warning("[预览生成] 配音解析失败（按无配音处理）", exc_info=True)
+            return [0.0] * count
+        try:
+            from app.api.routes.generation_tasks import _query_voice_durations
 
-    # 每条预览都关联独立克隆 plan：多预览前端为 N 次并发调用，若共用同一 plan
-    # 则 N 条预览片段完全相同；克隆时片段起点按持久化历史区间重算（含受控复用），
-    # 保证各预览版本内容不同
-    if task.source_edit_plan_id:
+            return _query_voice_durations(db, voices)
+        except Exception:
+            return [0.0] * count
+
+    voice_durations = _preview_voice_durations()
+
+    if source_plan_id and count == 1:
+        # 单预览：克隆一份（原逻辑）+ 配音时长分配
         try:
             from app.services.edit_plan_service import EditPlanService
 
             _plan_svc = EditPlanService(db)
-            _preview_plan = _plan_svc.clone_plan_for_variant(
-                task.source_edit_plan_id,
+            variant_plan = _plan_svc.clone_plan_for_variant(
+                source_plan_id,
                 created_by_user_id=user_id,
                 name_suffix="预览变体",
             )
-            task.source_edit_plan_id = _preview_plan.id
-            generation_task_repository.update(task)
-            logger.info(
-                "[预览生成] 预览关联独立克隆 plan: task_id=%s clone_plan_id=%s",
-                task.id,
-                _preview_plan.id,
-            )
-        except Exception as clone_err:
-            # 不退回共用原 plan（否则多条预览内容相同，违反去重诉求）：
-            # 标记任务失败并中断，前端可重新发起预览
-            logger.error(
-                "[预览生成] 克隆预览变体 plan 失败，任务标记失败: task_id=%s error=%s",
-                task.id,
-                clone_err,
-                exc_info=True,
-            )
-            _mark_task_failed(generation_task_repository, task, "预览变体计划创建失败")
+            if voice_durations and voice_durations[0] > 0:
+                try:
+                    _plan_svc.apply_voice_duration_to_plan(variant_plan.id, voice_durations[0])
+                except Exception:
+                    logger.exception("[预览生成] 变体0 配音分配失败（不阻断）: plan=%s", variant_plan.id)
+            variant_plan_ids.append(variant_plan.id)
+        except Exception as e:
+            logger.error("[预览生成] 克隆预览 plan 异常: %s", e, exc_info=True)
+            for t in created_tasks:
+                _mark_task_failed(generation_task_repository, t, "预览计划创建失败")
             raise HTTPException(
                 status_code=500,
                 detail="创建预览任务失败：无法生成独立剪辑计划，请重试",
-            ) from clone_err
+            ) from e
+    elif source_plan_id and count > 1:
+        try:
+            from app.services.edit_plan_service import EditPlanService
 
-    # 入队执行；若入队失败则标记任务为 failed 避免僵尸数据
-    try:
-        if not safe_enqueue_generation_task(
-            task,
-            generation_task_repository,
-            user_id=user_id,
-            log_prefix="[预览生成]",
-            log_task_status=True,
-        ):
-            logger.warning("[预览生成] 任务入队失败: task_id=%s", task.id)
-            _mark_task_failed(generation_task_repository, task, "任务入队失败")
-            raise HTTPException(status_code=500, detail="任务入队失败，请稍后重试")
-    except UserPendingLimitExceeded as e:
-        _mark_task_failed(generation_task_repository, task, "待处理任务超限")
-        raise HTTPException(
-            status_code=429,
-            detail=f"您的待处理任务过多（当前 {e.pending_count - 1}/{e.limit}），请等待后再提交",
-        ) from None
-    except GlobalQueueFull:
-        _mark_task_failed(generation_task_repository, task, "系统队列已满")
+            _plan_svc = EditPlanService(db)
+            # #1749：变体 0 也 clone（不污染源 plan）+ 配音分配；变体 1..N-1 独立选片
+            _plan0 = _plan_svc.clone_plan_for_variant(
+                source_plan_id,
+                created_by_user_id=user_id,
+                name_suffix="预览变体1",
+            )
+            if voice_durations and voice_durations[0] > 0:
+                try:
+                    _plan_svc.apply_voice_duration_to_plan(_plan0.id, voice_durations[0])
+                except Exception:
+                    logger.exception("[预览生成] 变体0 配音分配失败（不阻断）: plan=%s", _plan0.id)
+            variant_plan_ids.append(_plan0.id)
+            batch_asset_pool = list(dict.fromkeys(request.asset_ids or []))
+            for variant_index in range(1, count):
+                last_err: Exception | None = None
+                variant_plan = None
+                for _attempt in range(2):  # 1 次重试，抗 DB 瞬时抖动
+                    try:
+                        variant_plan = _plan_svc.reselect_plan_for_variant(
+                            source_plan_id,
+                            batch_asset_pool,
+                            created_by_user_id=user_id,
+                            name_suffix=f"预览变体{variant_index + 1}",
+                            voice_duration=(
+                                voice_durations[variant_index] if variant_index < len(voice_durations) else 0.0
+                            ),
+                        )
+                        break
+                    except ValueError as ve:
+                        logger.warning("[预览生成] 变体独立选片失败(素材不足): %s", ve)
+                        for t in created_tasks:
+                            _mark_task_failed(generation_task_repository, t, "预览变体选片失败")
+                        raise HTTPException(
+                            status_code=400,
+                            detail=f"批量预览第 {variant_index + 1} 个视频无法独立选片：{ve}。"
+                            "请增加素材库中的视频素材后重试。",
+                        ) from ve
+                    except Exception as reselection_err:  # noqa: PERF203
+                        last_err = reselection_err
+                        logger.warning(
+                            "[预览生成] 变体独立选片失败(尝试%d/2): variant=%d error=%s",
+                            _attempt + 1,
+                            variant_index,
+                            reselection_err,
+                            exc_info=True,
+                        )
+                if variant_plan is None:
+                    logger.error(
+                        "[预览生成] 变体独立选片重试仍失败: variant=%d source=%s",
+                        variant_index,
+                        source_plan_id,
+                        exc_info=last_err,
+                    )
+                    for t in created_tasks:
+                        _mark_task_failed(generation_task_repository, t, "预览变体计划创建失败")
+                    raise HTTPException(
+                        status_code=500,
+                        detail="创建预览任务失败：无法生成独立剪辑计划，请重试",
+                    ) from last_err
+                variant_plan_ids.append(variant_plan.id)
+        except HTTPException:
+            raise
+        except Exception as e:
+            logger.error("[预览生成] 变体 plan 生成异常: %s", e, exc_info=True)
+            for t in created_tasks:
+                _mark_task_failed(generation_task_repository, t, "预览变体计划创建失败")
+            raise HTTPException(
+                status_code=500,
+                detail="创建预览任务失败：无法生成独立剪辑计划，请重试",
+            ) from e
+
+    # 关联变体 plan 并回写标题配置
+    for variant_index, task in enumerate(created_tasks):
+        if variant_plan_ids:
+            task.source_edit_plan_id = variant_plan_ids[variant_index]
+            generation_task_repository.update(task)
+        # 回写变体标题到 plan config（worker 渲染时从 plan 读取 title 配置）
+        if task.source_edit_plan_id and (task.title_config or {}).get("text", "").strip():
+            try:
+                from app.api.routes.generation_tasks import _writeback_edit_plan_config
+
+                _writeback_edit_plan_config(
+                    plan_id=task.source_edit_plan_id,
+                    task_id=task.id,
+                    title_config=task.title_config,
+                    db=db,
+                )
+            except Exception:
+                logger.warning(
+                    "[预览生成] 回写标题配置失败（不影响主流程）: task_id=%s",
+                    task.id,
+                    exc_info=True,
+                )
+
+    # ── 入队 ──
+    responses: list[PreviewGenerationTaskResponse] = []
+    rate_limit_exc: Exception | None = None  # 记录首个限流异常，全部失败时返回结构化提示
+    for variant_index, task in enumerate(created_tasks):
+        try:
+            enqueued = safe_enqueue_generation_task(
+                task,
+                generation_task_repository,
+                user_id=user_id,
+                log_prefix=f"[预览生成][变体{variant_index + 1}]",
+                log_task_status=True,
+            )
+            if not enqueued:
+                logger.warning("[预览生成] 任务入队失败: task_id=%s", task.id)
+                _mark_task_failed(generation_task_repository, task, "任务入队失败")
+        except UserPendingLimitExceeded as e:
+            _mark_task_failed(generation_task_repository, task, "待处理任务超限")
+            rate_limit_exc = rate_limit_exc or e
+        except GlobalQueueFull as e:
+            _mark_task_failed(generation_task_repository, task, "系统队列已满")
+            rate_limit_exc = rate_limit_exc or e
+        except Exception:
+            logger.exception("[预览生成] 入队异常: task_id=%s", task.id)
+            _mark_task_failed(generation_task_repository, task, "任务入队异常")
+        # enqueue 会原地更新 task 状态/进度，直接用 task 构造响应
+        responses.append(_to_preview_response(task))
+
+    # 队列满/限流时若全部失败，返回结构化错误码（前端区分"排队"与"创建失败"）
+    if all(r.status == "failed" for r in responses) and rate_limit_exc is not None:
+        if isinstance(rate_limit_exc, UserPendingLimitExceeded):
+            raise HTTPException(
+                status_code=429,
+                detail=build_rate_limit_detail(rate_limit_exc, generation_task_repository, scope="user"),
+            )
         raise HTTPException(
             status_code=503,
-            detail="系统繁忙，请稍后再试",
-        ) from None
+            detail=build_rate_limit_detail(rate_limit_exc, generation_task_repository, scope="global"),
+        )
 
-    return _to_preview_response(task)
+    logger.info(
+        "[预览生成] 创建完成: %d 个变体任务, task_ids=%s",
+        len(responses),
+        [r.task_id for r in responses],
+    )
+    return BatchPreviewGenerationTaskResponse(items=responses, total=len(responses))
 
 
 @router.get("/preview/{task_id}", response_model=PreviewGenerationTaskResponse)

apps/api/app/api/routes/generation_tasks.py

@@ -10,6 +10,7 @@ from app.core.task_enqueue import (
     USER_PENDING_LIMIT,
     GlobalQueueFull,
     UserPendingLimitExceeded,
+    build_rate_limit_detail,
     safe_enqueue_generation_task,
 )
 from app.dependencies import (
@@ -47,6 +48,39 @@ logger = logging.getLogger(__name__)
 router = APIRouter()
 
 
+def _variant_value(values: list[str], index: int, fallback: str = "") -> str:
+    """从变体数组中取值：长度1=共用，长度>N=按索引，空数组=回退 fallback。"""
+    if not values:
+        return fallback
+    if len(values) == 1:
+        return values[0]
+    return values[index] if index < len(values) else fallback
+
+
+def _query_voice_durations(db: Session, voice_ids: list[str]) -> list[float]:
+    """批量查询配音素材时长（秒），#1749 配音时长分配用。
+
+    逐项 try/float 硬化：MagicMock/异常/缺失 → 0.0（无配音不分配，不阻断）。
+    """
+    ids = [v for v in dict.fromkeys(voice_ids or []) if v]
+    if not ids:
+        return []
+    try:
+        from packages.adapters.sqlalchemy_impl.models import AssetModel
+
+        rows = db.query(AssetModel.id, AssetModel.duration).filter(AssetModel.id.in_(ids)).all()
+        dur_map: dict[str, float] = {}
+        for row in rows:
+            try:
+                dur_map[row[0]] = float(row[1] or 0.0)
+            except (TypeError, ValueError):
+                dur_map[row[0]] = 0.0
+        return [dur_map.get(v, 0.0) for v in ids]
+    except Exception:
+        logger.warning("[生成任务] 配音时长查询失败（按无配音处理，不阻断）", exc_info=True)
+        return [0.0 for _ in ids]
+
+
 def _to_generation_task_response(task) -> GenerationTaskResponse:
     return GenerationTaskResponse(
         id=task.id,
@@ -113,6 +147,7 @@ def _select_assets_from_library(
     assets: list,
     mode: str,
     count: int,
+    rng=None,
 ) -> list[str]:
     """根据选取模式从素材库中选取 ready 状态的视频素材 ID。
 
@@ -120,6 +155,8 @@ def _select_assets_from_library(
         assets: 素材库中所有素材（Asset 实体列表）
         mode: 选取模式 — all=全部, smart=智能匹配（多维度评分+多样性）
         count: 选取数量，0 表示全部（仅 smart 模式有效）
+        rng: 可选随机源（smart 模式排序噪声用），生产环境不传则内部随机；
+            测试可注入固定种子或零噪声随机源获得确定性结果。
 
     Returns:
         选中的素材 ID 列表
@@ -132,8 +169,9 @@ def _select_assets_from_library(
     if mode == "smart":
         # 智能匹配：统一使用 packages/domain/smart_match.py 的多维评分+多样性选取
         # 评分维度：质量分(40%) + 时长适配(30%) + 新鲜度(20%) + 未使用加分(10%)
+        # 排序注入随机噪声（#1743）：同分素材每次选出不同组合，从素材组合层面降重
         limit = count if count > 0 else None
-        results = smart_select_assets(ready_video_assets, limit=limit, kind="video")
+        results = smart_select_assets(ready_video_assets, limit=limit, kind="video", rng=rng)
         return [r.asset.id for r in results]
 
     # 默认 all 模式：返回全部 ready 视频素材
@@ -408,12 +446,12 @@ def create_generation_task(
     except UserPendingLimitExceeded as e:
         raise HTTPException(
             status_code=429,
-            detail=f"您的待处理任务过多（当前 {e.pending_count - count}/{e.limit}，本次提交 {count} 个），请等待完成后再提交",
+            detail=build_rate_limit_detail(e, generation_task_repository, scope="user"),
         ) from e
     except GlobalQueueFull as e:
         raise HTTPException(
             status_code=503,
-            detail="系统繁忙，请稍后再试",
+            detail=build_rate_limit_detail(e, generation_task_repository, scope="global"),
         ) from e
 
     # 画中画已下线：strategy_id 中的 pip/voice_pip 统一映射为 one_take
@@ -422,66 +460,218 @@ def create_generation_task(
         logger.info("画中画已下线，strategy_id %s → one_take", effective_strategy_id)
         effective_strategy_id = "one_take"
 
-    # 批量生成时每个任务关联独立克隆 plan（片段起点重算），
-    # 禁止 N 条任务共用同一 source_edit_plan_id 导致片段一模一样。
-    # 在创建任何任务【之前】预克隆全部变体：克隆失败直接中断（此时无脏数据），
-    # 绝不静默退回共用源 plan（否则批量视频内容重复，违反去重诉求）。
+    # 批量生成（count>1）：每个变体必须走与单视频完全相同的独立选片流程（#1743/#1749）。
+    # - 变体 0：clone 源 plan（不污染源 plan），变体 1..N-1 用 reselect_plan_for_variant
+    #   完整重跑选片（素材级去重：fresh 优先 → 受控复用 overlap≤20% → 短素材禁复用）；
+    # - #1749：前端可回传 variant-plans 接口预生成的 plan_id（variant_plan_ids），直接复用；
+    #   回传 plan 仍按各变体配音幂等重分配段长（防 variant-plans 阶段未带配音/占位时长）；
+    # - 配音时长：独立配音各自时长、统一配音同值，逐变体 apply_voice_duration_to_plan，
+    #   成片总时长=配音时长（素材短→末帧冻结，禁慢放/禁截配音）；
+    # - count>1 但没有源 plan 时，不允许 N 个任务兜底共用同一 plan，直接 4xx 中断。
+    # 在创建任何任务【之前】预生成/校验全部变体 plan：失败直接中断（此时无脏数据）。
     variant_plan_ids: list[str] = []
-    if count > 1 and request.source_edit_plan_id:
+    if count > 1:
         from app.services.edit_plan_service import EditPlanService
 
+        from packages.domain.variant_voice_resolver import VariantVoiceError, resolve_variant_voice_ids
+
         _plan_svc = EditPlanService(db)
-        for task_index in range(1, count):
-            variant = None
-            last_err: Exception | None = None
-            for _attempt in range(2):  # 1 次重试，抗 DB 瞬时抖动
-                try:
-                    variant = _plan_svc.clone_plan_for_variant(
-                        request.source_edit_plan_id,
-                        created_by_user_id=user_id,
-                        name_suffix=f"批量{task_index + 1}",
+
+        # 解析每变体配音（严格守卫：独立配音长度/缺值 → 400，禁静默 fallback）
+        try:
+            variant_voices = resolve_variant_voice_ids(
+                count=count,
+                voice_library_id=request.voice_library_id,
+                voice_library_ids=request.voice_library_ids or None,
+            )
+        except VariantVoiceError as ve:
+            raise HTTPException(status_code=400, detail=str(ve)) from ve
+        # 各变体配音时长（查询硬化：异常 → 0.0 不阻断）
+        voice_durations = _query_voice_durations(db, variant_voices)
+
+        # 解析批量源 plan：优先前端传入；否则按 template_id + user 查最新（与单任务兜底同源）
+        batch_source_plan_id = request.source_edit_plan_id
+        if not batch_source_plan_id and request.template_id:
+            try:
+                from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+
+                _latest = (
+                    db.query(EditPlanModel)
+                    .filter(
+                        EditPlanModel.template_id == request.template_id,
+                        EditPlanModel.created_by_user_id == user_id,
                     )
-                    break
-                except Exception as clone_err:  # noqa: PERF203
-                    last_err = clone_err
-                    logger.warning(
-                        "[生成任务] 克隆变体 plan 失败(尝试%d/2): source=%s error=%s",
-                        _attempt + 1,
-                        request.source_edit_plan_id,
-                        clone_err,
-                        exc_info=True,
-                    )
-            if variant is None:
-                logger.error(
-                    "[生成任务] 克隆变体 plan 重试仍失败，中断批量创建: source=%s",
-                    request.source_edit_plan_id,
-                    exc_info=last_err,
+                    .order_by(EditPlanModel.created_at.desc())
+                    .first()
                 )
+                if _latest:
+                    batch_source_plan_id = _latest.id
+            except Exception:
+                logger.warning("[生成任务] 批量源 plan 解析失败", exc_info=True)
+
+        if not batch_source_plan_id and not request.variant_plan_ids:
+            # 无任何可用源 plan：批量变体无从选片，明确报错，严禁静默共用/同源
+            logger.error("[生成任务] 批量 count=%d 但无可编辑计划（无 source_edit_plan_id/template plan）", count)
+            raise HTTPException(
+                status_code=400,
+                detail="批量生成需要先完成预览生成（缺少剪辑计划）。请先生成预览后再批量创建。",
+            )
+
+        # 批次素材池：请求显式素材 + 库自动匹配素材（resolved_asset_ids）
+        batch_asset_pool = list(dict.fromkeys(resolved_asset_ids or []))
+
+        if request.variant_plan_ids:
+            # ① 前端回传 variant-plans 预生成结果：直接复用（轻量选片接口已建好 plan）
+            if len(request.variant_plan_ids) != count:
                 raise HTTPException(
-                    status_code=500,
-                    detail="创建批量任务失败：无法生成独立剪辑计划，请重试",
-                ) from last_err
-            variant_plan_ids.append(variant.id)
+                    status_code=400,
+                    detail=f"variant_plan_ids 数量({len(request.variant_plan_ids)})与视频数量({count})不一致",
+                )
+            # 校验归属权
+            from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+
+            for _pid in request.variant_plan_ids:
+                _pm = db.query(EditPlanModel).filter(EditPlanModel.id == _pid).first()
+                if _pm is None:
+                    raise HTTPException(status_code=400, detail=f"剪辑计划不存在: {_pid}")
+                if _pm.created_by_user_id and _pm.created_by_user_id != user_id:
+                    raise HTTPException(status_code=403, detail=f"无权使用剪辑计划: {_pid}")
+            variant_plan_ids = list(request.variant_plan_ids)
+        else:
+            # ② 服务端选片：变体 0 clone 源 plan（不污染源 plan）
+            try:
+                _plan0 = _plan_svc.clone_plan_for_variant(
+                    batch_source_plan_id,
+                    created_by_user_id=user_id,
+                    name_suffix="批量1",
+                )
+            except Exception as clone_err:
+                logger.error("[生成任务] 变体0 clone 失败: %s", clone_err, exc_info=True)
+                raise HTTPException(
+                    status_code=500, detail="创建批量任务失败：无法生成独立剪辑计划，请重试"
+                ) from clone_err
+            variant_plan_ids.append(_plan0.id)
+
+            # 变体 1..N-1 独立选片
+            for task_index in range(1, count):
+                variant = None
+                last_err: Exception | None = None
+                for _attempt in range(2):  # 1 次重试，抗 DB 瞬时抖动
+                    try:
+                        variant = _plan_svc.reselect_plan_for_variant(
+                            batch_source_plan_id,
+                            batch_asset_pool,
+                            created_by_user_id=user_id,
+                            name_suffix=f"批量{task_index + 1}",
+                            voice_duration=voice_durations[task_index] if task_index < len(voice_durations) else 0.0,
+                        )
+                        break
+                    except ValueError as ve:
+                        # 素材不足等可预期错误：不重试，直接中断并给出明确提示
+                        logger.warning("[生成任务] 变体独立选片失败(素材不足): %s", ve)
+                        raise HTTPException(
+                            status_code=400,
+                            detail=f"批量生成第 {task_index + 1} 个视频无法独立选片：{ve}。"
+                            "请增加素材库中的视频素材后重试。",
+                        ) from ve
+                    except Exception as reselection_err:  # noqa: PERF203
+                        last_err = reselection_err
+                        logger.warning(
+                            "[生成任务] 变体独立选片失败(尝试%d/2): source=%s error=%s",
+                            _attempt + 1,
+                            batch_source_plan_id,
+                            reselection_err,
+                            exc_info=True,
+                        )
+                if variant is None:
+                    logger.error(
+                        "[生成任务] 变体独立选片重试仍失败，中断批量创建: source=%s",
+                        batch_source_plan_id,
+                        exc_info=last_err,
+                    )
+                    raise HTTPException(
+                        status_code=500,
+                        detail="创建批量任务失败：无法生成独立剪辑计划，请重试",
+                    ) from last_err
+                variant_plan_ids.append(variant.id)
+
+        # ③ 配音时长分配（回传 plan / clone 变体0 均需幂等分配；reselect 已在选片时分配）
+        for _vi, _pid in enumerate(variant_plan_ids):
+            _vd = voice_durations[_vi] if _vi < len(voice_durations) else 0.0
+            if _vd > 0:
+                try:
+                    _plan_svc.apply_voice_duration_to_plan(_pid, _vd)
+                except Exception:
+                    logger.exception("[生成任务] 变体%d 配音时长分配失败（不阻断）: plan=%s", _vi, _pid)
+
+    # N=1 正式生成：渲染侧全局慢放兜底已删除（#1749），enqueue 前也必须按配音分配段长
+    if count == 1 and not request.is_preview:
+        from packages.domain.variant_voice_resolver import VariantVoiceError, resolve_variant_voice_ids
+
+        try:
+            _voices = resolve_variant_voice_ids(
+                count=1,
+                voice_library_id=request.voice_library_id,
+                voice_library_ids=request.voice_library_ids or None,
+            )
+            _single_vd: list[float] = _query_voice_durations(db, _voices)
+            _single_dur = _single_vd[0] if _single_vd else 0.0
+            _single_plan = request.source_edit_plan_id
+            if not _single_plan and request.template_id:
+                try:
+                    from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+
+                    _latest = (
+                        db.query(EditPlanModel)
+                        .filter(
+                            EditPlanModel.template_id == request.template_id,
+                            EditPlanModel.created_by_user_id == user_id,
+                        )
+                        .order_by(EditPlanModel.created_at.desc())
+                        .first()
+                    )
+                    if _latest:
+                        _single_plan = _latest.id
+                except Exception:
+                    logger.warning("[生成任务] 单任务源 plan 解析失败", exc_info=True)
+            if _single_dur > 0 and _single_plan:
+                from app.services.edit_plan_service import EditPlanService
+
+                try:
+                    EditPlanService(db).apply_voice_duration_to_plan(_single_plan, _single_dur)
+                except Exception:
+                    logger.exception("[生成任务] N=1 配音时长分配失败（不阻断）: plan=%s", _single_plan)
+        except VariantVoiceError as ve:
+            raise HTTPException(status_code=400, detail=str(ve)) from ve
+        except Exception:
+            logger.exception("[生成任务] N=1 配音分配兜底异常（不阻断）")
 
     try:
         for task_index in range(count):
-            # 第 1 条复用源 plan（保留用户编辑结果）；其余使用预克隆的独立变体 plan。
-            # 无源 plan（source_edit_plan_id 为空）时无可克隆对象，variant_plan_ids
-            # 为空列表：各任务走自身随机选片流程，不做索引访问（防 IndexError）
-            effective_plan_id = request.source_edit_plan_id
-            if task_index > 0 and variant_plan_ids:
-                effective_plan_id = variant_plan_ids[task_index - 1]
+            # #1749：count>1 时每个变体（含变体0）都关联各自独立 plan（clone/reselect/variant-plans）。
+            if count > 1 and variant_plan_ids:
+                effective_plan_id = variant_plan_ids[task_index]
+            else:
+                effective_plan_id = request.source_edit_plan_id
+
+            # 变体级独立配置：titles[]/voice_library_ids[]/cover_urls[]
+            # 长度1=所有变体共用，长度=count=每个变体独立，空数组=回退单值字段
+            variant_title_text = _variant_value(request.titles, task_index, "")
+            variant_title_config = dict(request.title_config or {})
+            if variant_title_text.strip():
+                variant_title_config["text"] = variant_title_text.strip()
+            variant_voice_library_id = _variant_value(request.voice_library_ids, task_index, request.voice_library_id)
+            variant_cover_url = _variant_value(request.cover_urls, task_index, request.cover_url)
 
             task = use_case.execute(
                 CreateGenerationTaskCommand(
                     project_id=project_id,
                     asset_library_id=asset_library_id,
                     strategy_id=effective_strategy_id,
-                    voice_library_id=request.voice_library_id,
+                    voice_library_id=variant_voice_library_id,
                     template_id=request.template_id,
                     asset_ids=resolved_asset_ids,
                     title_ids=request.title_ids,
-                    voice_ids=request.voice_ids,
                     created_by_user_id=user_id,
                     source_edit_plan_id=effective_plan_id,
                     asset_select_mode=request.asset_select_mode,
@@ -495,14 +685,28 @@ def create_generation_task(
                     source_task_id=request.source_task_id,
                     output_width=request.output_width,
                     output_height=request.output_height,
-                    cover_url=request.cover_url,
-                    title_config=request.title_config or {},
+                    cover_url=variant_cover_url,
+                    title_config=variant_title_config,
                 )
             )
+            # 变体序号写入 extra_meta（响应/排查时可辨识）
+            task.extra_meta["variant_index"] = task_index
             try:
                 # 兜底关联编辑计划：前端未传 source_edit_plan_id 时，
                 # 通过 template_id + user_id 在 DB 层直接查找最新的 plan。
-                # 必须在 enqueue 之前执行，避免 worker 读取时 source_edit_plan_id 为空（竞态条件）
+                # 必须在 enqueue 之前执行，避免 worker 读取时 source_edit_plan_id 为空（竞态条件）。
+                # #1743：批量（count>1）场景严禁兜底共用——变体 plan 已在上方预生成，
+                # 走到这里还缺 plan 说明预生成漏配，直接报错中断，不允许 N 任务关联同一 plan。
+                if not task.source_edit_plan_id and count > 1:
+                    logger.error(
+                        "[生成任务] 批量任务缺少独立 plan（禁止共用兜底）: task_index=%d task_id=%s",
+                        task_index,
+                        task.id,
+                    )
+                    raise HTTPException(
+                        status_code=500,
+                        detail="创建批量任务失败：变体剪辑计划缺失，请重新预览后再批量生成。",
+                    )
                 if not task.source_edit_plan_id and request.template_id:
                     try:
                         from packages.adapters.sqlalchemy_impl.models import EditPlanModel
@@ -533,13 +737,13 @@ def create_generation_task(
 
                 # 回写 plan.config：必须在 enqueue 之前执行，
                 # 确保 worker 读取 plan 时 config 中已包含 generation_task_id。
-                # 只在首个任务时回写一次，避免批量生成时循环覆盖。
+                # 批量场景下每个变体关联独立 plan，需各自回写自己的变体标题配置。
                 _effective_plan_id = task.source_edit_plan_id
-                if _effective_plan_id and len(created_tasks) == 0:
+                if _effective_plan_id:
                     _writeback_edit_plan_config(
                         plan_id=_effective_plan_id,
                         task_id=task.id,
-                        title_config=request.title_config,
+                        title_config=variant_title_config,
                         db=db,
                     )
 
@@ -559,7 +763,7 @@ def create_generation_task(
                 if not created_tasks:
                     raise HTTPException(
                         status_code=429,
-                        detail="您的待处理任务过多，请等待完成后再提交",
+                        detail=build_rate_limit_detail(_e, generation_task_repository, scope="user"),
                     ) from _e
                 break
             except GlobalQueueFull as _e:
@@ -567,7 +771,7 @@ def create_generation_task(
                 if not created_tasks:
                     raise HTTPException(
                         status_code=503,
-                        detail="系统繁忙，请稍后再试",
+                        detail=build_rate_limit_detail(_e, generation_task_repository, scope="global"),
                     ) from _e
                 break
     except HTTPException:
@@ -669,7 +873,6 @@ def confirm_generation(
             template_id=source_task.template_id,
             asset_ids=source_task.asset_ids,
             title_ids=source_task.title_ids,
-            voice_ids=source_task.voice_ids,
             created_by_user_id=authenticated_user.user.id,
             source_edit_plan_id=source_task.source_edit_plan_id or "",
             asset_select_mode=source_task.asset_select_mode,
@@ -693,15 +896,15 @@ def confirm_generation(
             log_task_status=True,
         ):
             logger.warning("[确认生成] 入队失败: task_id=%s", new_task.id)
-    except UserPendingLimitExceeded:
+    except UserPendingLimitExceeded as _e:
         raise HTTPException(
             status_code=429,
-            detail="您的待处理任务过多，请等待完成后再提交",
+            detail=build_rate_limit_detail(_e, generation_task_repository, scope="user"),
         ) from None
-    except GlobalQueueFull:
+    except GlobalQueueFull as _e:
         raise HTTPException(
             status_code=503,
-            detail="系统繁忙，请稍后再试",
+            detail=build_rate_limit_detail(_e, generation_task_repository, scope="global"),
         ) from None
 
     return BatchGenerationTaskResponse(
@@ -783,12 +986,24 @@ def retry_generation_task(
     if user_pending >= USER_PENDING_LIMIT:
         raise HTTPException(
             status_code=429,
-            detail=f"您的待处理任务过多（当前 {user_pending}/{USER_PENDING_LIMIT}），请等待完成后再提交",
+            detail=build_rate_limit_detail(
+                UserPendingLimitExceeded(
+                    user_id=user_id,
+                    pending_count=user_pending,
+                    limit=USER_PENDING_LIMIT,
+                ),
+                generation_task_repository,
+                scope="user",
+            ),
         )
     if global_pending >= GLOBAL_PENDING_LIMIT:
         raise HTTPException(
             status_code=503,
-            detail="系统繁忙，请稍后再试",
+            detail=build_rate_limit_detail(
+                GlobalQueueFull(pending_count=global_pending, limit=GLOBAL_PENDING_LIMIT),
+                generation_task_repository,
+                scope="global",
+            ),
         )
 
     use_case = CreateGenerationTaskUseCase(generation_task_repository)
@@ -801,7 +1016,6 @@ def retry_generation_task(
             template_id=task.template_id,
             asset_ids=task.asset_ids,
             title_ids=task.title_ids,
-            voice_ids=task.voice_ids,
             created_by_user_id=user_id,
             source_edit_plan_id=task.source_edit_plan_id or "",
             asset_select_mode=getattr(task, "asset_select_mode", ""),
@@ -823,15 +1037,15 @@ def retry_generation_task(
             log_task_status=True,
         ):
             logger.warning("[生成任务] 重试入队失败: task_id=%s", retried.id)
-    except UserPendingLimitExceeded:
+    except UserPendingLimitExceeded as _e:
         raise HTTPException(
             status_code=429,
-            detail="您的待处理任务过多，请等待完成后再提交",
+            detail=build_rate_limit_detail(_e, generation_task_repository, scope="user"),
         ) from None
-    except GlobalQueueFull:
+    except GlobalQueueFull as _e:
         raise HTTPException(
             status_code=503,
-            detail="系统繁忙，请稍后再试",
+            detail=build_rate_limit_detail(_e, generation_task_repository, scope="global"),
         ) from None
     return _to_generation_task_response(retried)
 

apps/api/app/api/routes/generation_variant_plans.py

@@ -0,0 +1,178 @@
+"""轻量选片接口 POST /generation/variant-plans（#1749）。
+
+与正式生成共用同一套选片函数（EditPlanService.ensure_variant_plans →
+clone_plan_for_variant / reselect_plan_for_variant → variant_plan_selector），
+但**不建任务、不入队、不渲染**：
+
+- 仅为 N 个变体创建/选好 EditPlan + clips，返回 plan_id 与片段列表；
+- 前端确认后调正式生成接口回传 variant_plan_ids，直接复用这些 plan，
+  不再重复选片（回传后仍按各变体配音幂等重分配段长）；
+- 配音守卫：voice_library_ids 长度/缺值 → 400（variant_voice_resolver），
+  禁静默 fallback；
+- 素材不足等选片失败 → 400（与正式生成同口径）；除此之外不报错打断。
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any
+
+from app.auth import AuthenticatedUser, get_current_user
+from app.dependencies import get_db_session
+from fastapi import APIRouter, Depends, HTTPException
+from pydantic import BaseModel, Field, model_validator
+from sqlalchemy.orm import Session
+
+from packages.domain.variant_voice_resolver import VariantVoiceError, resolve_variant_voice_ids
+
+logger = logging.getLogger(__name__)
+
+router = APIRouter()
+
+
+class VariantPlanRequest(BaseModel):
+    """轻量选片请求体（与前端 variantPlans.ts 契约一致）。"""
+
+    template_id: str = Field(default="", description="模板 ID（无 source_edit_plan_id 时用于查找骨架 plan）")
+    asset_ids: list[str] = Field(default_factory=list, description="批次素材池")
+    count: int = Field(default=1, ge=1, le=50, description="变体数量")
+    source_edit_plan_id: str = Field(default="", description="源剪辑计划 ID（优先）")
+    # 配音（可选；传独立配音时严格守卫）
+    voice_library_id: str = Field(default="", description="统一配音 ID")
+    voice_library_ids: list[str] = Field(default_factory=list, description="独立配音 ID 列表（长度须=count）")
+
+    @model_validator(mode="after")
+    def _validate(self) -> "VariantPlanRequest":
+        if not self.template_id.strip() and not self.source_edit_plan_id.strip():
+            raise ValueError("template_id 与 source_edit_plan_id 至少需要提供一个")
+        try:
+            resolve_variant_voice_ids(
+                count=self.count,
+                voice_library_id=self.voice_library_id,
+                voice_library_ids=self.voice_library_ids or None,
+            )
+        except VariantVoiceError as exc:
+            raise ValueError(str(exc)) from exc
+        return self
+
+
+class VariantPlanItem(BaseModel):
+    variant_index: int
+    plan_id: str
+    clips: list[dict[str, Any]] = Field(default_factory=list)
+
+
+class VariantPlanResponse(BaseModel):
+    items: list[VariantPlanItem]
+    total: int
+
+
+@router.post("/variant-plans", response_model=VariantPlanResponse)
+def create_variant_plans(
+    request: VariantPlanRequest,
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    db: Session = Depends(get_db_session),
+) -> VariantPlanResponse:
+    """轻量选片：为 N 个变体创建独立 EditPlan + clips，不建任务/不渲染。
+
+    Returns:
+        200 + {items: [{variant_index, plan_id, clips}], total}
+    """
+    user_id = authenticated_user.user.id
+
+    # 配音严格守卫（schema 已校验，此处复用解析取每变体配音）
+    try:
+        voices = resolve_variant_voice_ids(
+            count=request.count,
+            voice_library_id=request.voice_library_id,
+            voice_library_ids=request.voice_library_ids or None,
+        )
+    except VariantVoiceError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+
+    # 解析源 plan：显式传入优先；否则按 template_id + user 查最新
+    source_plan_id = request.source_edit_plan_id.strip()
+    if not source_plan_id and request.template_id.strip():
+        try:
+            from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+
+            _latest = (
+                db.query(EditPlanModel)
+                .filter(
+                    EditPlanModel.template_id == request.template_id.strip(),
+                    EditPlanModel.created_by_user_id == user_id,
+                )
+                .order_by(EditPlanModel.created_at.desc())
+                .first()
+            )
+            if _latest:
+                source_plan_id = _latest.id
+        except Exception:
+            logger.warning("[variant-plans] 源 plan 解析失败", exc_info=True)
+
+    if not source_plan_id:
+        raise HTTPException(
+            status_code=400,
+            detail="缺少剪辑计划：请先完成一次预览生成（或传入 source_edit_plan_id）后再试。",
+        )
+
+    # 配音时长（硬化：异常 → 0.0 不阻断选片）
+    try:
+        from app.api.routes.generation_tasks import _query_voice_durations
+
+        voice_durations = _query_voice_durations(db, voices)
+    except Exception:
+        logger.warning("[variant-plans] 配音时长查询失败（按占位段长选片）", exc_info=True)
+        voice_durations = [0.0] * request.count
+
+    from app.services.edit_plan_service import EditPlanService
+
+    svc = EditPlanService(db)
+    try:
+        plan_ids = svc.ensure_variant_plans(
+            source_plan_id,
+            request.count,
+            list(dict.fromkeys(request.asset_ids or [])),
+            created_by_user_id=user_id,
+            voice_durations=voice_durations,
+        )
+    except ValueError as ve:
+        # 素材池为空/时长全未知等可预期错误 → 400（与正式生成同口径）
+        logger.warning("[variant-plans] 选片失败: %s", ve)
+        raise HTTPException(status_code=400, detail=f"变体选片失败：{ve}。请增加素材后重试。") from ve
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error("[variant-plans] 选片异常: %s", e, exc_info=True)
+        raise HTTPException(status_code=500, detail="选片失败，请稍后重试") from e
+
+    # 组装 clips 响应
+    items: list[VariantPlanItem] = []
+    for idx, pid in enumerate(plan_ids):
+        clips = svc.list_clips(pid)
+        clip_dicts = [
+            {
+                "id": c.id,
+                "order": c.order,
+                "asset_id": c.asset_id,
+                "start_time": float(c.start_time or 0.0),
+                "duration": float(c.duration or 0.0),
+                "clip_type": c.clip_type,
+                "transition_effect": c.transition_effect,
+                "transition_duration": float(c.transition_duration or 0.0),
+                "playback_speed": float(c.playback_speed or 1.0),
+                "text_content": c.text_content or "",
+                "status": c.status or "ready",
+            }
+            for c in clips
+        ]
+        items.append(VariantPlanItem(variant_index=idx, plan_id=pid, clips=clip_dicts))
+
+    logger.info(
+        "[variant-plans] 轻量选片完成: user=%s source=%s count=%d plans=%d",
+        user_id,
+        source_plan_id,
+        request.count,
+        len(plan_ids),
+    )
+    return VariantPlanResponse(items=items, total=len(items))

apps/api/app/api/routes/ingest_jobs.py

@@ -43,7 +43,13 @@ def submit_ingest_job(
         )
     )
 
-    celery_app.send_task("worker.ingest_asset", args=[job.id])
+    celery_result = celery_app.send_task("worker.ingest_asset", args=[job.id])
+    if getattr(celery_result, "id", ""):
+        try:
+            job.celery_task_id = celery_result.id
+            ingest_job_repository.update(job)
+        except Exception:  # noqa: BLE001
+            pass
 
     return IngestJobResponse(
         id=job.id,

apps/api/app/api/routes/lipsync.py

@@ -0,0 +1,199 @@
+"""对口型 API 路由 — #1796 MediaKit 对口型, #1809 参数调整.
+
+接口：
+  POST   /api/v1/lipsync/jobs          提交对口型任务
+  GET    /api/v1/lipsync/jobs          任务列表
+  GET    /api/v1/lipsync/jobs/{id}     任务详情
+  POST   /api/v1/lipsync/jobs/{id}/refresh  刷新任务状态
+  POST   /api/v1/lipsync/jobs/{id}/cancel   取消任务
+"""
+
+from __future__ import annotations
+
+import logging
+
+from app.auth import AuthenticatedUser, get_current_user
+from app.dependencies import (
+    get_cosyvoice_service,
+    get_db_session,
+    get_voice_clone_profile_repository,
+)
+from app.schemas.lipsync import CreateLipsyncJobRequest, LipsyncJobResponse
+from app.services.lipsync_service import LipsyncService
+from app.services.mediakit_client import MediaKitError
+from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Query
+from sqlalchemy.orm import Session
+
+from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
+
+logger = logging.getLogger(__name__)
+
+router = APIRouter()
+
+
+def _get_service(
+    db: Session = Depends(get_db_session),
+    voice_clone_repo=Depends(get_voice_clone_profile_repository),
+    cosyvoice_service: CosyVoiceService = Depends(get_cosyvoice_service),
+) -> LipsyncService:
+    # voice_clone_repo 用于克隆音色 profile 解析；cosyvoice_service 用于 TTS 直生
+    # （TTS 合成、音色解析、错误码归一化都在 LipsyncService 内部完成）
+    return LipsyncService(
+        db,
+        cosyvoice_service=cosyvoice_service,
+        voice_clone_repo=voice_clone_repo,
+    )
+
+
+# ── POST /jobs — 提交对口型任务 ───────────────────────────────────────────
+
+
+@router.post("/jobs", response_model=LipsyncJobResponse, status_code=201)
+def create_lipsync_job(
+    body: CreateLipsyncJobRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: LipsyncService = Depends(_get_service),
+):
+    """提交对口型任务.
+
+    #1809/#1822: 前端传 {video_url, voice_id, script_text, speed?, emotion?}，
+    后端内部解析音色、调 TTS 合成音频、转存 OSS，再提交 MediaKit；
+    也支持直接传 {video_url, audio_url}。
+    """
+    try:
+        job = svc.create_job(
+            user_id=current_user.user.id,
+            video_url=body.video_url,
+            audio_url=body.audio_url,
+            voice_id=body.voice_id,
+            script_text=body.script_text,
+            speed=body.speed,
+            emotion=body.emotion,
+            enable_video_loop=body.enable_video_loop,
+            project_id=body.project_id,
+        )
+    except ValueError as exc:
+        # 参数无效（如 voice_id 格式不对、文本过长等）
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+    except CosyVoiceError as exc:
+        # TTS 合成基础设施失败（API/网络/认证）
+        raise HTTPException(
+            status_code=502,
+            detail={"code": "TTSSynthesisFailed", "message": str(exc)},
+        ) from exc
+    except MediaKitError as exc:
+        # TTS 合成失败 / 音色无权访问 → 400/403；MediaKit 提交失败 → 502
+        status_code = 502
+        if exc.code in ("VoiceForbidden",):
+            status_code = 403
+        elif exc.code in ("InvalidInput", "TTSInvalidParam", "VoiceNotReady"):
+            status_code = 400
+        elif exc.code == "TTSSynthesisFailed":
+            status_code = 502
+        raise HTTPException(
+            status_code=status_code,
+            detail={
+                "code": exc.code,
+                "message": str(exc),
+                "request_id": getattr(exc, "request_id", ""),
+            },
+        ) from exc
+    except Exception as exc:
+        # 兜底：任何未预期的错误返回 400 而非 500
+        logger.error("创建对口型任务异常: %s", exc, exc_info=True)
+        raise HTTPException(
+            status_code=400,
+            detail=f"创建对口型任务失败: {exc}",
+        ) from exc
+
+    return job
+
+
+# ── GET /jobs — 任务列表 ─────────────────────────────────────────────────
+
+
+@router.get("/jobs", response_model=dict)
+def list_lipsync_jobs(
+    project_id: str = Query("", description="项目 ID 过滤"),
+    status: str = Query("", description="状态过滤"),
+    offset: int = Query(0, ge=0),
+    limit: int = Query(20, ge=1, le=100),
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: LipsyncService = Depends(_get_service),
+):
+    """获取对口型任务列表."""
+    items, total = svc.list_jobs(
+        user_id=current_user.user.id,
+        project_id=project_id,
+        status=status,
+        offset=offset,
+        limit=limit,
+    )
+    return {
+        "items": [LipsyncJobResponse.model_validate(j) for j in items],
+        "total": total,
+        "offset": offset,
+        "limit": limit,
+    }
+
+
+# ── GET /jobs/{job_id} — 任务详情 ────────────────────────────────────────
+
+
+@router.get("/jobs/{job_id}", response_model=LipsyncJobResponse)
+def get_lipsync_job(
+    job_id: str,
+    background: BackgroundTasks,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: LipsyncService = Depends(_get_service),
+):
+    """获取对口型任务详情.
+
+    非终态任务：先返回 DB 缓存，挂后台刷新（下次轮询拿到新状态），
+    避免 MediaKit 慢响应阻塞前端轮询。
+    """
+    job = svc.get_job(job_id, current_user.user.id)
+    if job is None:
+        raise HTTPException(status_code=404, detail="任务不存在")
+
+    if job.status not in ("completed", "failed"):
+        background.add_task(svc.refresh_job_status, job_id, current_user.user.id)
+
+    return job
+
+
+# ── POST /jobs/{job_id}/refresh — 刷新状态 ───────────────────────────────
+
+
+@router.post("/jobs/{job_id}/refresh", response_model=LipsyncJobResponse)
+def refresh_lipsync_job(
+    job_id: str,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: LipsyncService = Depends(_get_service),
+):
+    """从 MediaKit 拉取最新状态并更新."""
+    job = svc.refresh_job_status(job_id, current_user.user.id)
+    if job is None:
+        raise HTTPException(status_code=404, detail="任务不存在")
+    return job
+
+
+# ── POST /jobs/{job_id}/cancel — 取消任务 ────────────────────────────────
+
+
+@router.post("/jobs/{job_id}/cancel", response_model=LipsyncJobResponse)
+def cancel_lipsync_job(
+    job_id: str,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: LipsyncService = Depends(_get_service),
+):
+    """取消对口型任务（仅 pending/submitted 状态可取消）."""
+    job = svc.cancel_job(job_id, current_user.user.id)
+    if job is None:
+        raise HTTPException(status_code=404, detail="任务不存在")
+    if job.status != "cancelled":
+        raise HTTPException(
+            status_code=400,
+            detail=f"任务状态 {job.status} 不可取消，仅 pending/submitted 可取消",
+        )
+    return job

apps/api/app/api/routes/projects.py

@@ -1,13 +1,14 @@
 from typing import Any
 
 from app.auth import AuthenticatedUser, get_current_user
-from app.dependencies import get_project_repository
+from app.dependencies import get_asset_library_repository, get_project_repository
 from app.schemas.project import (
     CreateProjectRequest,
     ListProjectsResponse,
     ProjectResponse,
 )
 from fastapi import APIRouter, Depends, HTTPException, Response, status
+from pydantic import BaseModel
 
 from packages.application import (
     CreateProjectCommand,
@@ -16,10 +17,20 @@ from packages.application import (
     GetProjectUseCase,
     ListProjectsUseCase,
 )
+from packages.domain import AssetLibraryKind
 
 router = APIRouter()
 
 
+class DefaultContextResponse(BaseModel):
+    """幂等默认上下文响应（Issue #1775）：默认项目 + 各类型默认素材库 ID。"""
+
+    project_id: str
+    image_library_id: str
+    video_library_id: str
+    voice_library_id: str
+
+
 def _to_project_response(item) -> ProjectResponse:
     return ProjectResponse(
         id=item.id,
@@ -72,6 +83,35 @@ def create_project(
     return _to_project_response(project)
 
 
+@router.post("/ensure-default", response_model=DefaultContextResponse)
+def ensure_default_project_and_libraries(
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    project_repository: Any = Depends(get_project_repository),
+    asset_library_repository: Any = Depends(get_asset_library_repository),
+) -> DefaultContextResponse:
+    """幂等获取/创建当前用户的默认项目和三类默认素材库（Issue #1775）。
+
+    - 同一用户永远只有一个默认项目（部分唯一索引 uq_projects_owner_default）
+    - 同一项目同 kind 永远只有一个默认素材库（唯一约束 uq_asset_libraries_project_kind）
+    - 并发调用/失败重试：唯一约束冲突时返回已存在记录，不报 500
+    - 项目和素材库的创建各自在仓储事务内幂等，冲突回滚后重查返回同一条
+    """
+    user_id = authenticated_user.user.id
+    project = project_repository.get_or_create_default_project(user_id)
+
+    libraries = {}
+    for kind in (AssetLibraryKind.VIDEO, AssetLibraryKind.VOICE, AssetLibraryKind.IMAGE):
+        library = asset_library_repository.get_or_create_default_library(project.id, kind)
+        libraries[kind] = library.id
+
+    return DefaultContextResponse(
+        project_id=project.id,
+        image_library_id=libraries[AssetLibraryKind.IMAGE],
+        video_library_id=libraries[AssetLibraryKind.VIDEO],
+        voice_library_id=libraries[AssetLibraryKind.VOICE],
+    )
+
+
 @router.delete("/{project_id}", status_code=status.HTTP_204_NO_CONTENT, response_model=None, response_class=Response)
 def delete_project(
     project_id: str,

apps/api/app/api/routes/scripts.py

@@ -0,0 +1,123 @@
+"""Script (口播文案库) CRUD routes — Issue #1795."""
+
+from __future__ import annotations
+
+from typing import Optional
+
+from app.auth import AuthenticatedUser, get_current_user
+from app.dependencies import get_db_session
+from app.schemas.script import (
+    CreateScriptRequest,
+    ScriptListResponse,
+    ScriptResponse,
+    ScriptSegment,
+    UpdateScriptRequest,
+)
+from app.services.script_service import ScriptNotFoundError, ScriptService
+from fastapi import APIRouter, Depends, HTTPException, Query, Response, status
+from sqlalchemy.orm import Session
+
+router = APIRouter()
+
+
+def _get_service(session: Session = Depends(get_db_session)) -> ScriptService:
+    return ScriptService(session)
+
+
+def _to_response(script) -> ScriptResponse:
+    segments = script.segments or []
+    return ScriptResponse(
+        id=script.id,
+        user_id=script.user_id,
+        title=script.title,
+        content=script.content,
+        segments=[
+            ScriptSegment(text=s.get("text", ""), duration=s.get("duration")) if isinstance(s, dict) else s
+            for s in segments
+        ],
+        tags=script.tags or [],
+        created_at=script.created_at,
+        updated_at=script.updated_at,
+    )
+
+
+@router.get("", response_model=ScriptListResponse)
+def list_scripts(
+    skip: int = Query(0, ge=0),
+    limit: int = Query(50, ge=1, le=200),
+    tag: Optional[str] = Query(None, description="按标签筛选"),
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    svc: ScriptService = Depends(_get_service),
+) -> ScriptListResponse:
+    user_id = authenticated_user.user.id
+    items, total = svc.list_scripts(user_id, skip=skip, limit=limit, tag=tag)
+    return ScriptListResponse(
+        items=[_to_response(i) for i in items],
+        total=total,
+    )
+
+
+@router.post("", response_model=ScriptResponse, status_code=status.HTTP_201_CREATED)
+def create_script(
+    request: CreateScriptRequest,
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    svc: ScriptService = Depends(_get_service),
+) -> ScriptResponse:
+    user_id = authenticated_user.user.id
+    script = svc.create_script(
+        user_id=user_id,
+        title=request.title,
+        content=request.content,
+        segments=[s.model_dump() for s in request.segments],
+        tags=request.tags,
+    )
+    return _to_response(script)
+
+
+@router.get("/{script_id}", response_model=ScriptResponse)
+def get_script(
+    script_id: str,
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    svc: ScriptService = Depends(_get_service),
+) -> ScriptResponse:
+    user_id = authenticated_user.user.id
+    try:
+        script = svc.get_script(script_id, user_id)
+    except ScriptNotFoundError as exc:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Script not found") from exc
+    return _to_response(script)
+
+
+@router.put("/{script_id}", response_model=ScriptResponse)
+def update_script(
+    script_id: str,
+    request: UpdateScriptRequest,
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    svc: ScriptService = Depends(_get_service),
+) -> ScriptResponse:
+    user_id = authenticated_user.user.id
+    try:
+        script = svc.update_script(
+            script_id=script_id,
+            user_id=user_id,
+            title=request.title,
+            content=request.content,
+            segments=[s.model_dump() for s in request.segments] if request.segments is not None else None,
+            tags=request.tags,
+        )
+    except ScriptNotFoundError as exc:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Script not found") from exc
+    return _to_response(script)
+
+
+@router.delete("/{script_id}", status_code=status.HTTP_204_NO_CONTENT, response_model=None, response_class=Response)
+def delete_script(
+    script_id: str,
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    svc: ScriptService = Depends(_get_service),
+) -> Response:
+    user_id = authenticated_user.user.id
+    deleted = svc.delete_script(script_id, user_id)
+    if not deleted:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Script not found")
+    return

apps/api/app/api/routes/task_center.py

@@ -375,7 +375,13 @@ def retry_project_task(
                 storage_key=job.storage_key,
             )
         )
-        celery_app.send_task("worker.ingest_asset", args=[retried.id])
+        celery_result = celery_app.send_task("worker.ingest_asset", args=[retried.id])
+        if getattr(celery_result, "id", ""):
+            try:
+                retried.celery_task_id = celery_result.id
+                ingest_job_repository.update(retried)
+            except Exception:  # noqa: BLE001
+                pass
         return ProjectTaskResponse(
             id=f"ingest:{retried.id}",
             task_type="ingest",

apps/api/app/api/routes/templates.py

@@ -106,6 +106,10 @@ def list_templates(
     tag: str | None = Query(None, description="按标签筛选"),
     keyword: str | None = Query(None, description="按名称关键词搜索"),
     mode: str | None = Query(None, description="按剪辑模式筛选"),
+    valid_only: bool = Query(
+        False,
+        description="仅返回已配置片段的模板（剪辑页传 true；模板编辑器不传，可查看全部模板含草稿）",
+    ),
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     template_repository: SQLAlchemyTemplateRepository = Depends(_get_template_repository),
 ) -> ListTemplatesResponse:
@@ -116,6 +120,7 @@ def list_templates(
             tag=tag,
             keyword=keyword,
             mode=mode,
+            valid_only=valid_only,
         )
         use_case = ListTemplatesUseCase(template_repository)
         templates = use_case.execute(user_id, skip=skip, limit=limit, filter=tpl_filter)

apps/api/app/api/routes/templates_editor/clips.py

@@ -36,17 +36,11 @@ from app.services.asset_segment_tracker import (
     remove_used_segment,
 )
 from app.services.edit_plan_service import EditPlanService
-from app.services.edit_template_service import EditTemplateService
+from app.services.edit_template_service import EditTemplateService, TemplateNotFoundError
 from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Query, status
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl.asset_repository import SQLAlchemyAssetRepository
-from packages.adapters.sqlalchemy_impl.template_clip_config_repository import (
-    SQLAlchemyTemplateClipConfigRepository,
-)
-from packages.adapters.sqlalchemy_impl.template_repository import (
-    SQLAlchemyTemplateRepository,
-)
 from packages.domain.plan_generator_utils import (
     _calc_random_start_time,
     build_scene_segments,
@@ -399,68 +393,42 @@ def _safe_segment_duration(value, default: float) -> float:
 
 def _get_template_segments(
     template_id: str,
+    user_id: str,
     tpl_svc: EditTemplateService,
-    db: Session,
 ) -> list[tuple[int, float, float]]:
     """获取模板的片段配置（顺序、最短时长、最长时长）.
 
-    优先从新模板系统（template_clip_configs）查询，
-    若不存在则回退到旧模板系统（template_segments）。
+    单一数据源：模板主表为 ``templates``（用户自建，归属 user_id）/
+    ``edit_templates``（全局模板库），片段配置主表为 ``template_clip_configs``
+    （由 ``EditTemplateService.list_clip_configs_for_editor`` 统一读取）。
+
+    不再使用"新表抛异常 → 降级直查配置表 → 再降级查 segments"的异常控制流，
+    也不在正常请求中打印 ``ValueError: 模板不存在`` 堆栈。
+
+    Args:
+        template_id: 模板 ID
+        user_id: 当前登录用户 ID（用于归属校验）
+        tpl_svc: 模板编辑器服务
 
     Returns:
-        [(segment_order, duration_min, duration_max), ...] 按 order 排序
+        [(segment_order, duration_min, duration_max), ...] 按 order 排序；
+        模板存在但未配置片段时返回空列表。
+
+    Raises:
+        TemplateNotFoundError: 模板不存在、已删除或不归属于当前用户。
     """
-    # 优先查新模板系统
-    try:
-        clip_configs = tpl_svc.list_clip_configs(template_id)
-        if clip_configs:
-            result = []
-            for cc in clip_configs:
-                dur_min = _safe_segment_duration(cc.min_duration, _DEFAULT_EDITOR_CLIP_DURATION)
-                dur_max = _safe_segment_duration(
-                    cc.max_duration or cc.min_duration,
-                    _DEFAULT_EDITOR_CLIP_DURATION,
-                )
-                dur_min, dur_max = min(dur_min, dur_max), max(dur_min, dur_max)
-                result.append((cc.order, dur_min, dur_max))
-            return sorted(result, key=lambda x: x[0])
-    except Exception:
-        logger.warning("新模板系统查询clip_configs失败（主表可能不存在），直接查clip_configs表", exc_info=True)
+    clip_configs = tpl_svc.list_clip_configs_for_editor(template_id, user_id)
 
-    # 兜底：直接查 template_clip_configs 表（片段表有 template_id 外键，不依赖模板主表）
-    try:
-        direct_repo = SQLAlchemyTemplateClipConfigRepository(db)
-        direct_configs = direct_repo.list_by_template(template_id)
-        if direct_configs:
-            result = []
-            for cc in direct_configs:
-                dur_min = _safe_segment_duration(cc.min_duration, _DEFAULT_EDITOR_CLIP_DURATION)
-                dur_max = _safe_segment_duration(
-                    cc.max_duration or cc.min_duration,
-                    _DEFAULT_EDITOR_CLIP_DURATION,
-                )
-                dur_min, dur_max = min(dur_min, dur_max), max(dur_min, dur_max)
-                result.append((cc.order, dur_min, dur_max))
-            return sorted(result, key=lambda x: x[0])
-    except Exception:
-        logger.warning("直接查clip_configs表也失败，继续回退旧系统", exc_info=True)
-
-    # 回退到旧模板系统（template_segments表）
-    try:
-        old_repo = SQLAlchemyTemplateRepository(db)
-        segments = old_repo.list_segments(template_id)
-        if segments:
-            result = []
-            for s in segments:
-                dur_min = _safe_segment_duration(s.duration_min, _DEFAULT_EDITOR_CLIP_DURATION)
-                dur_max = _safe_segment_duration(s.duration_max, _DEFAULT_EDITOR_CLIP_DURATION)
-                dur_min, dur_max = min(dur_min, dur_max), max(dur_min, dur_max)
-                result.append((s.segment_order, dur_min, dur_max))
-            return sorted(result, key=lambda x: x[0])
-    except Exception:
-        logger.warning("旧模板系统查询segments失败", exc_info=True)
-
-    return []
+    result = []
+    for cc in clip_configs:
+        dur_min = _safe_segment_duration(cc.min_duration, _DEFAULT_EDITOR_CLIP_DURATION)
+        dur_max = _safe_segment_duration(
+            cc.max_duration or cc.min_duration,
+            _DEFAULT_EDITOR_CLIP_DURATION,
+        )
+        dur_min, dur_max = min(dur_min, dur_max), max(dur_min, dur_max)
+        result.append((cc.order, dur_min, dur_max))
+    return sorted(result, key=lambda x: x[0])
 
 
 def _recommended_time_conflicts(
@@ -663,13 +631,21 @@ def create_clips_from_assets_editor(
     7. 素材时长为 0 或缺失时报 400，不创建无效片段
     """
     tpl_svc, plan_svc = services
+    user_id = str(current_user.user.id)
 
-    # 1. 查询模板 segments
-    segments = _get_template_segments(template_id, tpl_svc, db)
+    # 1. 查询模板片段配置。模板不存在/已删除/无权限 → 404；
+    #    模板存在但确实未配置片段 → 422（配置错误，与 404 区分）。
+    try:
+        segments = _get_template_segments(template_id, user_id, tpl_svc)
+    except TemplateNotFoundError as exc:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="模板不存在或无权访问",
+        ) from exc
     if not segments:
         raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST,
-            detail="模板没有片段配置，无法创建片段",
+            status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
+            detail="模板未配置片段",
         )
 
     # 防御：schema validator 已过滤 null/空串，这里再归一化一次，

apps/api/app/api/routes/templates_editor/dependencies.py

@@ -41,29 +41,33 @@ def get_draft_plan_id(
     这是模板编辑器路由的核心依赖——所有编辑器端点都先经过这里，
     确保 template_id → plan_id 的映射始终存在。
 
-    兼容策略：优先从新模板系统（edit_templates 表）查找，
-    若不存在则回退到旧模板系统（templates 表），确保用户自建模板可用。
+    模板读取遵循单一数据源、显式判定（不使用异常降级）：
+    - 用户自建模板在旧表 ``templates``（归属 user_id，is_active=True）；
+    - 全局模板在新表 ``edit_templates``（无 user_id，全局可读）。
+    模板不存在、已删除或不归属于当前用户时，一律返回 404。
     """
     tpl_svc, plan_svc = services
     user_id = str(current_user.user.id)
 
+    # 0. 门禁：校验模板存在且可访问（即使草稿已缓存命中也要校验，
+    #    避免模板被删除/无权访问后仍可通过既有草稿 plan 继续操作）。
+    old_repo = SQLAlchemyTemplateRepository(db)
+    old_template = old_repo.get_active(template_id, user_id)
+    is_global_template = tpl_svc.get_template(template_id) is not None
+    if old_template is None and not is_global_template:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="模板不存在")
+
     # 1. 草稿已存在 → 直接返回
     draft = tpl_svc.get_template_draft(template_id)
     if draft is not None:
         return draft.id
 
-    # 2. 新系统有模板 → 用新服务创建草稿
-    if tpl_svc.get_template(template_id) is not None:
+    # 2. 全局模板（新系统）→ 用新服务创建草稿
+    if is_global_template:
         draft = tpl_svc.create_template_draft(template_id, user_id=user_id)
         return draft.id
 
-    # 3. 回退到旧模板系统（templates 表）
-    old_repo = SQLAlchemyTemplateRepository(db)
-    old_template = old_repo.get(template_id, user_id=user_id)
-    if old_template is None:
-        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="模板不存在")
-
-    # 4. 基于旧模板创建草稿计划
+    # 3. 旧模板（templates 表）→ 基于旧模板创建草稿计划
     from app.services.plan_generator_service import PlanGeneratorService
 
     from packages.domain.edit_template import EditTemplate, EditTemplateStatus

apps/api/app/api/routes/templates_editor/draft.py

@@ -150,7 +150,7 @@ def rollback_template(
     try:
         tpl = tpl_svc.rollback_to_version(template_id, request.version)
     except ValueError as exc:
-        raise HTTPException(status_code=400, detail=str(exc)) from exc
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(exc)) from exc
 
     clip_configs = tpl_svc.list_clip_configs(template_id)
     return EditorRollbackResponse(

apps/api/app/api/routes/tts.py

@@ -173,6 +173,14 @@ def synthesize(
         # job.voice_id 统一存解析后的 CosyVoice voice_id
         actual_voice_id = resolved_profile.voice_id
 
+    # 语速/情绪等合成参数随 metadata 落库，workflow 提交 CosyVoice 时读取透传
+    synthesis_meta = {
+        "speed": request.speed,
+        "emotion": request.emotion or "",
+    }
+    if request.metadata_:
+        synthesis_meta.update(request.metadata_)
+
     use_case = CreateTTSJobUseCase(repository)
     job = use_case.execute(
         user_id=user_id,
@@ -180,7 +188,7 @@ def synthesize(
         voice_id=actual_voice_id,
         voice_model=request.voice_model,
         voice_clone_profile_id=voice_clone_profile_id,
-        metadata=request.metadata_,
+        metadata=synthesis_meta,
     )
 
     # 提交 CosyVoice 合成任务
@@ -567,6 +575,7 @@ def preview_tts(
             text=request.text,
             voice_id=actual_voice_id,
             speed=request.speed,
+            emotion=request.emotion,
         )
     except CosyVoiceError as e:
         raise HTTPException(

apps/api/app/api/routes/upload.py

@@ -85,12 +85,22 @@ def _infer_mime_type_from_storage_key(storage_key: str) -> str:
     """从 storage_key 推断 MIME 类型（与 worker 端保持一致）。"""
     lower_filename = storage_key.rsplit("/", 1)[-1].lower()
     _MIME_MAP = {
-        ".mov": "video/quicktime", ".mp4": "video/mp4", ".avi": "video/x-msvideo",
-        ".mkv": "video/x-matroska", ".webm": "video/webm",
-        ".png": "image/png", ".gif": "image/gif", ".bmp": "image/bmp",
-        ".svg": "image/svg+xml", ".jpg": "image/jpeg", ".jpeg": "image/jpeg",
-        ".mp3": "audio/mpeg", ".wav": "audio/wav", ".ogg": "audio/ogg",
-        ".flac": "audio/flac", ".m4a": "audio/x-m4a",
+        ".mov": "video/quicktime",
+        ".mp4": "video/mp4",
+        ".avi": "video/x-msvideo",
+        ".mkv": "video/x-matroska",
+        ".webm": "video/webm",
+        ".png": "image/png",
+        ".gif": "image/gif",
+        ".bmp": "image/bmp",
+        ".svg": "image/svg+xml",
+        ".jpg": "image/jpeg",
+        ".jpeg": "image/jpeg",
+        ".mp3": "audio/mpeg",
+        ".wav": "audio/wav",
+        ".ogg": "audio/ogg",
+        ".flac": "audio/flac",
+        ".m4a": "audio/x-m4a",
     }
     for ext, mime in _MIME_MAP.items():
         if lower_filename.endswith(ext):
@@ -98,10 +108,139 @@ def _infer_mime_type_from_storage_key(storage_key: str) -> str:
     return "video/mp4"  # default
 
 
+# 兜底去重：无 file_hash / client_upload_id 且大小已知时，同库同名同大小近期活动记录视为重复
+FALLBACK_DEDUP_WINDOW_MINUTES = 30
+
+
+def _find_duplicate_asset(
+    asset_repository: Any,
+    *,
+    library_id: str,
+    file_hash: str,
+    client_upload_id: str,
+    filename: str,
+    file_size: int = 0,
+) -> Any:
+    """complete/上传幂等去重，按优先级查找已存在的素材。
+
+    1. client_upload_id（客户端幂等 token，同一次上传的重试保持一致）
+    2. file_hash（内容哈希，不同上传只要内容相同即去重）
+    3. 兜底（严格模式，宁可漏判不可误杀）：file_hash 与 client_upload_id
+       均缺失、且 file_size > 0 时，同库 + 同文件名 + **同大小** 且 30 分钟内
+       仍处 uploading/processing 的记录才判重。
+       - file_hash 非空时跳过兜底（hash 已代表内容；同名但内容全新的视频
+         如 iPhone 的 IMG_xxxx.MOV 绝不能被同名占位误杀）
+       - file_size=0（未知）时不允许仅凭同名 + processing 判重，直接放行
+
+    全部为鸭子类型调用：旧仓储无对应方法时静默跳过，不破坏既有实现。
+    """
+    if client_upload_id:
+        find = getattr(asset_repository, "find_by_library_and_client_upload_id", None)
+        if callable(find):
+            existing = find(library_id=library_id, client_upload_id=client_upload_id)
+            if existing is not None:
+                logger.info(
+                    "素材幂等命中(client_upload_id): library=%s token=%s asset=%s",
+                    library_id,
+                    client_upload_id,
+                    getattr(existing, "id", "?"),
+                )
+                return existing
+    if file_hash:
+        existing = asset_repository.find_by_library_and_file_hash(
+            library_id=library_id,
+            file_hash=file_hash,
+        )
+        if existing is not None:
+            logger.info(
+                "素材去重命中(file_hash): library=%s hash=%s asset=%s",
+                library_id,
+                file_hash,
+                existing.id,
+            )
+            return existing
+    # 同名兜底去重（最后防线，严格模式）：
+    # - 仅当 file_hash / client_upload_id 均缺失时启用（hash 能代表内容时不靠同名猜）
+    # - file_size 必须 > 0 且与记录大小严格一致；大小未知（0）直接放行
+    # - 只命中近期 UPLOADING/PROCESSING 活动记录（READY 历史素材不拦）
+    if filename and not file_hash and not client_upload_id and file_size and file_size > 0:
+        find_recent = getattr(asset_repository, "find_recent_active_by_library_and_name", None)
+        if callable(find_recent):
+            existing = find_recent(
+                library_id=library_id,
+                name=filename,
+                within_minutes=FALLBACK_DEDUP_WINDOW_MINUTES,
+                file_size=file_size,
+            )
+            if existing is not None:
+                logger.info(
+                    "素材幂等兜底命中(近期同名同大小活动记录): library=%s name=%s asset=%s status=%s size=%s",
+                    library_id,
+                    filename,
+                    getattr(existing, "id", "?"),
+                    getattr(existing, "status", None),
+                    file_size,
+                )
+                return existing
+    elif filename and not file_hash and not client_upload_id and not file_size:
+        logger.debug(
+            "同名兜底去重跳过(file_size 未知，宁可放行不可误杀): library=%s name=%s",
+            library_id,
+            filename,
+        )
+    return None
+
+
 def _create_pending_asset(
-    asset_repository, project_id, library_id, storage_key, filename, mime_type, user_id, file_hash=""
+    asset_repository,
+    project_id,
+    library_id,
+    storage_key,
+    filename,
+    mime_type,
+    user_id,
+    file_hash="",
+    client_upload_id="",
+    file_size: int = 0,
 ):
-    """立即创建一条 PROCESSING 状态的 Asset 记录，使前端能马上看到新素材。"""
+    """立即创建或复用一条 PROCESSING 状态的 Asset 记录。
+
+    find-or-create：prepare 阶段已按 file_hash/client_upload_id 预建的占位记录
+    会被 find_by_library_and_file_hash/find_by_library_and_client_upload_id 命中，
+    直接复用并补齐字段（避免 pre-create + complete 重复建两条）。
+
+    Issue #1776: 素材库计数由 asset_repository.create() 自动维护。
+    """
+    # 1. 按 client_upload_id / file_hash 查找现有记录
+    existing = None
+    if client_upload_id:
+        find_by_cuid = getattr(asset_repository, "find_by_library_and_client_upload_id", None)
+        if callable(find_by_cuid):
+            existing = find_by_cuid(library_id=library_id, client_upload_id=client_upload_id)
+    if existing is None and file_hash:
+        existing = asset_repository.find_by_library_and_file_hash(library_id=library_id, file_hash=file_hash)
+    if existing is not None:
+        # 补齐字段（幂等：避免重复建记录，前端已拿到 asset_id）
+        changed = False
+        if file_hash and not existing.file_hash:
+            existing.file_hash = file_hash
+            changed = True
+        if client_upload_id and not existing.client_upload_id:
+            existing.client_upload_id = client_upload_id
+            changed = True
+        if file_size and not existing.file_size:
+            existing.file_size = file_size
+            changed = True
+        if existing.status not in (AssetStatus.PROCESSING, AssetStatus.UPLOADING):
+            existing.status = AssetStatus.PROCESSING
+            changed = True
+        if changed:
+            try:
+                asset_repository.update(existing)
+            except Exception:  # noqa: BLE001 — 字段补齐失败不阻塞主流程
+                pass
+        return existing
+
     asset = Asset.create(
         project_id=project_id,
         library_id=library_id,
@@ -111,16 +250,30 @@ def _create_pending_asset(
         status=AssetStatus.PROCESSING,
         uploaded_by_user_id=user_id,
         file_hash=file_hash,
+        client_upload_id=client_upload_id,
+        file_size=file_size,
     )
     return asset_repository.create(asset)
 
 
+def _persist_celery_task_id(repo: Any, job: Any, celery_task_id: str) -> None:
+    """记录 celery 消息 ID 到任务行，供孤儿清理时 revoke/清除队列消息（#1714）。"""
+    if not celery_task_id:
+        return
+    try:
+        job.celery_task_id = celery_task_id
+        repo.update(job)
+    except Exception:  # noqa: BLE001 — 记录失败不影响主流程（执行前状态守卫兜底）
+        pass
+
+
 def _submit_ingest_job(
     project_id: str,
     library_id: str,
     storage_key: str,
     ingest_job_repository: Any,
     file_hash: str = "",
+    asset_id: str = "",
 ) -> Any:
     use_case = SubmitIngestJobUseCase(ingest_job_repository)
     job = use_case.execute(
@@ -129,9 +282,11 @@ def _submit_ingest_job(
             library_id=library_id,
             storage_key=storage_key,
             file_hash=file_hash,
+            asset_id=asset_id,
         )
     )
-    celery_app.send_task("worker.ingest_asset", args=[job.id])
+    celery_result = celery_app.send_task("worker.ingest_asset", args=[job.id])
+    _persist_celery_task_id(ingest_job_repository, job, getattr(celery_result, "id", ""))
     return job
 
 
@@ -141,9 +296,15 @@ async def prepare_direct_upload(
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     project_repository: Any = Depends(get_project_repository),
     asset_library_repository: Any = Depends(get_asset_library_repository),
+    asset_repository: Any = Depends(get_asset_repository),
     storage_service: OSSStorageService = Depends(get_storage_service),
 ) -> DirectUploadPrepareResponse:
-    """创建浏览器直传 OSS 的短期表单签名。"""
+    """创建浏览器直传 OSS 的短期表单签名，并在签名前按 file_hash/client_upload_id 去重。
+
+    命中去重：直接返回 duplicated=True + skip_transfer=True（前端跳过 OSS 直传），
+    未命中：正常签名 OSS 并立即预建一条 PROCESSING 状态的 asset 记录占住
+    file_hash 闸门，响应带 asset_id 供前端/后续 complete 关联。
+    """
     settings = get_settings()
     max_size_bytes = settings.OSS_DIRECT_UPLOAD_MAX_MB * 1024 * 1024
     if request.file_size > max_size_bytes:
@@ -162,8 +323,39 @@ async def prepare_direct_upload(
         asset_library_repository,
     )
 
-    file_id = uuid4().hex[:8]
     safe_filename = request.filename.replace("/", "_").replace("\\", "_")
+
+    # ── prepare 阶段去重：OSS 签名之前先查已存在素材 ──
+    if request.file_hash or request.client_upload_id:
+        existing = _find_duplicate_asset(
+            asset_repository,
+            library_id=request.library_id,
+            file_hash=request.file_hash,
+            client_upload_id=request.client_upload_id,
+            filename=request.filename,
+            file_size=request.file_size,
+        )
+        if existing is not None:
+            logger.info(
+                "prepare 命中去重: library=%s hash=%s cuid=%s existing_asset=%s",
+                request.library_id,
+                request.file_hash,
+                request.client_upload_id,
+                existing.id,
+            )
+            return DirectUploadPrepareResponse(
+                upload_url="",
+                method="",
+                storage_key=existing.storage_key,
+                expires_at="",
+                fields={},
+                max_size_bytes=0,
+                duplicated=True,
+                skip_transfer=True,
+                asset_id=existing.id,
+            )
+
+    file_id = uuid4().hex[:8]
     storage_key = f"uploads/{file_id}/{safe_filename}"
     try:
         payload = storage_service.create_direct_upload_post(
@@ -182,6 +374,28 @@ async def prepare_direct_upload(
             detail=f"Failed to prepare upload: {type(error).__name__}",
         ) from error
 
+    # ── 预建 asset 占位：占住 file_hash/client_upload_id 闸门，避免并发重复上传 ──
+    pending_asset_id = ""
+    if request.file_hash or request.client_upload_id:
+        try:
+            pending = _create_pending_asset(
+                asset_repository=asset_repository,
+                project_id=request.project_id,
+                library_id=request.library_id,
+                storage_key=storage_key,
+                filename=safe_filename,
+                mime_type=validated_content_type,
+                user_id=authenticated_user.user.id,
+                file_hash=request.file_hash,
+                client_upload_id=request.client_upload_id,
+                file_size=request.file_size,
+            )
+            pending_asset_id = pending.id
+            # Issue #1776: 计数由 asset_repository.create() 自动维护
+        except Exception as error:
+            # 预建失败不阻塞签名：complete 仍可按 OSS 文件 + hash 兜底去重
+            logger.warning("预建 asset 占位失败，降级走 old flow: %s", error)
+
     return DirectUploadPrepareResponse(
         upload_url=str(payload["url"]),
         method=str(payload["method"]),
@@ -189,6 +403,9 @@ async def prepare_direct_upload(
         expires_at=str(payload["expires_at"]),
         fields={str(key): str(value) for key, value in dict(payload["fields"]).items()},
         max_size_bytes=max_size_bytes,
+        duplicated=False,
+        skip_transfer=False,
+        asset_id=pending_asset_id,
     )
 
 
@@ -202,7 +419,7 @@ async def complete_direct_upload(
     asset_repository: Any = Depends(get_asset_repository),
     storage_service: OSSStorageService = Depends(get_storage_service),
 ) -> DirectUploadCompleteResponse:
-    """确认浏览器直传完成并创建导入任务。"""
+    """确认浏览器直传完成并创建导入任务（幂等：重复 complete 返回同一素材）。"""
     require_project_and_library(
         request.project_id,
         request.library_id,
@@ -212,6 +429,29 @@ async def complete_direct_upload(
     normalized_key = storage_service._normalize_storage_key(request.storage_key)
     if not normalized_key.startswith("uploads/"):
         raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid upload key")
+
+    filename = normalized_key.rsplit("/", 1)[-1]
+
+    # ── 幂等去重（放在 OSS 检查之前）：complete 超时后前端重试时，
+    # 第一次 complete 可能已建好占位记录，此时即使 OSS 检查失败也必须返回
+    # 已存在记录，绝不能再建第二条。─
+    existing = _find_duplicate_asset(
+        asset_repository,
+        library_id=request.library_id,
+        file_hash=request.file_hash,
+        client_upload_id=request.client_upload_id,
+        filename=filename,
+        file_size=request.file_size,
+    )
+    if existing is not None:
+        return DirectUploadCompleteResponse(
+            storage_key=existing.storage_key,
+            ingest_job_id="",
+            duplicated=True,
+            asset_id=existing.id,
+            url=storage_service.get_url(existing.storage_key),
+        )
+
     try:
         file_exists = storage_service.file_exists(normalized_key)
     except Exception as error:
@@ -223,29 +463,7 @@ async def complete_direct_upload(
     if not file_exists:
         raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Uploaded file not found")
 
-    # ── 素材去重检测：同素材库 + 同 file_hash 视为重复 ──
-    if request.file_hash:
-        existing = asset_repository.find_by_library_and_file_hash(
-            library_id=request.library_id,
-            file_hash=request.file_hash,
-        )
-        if existing is not None:
-            logger.info(
-                "素材去重命中: library=%s hash=%s existing_asset=%s",
-                request.library_id,
-                request.file_hash,
-                existing.id,
-            )
-            return DirectUploadCompleteResponse(
-                storage_key=normalized_key,
-                ingest_job_id="",
-                duplicated=True,
-                asset_id=existing.id,
-                url=storage_service.get_url(normalized_key),
-            )
-
     # 立即创建 Asset 记录（PROCESSING 状态），使前端刷新后即可看到新素材
-    filename = normalized_key.rsplit("/", 1)[-1]
     mime_type = _infer_mime_type_from_storage_key(normalized_key)
     pending_asset = _create_pending_asset(
         asset_repository=asset_repository,
@@ -256,7 +474,10 @@ async def complete_direct_upload(
         mime_type=mime_type,
         user_id=authenticated_user.user.id,
         file_hash=request.file_hash,
+        client_upload_id=request.client_upload_id,
+        file_size=request.file_size,
     )
+    # Issue #1776: 计数由 asset_repository.create() 自动维护
 
     job = _submit_ingest_job(
         project_id=request.project_id,
@@ -264,6 +485,7 @@ async def complete_direct_upload(
         storage_key=normalized_key,
         ingest_job_repository=ingest_job_repository,
         file_hash=request.file_hash,
+        asset_id=pending_asset.id,
     )
     return DirectUploadCompleteResponse(
         storage_key=normalized_key,
@@ -283,7 +505,8 @@ async def upload_asset(
     project_id: str = Form(..., min_length=1, description="项目 ID"),
     library_id: str = Form(..., min_length=1, description="素材库 ID"),
     file: UploadFile = File(..., description="要上传的文件（视频、音频、图片等）"),
-    file_hash: str = Form(default="", description="文件 MD5 哈希，用于去重检测"),
+    file_hash: str = Form(default="", description="文件哈希，用于去重检测"),
+    client_upload_id: str = Form(default="", description="客户端幂等 token（同一次上传的重试保持一致）"),
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     ingest_job_repository: Any = Depends(get_ingest_job_repository),
     project_repository: Any = Depends(get_project_repository),
@@ -294,32 +517,31 @@ async def upload_asset(
     """上传素材文件并触发导入流水线。"""
     require_project_and_library(project_id, library_id, project_repository, asset_library_repository)
 
-    # ── 素材去重检测：上传前检查同素材库 + 同 file_hash ──
-    if file_hash:
-        existing = asset_repository.find_by_library_and_file_hash(
-            library_id=library_id,
-            file_hash=file_hash,
-        )
-        if existing is not None:
-            logger.info(
-                "素材去重命中(multipart): library=%s hash=%s existing_asset=%s",
-                library_id,
-                file_hash,
-                existing.id,
-            )
-            return UploadAssetResponse(
-                storage_key=existing.storage_key,
-                ingest_job_id="",
-                url="",
-                duplicated=True,
-                asset_id=existing.id,
-            )
-
-    # P2-5: 服务端验证 MIME 类型
+    # P2-5: 服务端验证 MIME 类型（先验证，再幂等去重，避免非法类型绕过）
     validated_content_type = _validate_mime_type(file.content_type)
 
-    file_id = uuid4().hex[:8]
     safe_filename = file.filename.replace("/", "_").replace("\\", "_") if file.filename else "unknown"
+
+    # ── 幂等去重：client_upload_id → file_hash → 近期活动同名记录兜底 ──
+    # 放在 OSS 上传之前：重复提交直接返回，不占 OSS 流量、不建新记录。
+    existing = _find_duplicate_asset(
+        asset_repository,
+        library_id=library_id,
+        file_hash=file_hash,
+        client_upload_id=client_upload_id,
+        filename=safe_filename,
+        file_size=0,
+    )
+    if existing is not None:
+        return UploadAssetResponse(
+            storage_key=existing.storage_key,
+            ingest_job_id="",
+            url="",
+            duplicated=True,
+            asset_id=existing.id,
+        )
+
+    file_id = uuid4().hex[:8]
     storage_key = f"uploads/{file_id}/{safe_filename}"
 
     try:
@@ -348,7 +570,9 @@ async def upload_asset(
         mime_type=validated_content_type,
         user_id=authenticated_user.user.id,
         file_hash=file_hash,
+        client_upload_id=client_upload_id,
     )
+    # Issue #1776: 计数由 asset_repository.create() 自动维护
 
     job = _submit_ingest_job(
         project_id=project_id,
@@ -356,6 +580,7 @@ async def upload_asset(
         storage_key=storage_key,
         ingest_job_repository=ingest_job_repository,
         file_hash=file_hash,
+        asset_id=pending_asset.id,
     )
 
     return UploadAssetResponse(

apps/api/app/api/routes/videos.py

@@ -252,6 +252,10 @@ class RecomputeDedupRequest(BaseModel):
         None,
         description="指定视频 ID 列表。为空则对当前用户所有缺少查重数据的视频重新计算。",
     )
+    force: bool = Field(
+        False,
+        description="强制重算：即使视频已有查重数据也重新入队（#1702 查重算法升级后用于存量视频重算）。",
+    )
 
 
 class RecomputeDedupResponse(BaseModel):
@@ -291,15 +295,15 @@ def recompute_dedup(
     skipped = 0
 
     for video in target_videos:
-        # 已有完整查重数据的跳过
-        if video.duplicate_rate is not None and video.video_fingerprint:
+        # 已有完整查重数据的跳过（force=True 时强制重算，#1702 算法升级后存量视频需要重算指纹/分片）
+        if not request.force and video.duplicate_rate is not None and video.video_fingerprint:
             skipped += 1
             continue
 
         # 触发异步查重任务
         celery_app.send_task("worker.check_duplicate", args=[video.id])
         enqueued += 1
-        logger.info("Enqueued re-dedup for video %s (user=%s)", video.id, user_id)
+        logger.info("Enqueued re-dedup for video %s (user=%s, force=%s)", video.id, user_id, request.force)
 
     return RecomputeDedupResponse(
         enqueued=enqueued,

apps/api/app/core/celery_app.py

@@ -5,3 +5,11 @@ settings = get_settings()
 celery_app = Celery("xiaoxia-saas-api")
 celery_app.conf.broker_url = settings.CELERY_BROKER_URL
 celery_app.conf.result_backend = settings.CELERY_RESULT_BACKEND
+
+# #1714 队列隔离：视频生成走 generation 队列，素材转码走 transcode 队列
+try:
+    from packages.shared.celery_queues import apply_queue_settings
+
+    apply_queue_settings(celery_app)
+except Exception:  # noqa: BLE001 — 队列配置失败不阻断 API 启动
+    pass

apps/api/app/core/task_enqueue.py

@@ -8,27 +8,145 @@ logger = logging.getLogger(__name__)
 # ── 限流阈值常量（全系统统一管理，不要在业务代码里硬编码） ──
 USER_PENDING_LIMIT = 3  # 单用户 pending 上限
 GLOBAL_PENDING_LIMIT = 20  # 全局 pending 上限
+WORKER_CONCURRENCY = 4  # worker 渲染并发数（infra/docker/compose.yml WORKER_CONCURRENCY 默认值）
+
+# 限流错误码：前端据此区分"排队等待"与"创建失败"
+ERROR_CODE_USER_QUEUE_FULL = "USER_QUEUE_FULL"  # 429：用户自己的任务排队中
+ERROR_CODE_SYSTEM_QUEUE_FULL = "SYSTEM_QUEUE_FULL"  # 503：系统整体繁忙
 
 
 class UserPendingLimitExceeded(Exception):
     """用户 pending 任务数超限，返回 429。"""
 
-    def __init__(self, user_id: str, pending_count: int, limit: int):
+    def __init__(
+        self,
+        user_id: str,
+        pending_count: int,
+        limit: int,
+        *,
+        running_count: int = 0,
+        requested_count: int = 1,
+        queue_ahead: int = 0,
+        estimated_wait_seconds: int = 0,
+    ):
         self.user_id = user_id
         self.pending_count = pending_count
         self.limit = limit
+        # 排队上下文（用于 429 结构化提示，前端展示"排队中"而非"创建失败"）
+        self.running_count = running_count
+        self.requested_count = requested_count
+        self.queue_ahead = queue_ahead
+        self.estimated_wait_seconds = estimated_wait_seconds
         super().__init__(f"用户 {user_id} pending 任务数 {pending_count} 超过上限 {limit}")
 
 
 class GlobalQueueFull(Exception):
     """全局限流，返回 503。"""
 
-    def __init__(self, pending_count: int, limit: int):
+    def __init__(
+        self,
+        pending_count: int,
+        limit: int,
+        *,
+        running_count: int = 0,
+        queue_ahead: int = 0,
+        estimated_wait_seconds: int = 0,
+    ):
         self.pending_count = pending_count
         self.limit = limit
+        self.running_count = running_count
+        self.queue_ahead = queue_ahead
+        self.estimated_wait_seconds = estimated_wait_seconds
         super().__init__(f"系统 pending 任务数 {pending_count} 超过上限 {limit}")
 
 
+def _estimate_wait_seconds(queue_ahead: int, generation_task_repository: Any) -> int:
+    """根据排队任务数 + worker 并发数 + 历史平均任务耗时估算等待秒数。
+
+    估算公式：ceil(排队任务数 / 并发数) × 平均单任务耗时。
+    拿不到历史数据时仓储层返回默认 120 秒。
+    """
+    import math
+
+    if queue_ahead <= 0:
+        return 0
+    try:
+        estimator = getattr(generation_task_repository, "estimate_avg_duration_seconds", None)
+        avg_seconds = estimator() if estimator is not None else 120.0
+    except Exception:
+        avg_seconds = 120.0
+    return int(math.ceil(queue_ahead / WORKER_CONCURRENCY) * avg_seconds)
+
+
+def build_rate_limit_detail(
+    exc: Exception,
+    generation_task_repository: Any,
+    *,
+    scope: str = "user",
+) -> dict:
+    """构造结构化限流响应体（HTTPException 的 detail）。
+
+    前端按 detail.code 判断场景：
+    - USER_QUEUE_FULL (429)：用户自己的任务在排队，应提示"等待/继续排队"，不是创建失败
+    - SYSTEM_QUEUE_FULL (503)：系统繁忙，稍后重试
+
+    detail 字段：
+    - code: 错误码
+    - message: 可读中文提示（可直接展示）
+    - queued_count: 当前排队(pending)任务数
+    - running_count: 当前渲染中(running)任务数
+    - queue_ahead: 前方排队任务数（预计等待批次依据）
+    - estimated_wait_seconds: 预计等待秒数
+    - limit: 对应限流上限
+    """
+    if scope == "user" and isinstance(exc, UserPendingLimitExceeded):
+        running = exc.running_count
+        if not running:
+            try:
+                counter = getattr(generation_task_repository, "count_running_by_user", None)
+                running = counter(exc.user_id) if counter is not None else 0
+            except Exception:
+                running = 0
+        queue_ahead = exc.queue_ahead or max(exc.pending_count, 0)
+        wait = exc.estimated_wait_seconds or _estimate_wait_seconds(queue_ahead, generation_task_repository)
+        wait_minutes = max(1, round(wait / 60))
+        message = (
+            f"您有 {exc.pending_count} 个任务正在排队、{running} 个正在渲染，"
+            f"同一时间最多提交 {exc.limit} 个任务。请等待约 {wait_minutes} 分钟后再提交"
+        )
+        return {
+            "code": ERROR_CODE_USER_QUEUE_FULL,
+            "message": message,
+            "queued_count": exc.pending_count,
+            "running_count": running,
+            "queue_ahead": queue_ahead,
+            "estimated_wait_seconds": wait,
+            "limit": exc.limit,
+        }
+
+    # 全局繁忙
+    pending = getattr(exc, "pending_count", 0)
+    running = getattr(exc, "running_count", 0)
+    if not running:
+        try:
+            counter = getattr(generation_task_repository, "count_running_total", None)
+            running = counter() if counter is not None else 0
+        except Exception:
+            running = 0
+    queue_ahead = getattr(exc, "queue_ahead", 0) or pending
+    wait = getattr(exc, "estimated_wait_seconds", 0) or _estimate_wait_seconds(queue_ahead, generation_task_repository)
+    wait_minutes = max(1, round(wait / 60))
+    return {
+        "code": ERROR_CODE_SYSTEM_QUEUE_FULL,
+        "message": f"系统繁忙：当前 {pending} 个任务排队中、{running} 个渲染中，预计等待约 {wait_minutes} 分钟，请稍后再试",
+        "queued_count": pending,
+        "running_count": running,
+        "queue_ahead": queue_ahead,
+        "estimated_wait_seconds": wait,
+        "limit": getattr(exc, "limit", GLOBAL_PENDING_LIMIT),
+    }
+
+
 def check_queue_limits(
     user_id: str,
     generation_task_repository: Any,
@@ -161,7 +279,17 @@ def safe_enqueue_generation_task(
 
     # ── 发送 Celery 任务 ──
     try:
-        celery_app.send_task("worker.generate_video", args=[task.id])
+        celery_result = celery_app.send_task("worker.generate_video", args=[task.id])
+        # 记录 celery 消息 ID：孤儿清理/超时作废时据此 revoke + 清除队列消息（#1714）
+        celery_task_id = getattr(celery_result, "id", "")
+        if celery_task_id:
+            try:
+                task.celery_task_id = celery_task_id
+                generation_task_repository.update(task)
+            except Exception as persist_err:  # noqa: BLE001
+                logger.warning(
+                    "%s 持久化 celery_task_id 失败（不影响主流程）: task_id=%s err=%s", log_prefix, task.id, persist_err
+                )
     except Exception as e:
         logger.error(
             "%s 入队失败，标记为失败: task_id=%s error=%s",

apps/api/app/schemas/ai_avatar_render.py

@@ -0,0 +1,124 @@
+"""AI数字人渲染合成管线 API Schema — #1798."""
+
+from __future__ import annotations
+
+from datetime import datetime
+from typing import Any, Optional
+
+from pydantic import BaseModel, Field, field_validator
+
+
+class BRollSegment(BaseModel):
+    """B-roll 片段配置."""
+
+    script_segment_index: int = Field(..., ge=0, description="对应文案片段索引")
+    asset_url: str = Field(..., description="B-roll 素材 URL")
+    mode: str = Field(..., description="插入模式: fullscreen 或 pip")
+    start_time: float = Field(..., ge=0.0, description="在对口型视频中的起始时间(秒)")
+    end_time: float = Field(..., ge=0.0, description="在对口型视频中的结束时间(秒)")
+    pip_position: Optional[str] = Field("bottom_right", description="pip 模式位置")
+    pip_scale: Optional[float] = Field(0.3, ge=0.05, le=1.0, description="pip 模式缩放比例")
+
+    @field_validator("mode")
+    @classmethod
+    def validate_mode(cls, v: str) -> str:
+        v = v.strip().lower()
+        if v not in ("fullscreen", "pip"):
+            raise ValueError("mode 必须为 fullscreen 或 pip")
+        return v
+
+    @field_validator("asset_url")
+    @classmethod
+    def validate_asset_url(cls, v: str) -> str:
+        v = v.strip()
+        if not v:
+            raise ValueError("asset_url 不能为空")
+        if not v.startswith(("http://", "https://")):
+            raise ValueError("asset_url 必须是 HTTP/HTTPS URL")
+        return v
+
+    @field_validator("end_time")
+    @classmethod
+    def validate_end_time(cls, v: float, info: Any) -> float:
+        start = info.data.get("start_time", 0.0)
+        if v <= start:
+            raise ValueError("end_time 必须大于 start_time")
+        return v
+
+
+class CreateAiAvatarRenderRequest(BaseModel):
+    """创建渲染任务请求."""
+
+    lipsync_job_id: str = Field(..., description="对口型任务 ID")
+    script_id: str = Field("", description="文案 ID（选自文案库时传；手动输入文案直生场景可留空）")
+    b_roll_segments: list[BRollSegment] = Field(default_factory=list, description="B-roll 片段列表")
+    title_config: dict[str, Any] = Field(default_factory=dict, description="标题配置")
+    cover_config: dict[str, Any] = Field(default_factory=dict, description="封面配置")
+    project_id: str = Field("", description="项目 ID")
+
+    @field_validator("lipsync_job_id")
+    @classmethod
+    def validate_lipsync_job_id(cls, v: str) -> str:
+        v = v.strip()
+        if not v:
+            raise ValueError("lipsync_job_id 不能为空")
+        return v
+
+    @field_validator("script_id")
+    @classmethod
+    def validate_script_id(cls, v: str) -> str:
+        # script_id 可选：手动输入文案（TTS 直生）场景不关联文案库条目
+        return (v or "").strip()
+
+
+class AiAvatarRenderJobResponse(BaseModel):
+    """渲染任务响应."""
+
+    id: str
+    user_id: str
+    project_id: str
+    lipsync_job_id: str
+    script_id: str = ""
+    b_roll_segments: list[dict[str, Any]]
+    title_config: dict[str, Any]
+    cover_config: dict[str, Any]
+    status: str
+    progress: int
+    output_video_url: str
+    output_cover_url: str
+    output_duration: float
+    error_message: str
+    submitted_at: Optional[datetime] = None
+    started_at: Optional[datetime] = None
+    completed_at: Optional[datetime] = None
+    created_at: datetime
+    updated_at: datetime
+
+    class Config:
+        from_attributes = True
+
+
+class AiAvatarRenderProgressResponse(BaseModel):
+    """渲染进度响应."""
+
+    status: str
+    progress: int
+    output_video_url: str
+    output_cover_url: str
+    output_duration: float
+    error_message: str
+
+
+class SmartCoverRequest(BaseModel):
+    """智能封面请求 — MediaKit 抽帧 + 质量评分选最佳帧."""
+
+    video_url: str = Field(..., description="数字人视频 URL（对口型/渲染成片）")
+    max_frames: int = Field(5, ge=1, le=10, description="抽帧数量（默认 5）")
+
+
+class SmartCoverResponse(BaseModel):
+    """智能封面响应."""
+
+    cover_url: str = Field("", description="封面图公网 URL（OSS，非临时）；失败为空")
+    status: str = Field("completed", description="completed / fallback_failed")
+    message: str = Field("", description="失败原因（如有）")

apps/api/app/schemas/generation_task.py

@@ -25,14 +25,24 @@ class CreateGenerationTaskRequest(BaseModel):
     asset_library_id: str = ""
     strategy_id: str = ""
     voice_library_id: str = ""
+    # ── 多变体独立配音（批量生成）──
+    # 长度 1 = 所有变体共用；长度 = count = 每个变体独立配音；空数组 = 回退 voice_library_id
+    voice_library_ids: list[str] = Field(
+        default_factory=list,
+        description="各变体独立配音素材库ID数组：长度1=共用，长度=count=独立。为空时回退 voice_library_id",
+    )
     created_by_user_id: str = ""
     # ── 模板模式新增字段 ──
     template_id: str = ""
     asset_ids: list[str] = Field(default_factory=list)
     title_ids: list[str] = Field(default_factory=list)
-    voice_ids: list[str] = Field(default_factory=list)
     # ── 来源剪辑计划 ──
     source_edit_plan_id: str = ""
+    # ── variant-plans 轻量选片回传（#1749）：正式生成直接复用，不再重选 ──
+    variant_plan_ids: list[str] = Field(
+        default_factory=list,
+        description="POST /generation/variant-plans 返回的各变体 plan_id（长度须=count）；为空则走服务端选片",
+    )
     # ── 标题配置（结构化）──
     title_config: dict | None = Field(
         default=None,
@@ -75,6 +85,47 @@ class CreateGenerationTaskRequest(BaseModel):
     output_width: int = Field(default=1280, description="输出视频宽度")
     output_height: int = Field(default=720, description="输出视频高度")
     cover_url: str = Field(default="", description="封面图片 URL")
+    # ── 多变体独立封面（批量生成）──
+    # 长度 1 = 所有变体共用；长度 = count = 每个变体独立封面；空数组 = 回退 cover_url
+    cover_urls: list[str] = Field(
+        default_factory=list,
+        description="各变体独立封面URL数组：长度1=共用，长度=count=独立。为空时回退 cover_url",
+    )
+    # ── 多变体独立标题文字（批量生成）──
+    # 长度 1 = 所有变体共用；长度 = count = 每个变体独立标题文字；空数组 = 使用 title_config.text
+    titles: list[str] = Field(
+        default_factory=list,
+        description="各变体独立标题文字数组：长度1=共用，长度=count=独立。为空时使用 title_config.text",
+    )
+
+    @model_validator(mode="after")
+    def _check_variant_arrays(self) -> "CreateGenerationTaskRequest":
+        """变体数组字段长度校验 + #1749 配音严格守卫。
+
+        - cover_urls/titles：空（回退单值）、长度 1（共用）或长度 = count（独立）；
+        - voice_library_ids：独立配音长度必须恰好 = count 且逐项非空，禁止静默 fallback
+          （长度 1 的"共用"场景请用 voice_library_id 单值字段）；
+        - variant_plan_ids：非空时长度必须 = count。
+        """
+        for name in ("cover_urls", "titles"):
+            arr = getattr(self, name)
+            if arr and len(arr) != 1 and len(arr) != self.count:
+                raise ValueError(f"{name} 长度必须为 1（共用）或 {self.count}（与 count 一致），当前为 {len(arr)}")
+
+        from packages.domain.variant_voice_resolver import VariantVoiceError, resolve_variant_voice_ids
+
+        try:
+            resolve_variant_voice_ids(
+                count=self.count,
+                voice_library_id=self.voice_library_id,
+                voice_library_ids=self.voice_library_ids or None,
+            )
+        except VariantVoiceError as exc:
+            raise ValueError(str(exc)) from exc
+
+        if self.variant_plan_ids and len(self.variant_plan_ids) != self.count:
+            raise ValueError(f"variant_plan_ids 长度({len(self.variant_plan_ids)})必须与 count({self.count})一致")
+        return self
 
     @model_validator(mode="after")
     def _check_at_least_one_mode(self) -> "CreateGenerationTaskRequest":
@@ -83,9 +134,9 @@ class CreateGenerationTaskRequest(BaseModel):
         if not has_project and not has_template:
             raise ValueError("project_id 或 template_id 至少需要提供一个")
         has_library = bool(self.asset_library_id.strip())
-        has_assets = bool(self.asset_ids or self.title_ids or self.voice_ids)
+        has_assets = bool(self.asset_ids or self.title_ids)
         if not has_library and not has_assets:
-            raise ValueError("asset_library_id 或 asset_ids/title_ids/voice_ids 至少需要提供一个")
+            raise ValueError("asset_library_id 或 asset_ids/title_ids 至少需要提供一个")
         return self
 
 
@@ -162,7 +213,6 @@ class CreatePreviewGenerationTaskRequest(BaseModel):
     template_id: str
     asset_ids: list[str] = Field(default_factory=list)
     title_ids: list[str] = Field(default_factory=list)
-    voice_ids: list[str] = Field(default_factory=list)
     voice_library_id: str = Field(
         default="", description="配音素材库ID（用户上传的音频或AI配音），对应配音选择页面选择的配音素材"
     )
@@ -185,8 +235,44 @@ class CreatePreviewGenerationTaskRequest(BaseModel):
     )
     title_config: dict = Field(
         default_factory=dict,
-        description="标题配置（可选），渲染时烧录到预览视频中。支持字段: text/font/font_size/font_color/position/bold/stroke/shadow",
+        description="标题配置（可选），渲染时烧录到预览视频中。支持字段: text/font/font_size/font_color/position/bold/stroke/shadow。N个变体时样式全局共用",
     )
+    # ── 多变体独立配置（preview_count > 1）──
+    # 长度 1 = 所有变体共用；长度 = preview_count = 每个变体独立；空数组 = 回退单值字段
+    titles: list[str] = Field(
+        default_factory=list,
+        description="各变体独立标题文字数组：长度1=共用，长度=preview_count=独立。为空时使用 title_config.text",
+    )
+    voice_library_ids: list[str] = Field(
+        default_factory=list,
+        description="各变体独立配音素材库ID数组：长度1=共用，长度=preview_count=独立。为空时回退 voice_library_id",
+    )
+    cover_urls: list[str] = Field(
+        default_factory=list,
+        description="各变体独立封面URL数组：长度1=共用，长度=preview_count=独立（预览阶段通常为空）",
+    )
+
+    @model_validator(mode="after")
+    def _check_variant_arrays(self) -> "CreatePreviewGenerationTaskRequest":
+        """变体数组字段长度校验 + #1749 配音严格守卫。"""
+        for name in ("titles", "cover_urls"):
+            arr = getattr(self, name)
+            if arr and len(arr) != 1 and len(arr) != self.preview_count:
+                raise ValueError(
+                    f"{name} 长度必须为 1（共用）或 {self.preview_count}（与 preview_count 一致），当前为 {len(arr)}"
+                )
+
+        from packages.domain.variant_voice_resolver import VariantVoiceError, resolve_variant_voice_ids
+
+        try:
+            resolve_variant_voice_ids(
+                count=self.preview_count,
+                voice_library_id=self.voice_library_id,
+                voice_library_ids=self.voice_library_ids or None,
+            )
+        except VariantVoiceError as exc:
+            raise ValueError(str(exc)) from exc
+        return self
 
     @model_validator(mode="after")
     def _check_template_id(self) -> "CreatePreviewGenerationTaskRequest":
@@ -196,13 +282,13 @@ class CreatePreviewGenerationTaskRequest(BaseModel):
 
     @model_validator(mode="after")
     def _check_asset_ids(self) -> "CreatePreviewGenerationTaskRequest":
-        if not self.asset_ids and not self.title_ids and not self.voice_ids:
-            raise ValueError("asset_ids/title_ids/voice_ids 至少需要提供一个")
+        if not self.asset_ids and not self.title_ids:
+            raise ValueError("asset_ids/title_ids 至少需要提供一个")
         return self
 
 
 class PreviewGenerationTaskResponse(BaseModel):
-    """预览生成任务响应。
+    """单个预览变体任务响应。
 
     包含任务状态、进度、分辨率、生成结果 URL 等关键字段。
     """
@@ -211,6 +297,7 @@ class PreviewGenerationTaskResponse(BaseModel):
     status: str
     progress: float
     is_preview: bool = True
+    variant_index: int = 0
     resolution: str = ""
     video_url: str = ""
     duration: float = 0.0
@@ -219,7 +306,21 @@ class PreviewGenerationTaskResponse(BaseModel):
     transition_count: int = 0
     material_usage: dict = Field(default_factory=dict)
     error_message: str = ""
+    title_text: str = ""
+    voice_library_id: str = ""
     created_at: datetime | None = None
     started_at: datetime | None = None
     finished_at: datetime | None = None
     generate_duration: float = 0.0
+
+
+class BatchPreviewGenerationTaskResponse(BaseModel):
+    """批量预览任务响应：preview_count=N 时返回 N 个独立变体任务。
+
+    - items: 变体任务数组，按 variant_index 顺序排列，每个含独立 task_id/状态/预览视频URL
+    - total: 变体总数（= preview_count）
+    - 前端按 items[i].task_id 分别轮询 GET /preview/{task_id} 获取进度与结果
+    """
+
+    items: list[PreviewGenerationTaskResponse]
+    total: int

apps/api/app/schemas/lipsync.py

@@ -0,0 +1,100 @@
+"""对口型 API Schema 定义 — #1796 / #1809 / #1822.
+
+支持两种输入模式（二选一）：
+1. TTS 直生模式（推荐）：传 voice_id + script_text（+ speed/emotion），
+   后端内部先调 CosyVoice 合成音频，再提交 MediaKit 对口型。
+2. 直接音频模式：传 video_url + audio_url（音频已由调用方准备好）。
+"""
+
+from __future__ import annotations
+
+from datetime import datetime
+from typing import Optional
+
+from pydantic import BaseModel, Field, model_validator
+
+
+class LipsyncJobResponse(BaseModel):
+    """对口型任务响应."""
+
+    id: str
+    user_id: str
+    project_id: str
+    video_url: str
+    audio_url: str
+    enable_video_loop: bool
+    voice_id: str = ""
+    script_text: str = ""
+    speed: float = 1.0
+    emotion: str = ""
+    mediakit_task_id: str
+    status: str
+    output_video_url: str
+    output_duration: float
+    error_message: str
+    error_code: str
+    submitted_at: Optional[datetime] = None
+    completed_at: Optional[datetime] = None
+    created_at: datetime
+    updated_at: datetime
+
+    class Config:
+        from_attributes = True
+
+
+class CreateLipsyncJobRequest(BaseModel):
+    """创建对口型任务请求.
+
+    两种模式（二选一）：
+    - TTS 直生：voice_id + script_text 必填（+ 可选 speed/emotion）；audio_url 留空。
+    - 直接音频：video_url + audio_url 必填。
+    """
+
+    video_url: str = Field(..., description="人物视频 URL（MP4，≤30min，单人真人）")
+
+    # 模式 2：直接音频
+    audio_url: str = Field("", description="驱动音频 URL（mp3/aac/wav/m4a/flac）；直生模式留空")
+
+    # 模式 1：TTS 直生
+    voice_id: str = Field("", description="音色 ID（预置音色或克隆音色 profile UUID）")
+    script_text: str = Field("", description="要合成的文案（直生模式必填，最长 5000 字符）")
+    speed: float = Field(1.0, ge=0.5, le=2.0, description="语速（0.5-2.0），默认 1.0")
+    emotion: str = Field("", description="情绪（natural/excited/calm/friendly 或中文 自然/兴奋/沉稳/亲切）")
+
+    enable_video_loop: bool = Field(False, description="音频长于视频时是否循环画面")
+    project_id: str = Field("", description="项目 ID（可选）")
+
+    @model_validator(mode="after")
+    def _validate_input_mode(self) -> "CreateLipsyncJobRequest":
+        video = (self.video_url or "").strip()
+        if not video:
+            raise ValueError("video_url 不能为空")
+        if not video.startswith(("http://", "https://")):
+            raise ValueError("video_url 必须是 HTTP/HTTPS URL")
+        lower = video.lower().split("?")[0]
+        if not lower.endswith(".mp4"):
+            raise ValueError("video_url 仅支持 MP4 格式")
+
+        has_audio = bool((self.audio_url or "").strip())
+        has_tts = bool((self.voice_id or "").strip()) and bool((self.script_text or "").strip())
+
+        if not has_audio and not has_tts:
+            raise ValueError(
+                "必须提供驱动音频：要么传 audio_url（直接音频模式），"
+                "要么同时传 voice_id + script_text（TTS 直生模式）"
+            )
+
+        if has_tts and len(self.script_text) > 5000:
+            raise ValueError("script_text 最长 5000 字符")
+
+        if has_audio:
+            au = self.audio_url.strip()
+            if not au.startswith(("http://", "https://")):
+                raise ValueError("audio_url 必须是 HTTP/HTTPS URL")
+            au_lower = au.lower().split("?")[0]
+            allowed = (".mp3", ".aac", ".wav", ".m4a", ".flac")
+            if not any(au_lower.endswith(ext) for ext in allowed):
+                raise ValueError(f"audio_url 格式不支持，仅支持: {', '.join(allowed)}")
+            self.audio_url = au
+
+        return self

apps/api/app/schemas/script.py

@@ -0,0 +1,45 @@
+"""Script (口播文案库) Pydantic schemas — Issue #1795."""
+
+from __future__ import annotations
+
+from datetime import datetime
+from typing import List, Optional
+
+from pydantic import BaseModel, Field
+
+
+class ScriptSegment(BaseModel):
+    """单段文案."""
+
+    text: str
+    duration: Optional[float] = None
+
+
+class ScriptResponse(BaseModel):
+    id: str
+    user_id: str
+    title: str
+    content: str
+    segments: List[ScriptSegment] = Field(default_factory=list)
+    tags: List[str] = Field(default_factory=list)
+    created_at: datetime
+    updated_at: datetime
+
+
+class ScriptListResponse(BaseModel):
+    items: list[ScriptResponse]
+    total: int = 0
+
+
+class CreateScriptRequest(BaseModel):
+    title: str = Field(..., min_length=1, max_length=255)
+    content: str = ""
+    segments: List[ScriptSegment] = Field(default_factory=list)
+    tags: List[str] = Field(default_factory=list)
+
+
+class UpdateScriptRequest(BaseModel):
+    title: Optional[str] = Field(None, min_length=1, max_length=255)
+    content: Optional[str] = None
+    segments: Optional[List[ScriptSegment]] = None
+    tags: Optional[List[str]] = None

apps/api/app/schemas/tts.py

@@ -16,6 +16,7 @@ class TTSSynthesizeRequest(BaseModel):
     output_name: str = Field("", description="输出文件名")
     language: str = Field("zh-CN", description="语言")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
+    emotion: str = Field("", description="情绪（natural/excited/calm/friendly，或中文 自然/兴奋/沉稳/亲切）")
     voice_model: str = Field("", description="语音模型名称")
     voice_clone_profile_id: str = Field("", description="关联的音色克隆档案 ID")
     format: str = Field("mp3", description="输出格式（mp3/wav/pcm）")
@@ -109,6 +110,7 @@ class TTSPreviewRequest(BaseModel):
     text: str = Field(..., min_length=1, max_length=200, description="合成文本，限制 200 字")
     voice_id: str = Field(..., min_length=1, description="音色 ID")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
+    emotion: str = Field("", description="情绪（natural/excited/calm/friendly，或中文）")
     pitch: float = Field(1.0, ge=0.5, le=2.0, description="音调（预留，当前未使用）")
 
 

apps/api/app/schemas/upload.py

@@ -16,6 +16,7 @@ class DirectUploadPrepareRequest(BaseModel):
     content_type: str = Field(default="application/octet-stream", min_length=1, max_length=100)
     file_size: int = Field(..., gt=0)
     file_hash: str = Field(default="", max_length=64, description="文件 MD5 哈希，用于去重检测")
+    client_upload_id: str = Field(default="", max_length=64, description="客户端幂等 token（同一次上传的重试保持一致）")
 
 
 class DirectUploadPrepareResponse(BaseModel):
@@ -25,20 +26,25 @@ class DirectUploadPrepareResponse(BaseModel):
     expires_at: str
     fields: dict[str, str]
     max_size_bytes: int
+    duplicated: bool = False
+    skip_transfer: bool = False
+    asset_id: str = ""
 
 
 class DirectUploadCompleteRequest(BaseModel):
     project_id: str = Field(..., min_length=1)
     library_id: str = Field(..., min_length=1)
     storage_key: str = Field(..., min_length=1, max_length=255)
-    file_hash: str = Field(default="", max_length=64, description="文件 MD5 哈希，用于去重检测")
+    file_hash: str = Field(default="", max_length=64, description="文件哈希，用于去重检测")
+    client_upload_id: str = Field(default="", max_length=64, description="客户端幂等 token（同一次上传的重试保持一致）")
+    file_size: int = Field(default=0, ge=0, description="文件大小（字节），用于无 hash 时的兜底去重")
 
 
 class DirectUploadCompleteResponse(BaseModel):
     storage_key: str
     ingest_job_id: str
-    duplicated: bool = Field(default=False, description="是否为重复素材（命中去重）")
-    asset_id: str = Field(default="", description="重复素材的 asset_id（duplicated=true 时返回）")
+    duplicated: bool = Field(default=False, description="是否为重复素材/重复 complete（命中幂等去重）")
+    asset_id: str = Field(default="", description="素材 asset_id（重复 complete 时返回已存在记录）")
     url: str = Field(default="", description="Public URL of uploaded file")
 
 
@@ -46,5 +52,5 @@ class UploadAssetResponse(BaseModel):
     storage_key: str
     ingest_job_id: str
     url: str = Field(..., description="Public URL of uploaded file")
-    duplicated: bool = Field(default=False, description="是否为重复素材（命中去重）")
-    asset_id: str = Field(default="", description="重复素材的 asset_id（duplicated=true 时返回）")
+    duplicated: bool = Field(default=False, description="是否为重复素材/重复提交（命中幂等去重）")
+    asset_id: str = Field(default="", description="素材 asset_id（重复提交时返回已存在记录）")

apps/api/app/services/ai_avatar_cover_service.py

@@ -0,0 +1,162 @@
+"""AI 数字人封面服务 — 复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧.
+
+与 generation_cover.py 的智能选帧能力对齐（不再用 FFmpeg 简单截帧）：
+1. MediaKit extract_frames 抽取多帧（默认 5 帧，SpecifiedFrames 策略）
+2. cover_frame_scorer.score_frames 按清晰度/亮度/色彩评分选最佳
+3. 下载最佳帧并转存 OSS，返回公网封面 URL
+
+降级：MediaKit 不可用或抽帧失败时返回空字符串，由调用方决定回退策略。
+"""
+
+from __future__ import annotations
+
+import logging
+import tempfile
+import uuid
+from pathlib import Path
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+
+def select_best_cover_frame(video_url: str, *, max_frames: int = 5) -> str:
+    """从视频抽取多帧并评分选最佳帧，返回最佳帧的临时 URL.
+
+    Args:
+        video_url: 可公网访问的视频 URL
+        max_frames: 抽帧数量
+
+    Returns:
+        最佳帧图片 URL；失败返回空字符串
+    """
+    if not video_url:
+        return ""
+    try:
+        from packages.shared.cover_frame_scorer import score_frames
+        from packages.shared.mediakit_client import get_mediakit_client
+
+        mk = get_mediakit_client()
+        if not mk.is_available:
+            logger.warning("[数字人封面] MediaKit 未配置，无法智能抽帧")
+            return ""
+
+        snapshots = mk.extract_frames(
+            video_url=video_url,
+            strategy="SpecifiedFrames",
+            max_frames=max_frames,
+            poll_interval=2.0,
+            max_poll_attempts=5,
+            max_retries=0,
+        )
+        if not snapshots:
+            logger.warning("[数字人封面] MediaKit 未返回帧: %s", video_url[:80])
+            return ""
+
+        if len(snapshots) == 1:
+            return snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+
+        # 下载各帧评分
+        import httpx
+
+        candidates = []
+        for snap in snapshots:
+            url = snap.get("image_url") or snap.get("url") or ""
+            if not url:
+                continue
+            tmp_path: Optional[str] = None
+            try:
+                resp = httpx.get(url, timeout=15, follow_redirects=True)
+                resp.raise_for_status()
+                with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp:
+                    tmp.write(resp.content)
+                    tmp_path = tmp.name
+                candidates.append({"image_path": tmp_path, "url": url})
+            except Exception:
+                candidates.append({"image_path": None, "url": url, "score": 0.0})
+
+        if not candidates:
+            return snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+
+        scored = score_frames(candidates)
+        best = scored[0] if scored else None
+        best_url = best.get("url", "") if best else ""
+
+        # 清理临时文件
+        for c in candidates:
+            p = c.get("image_path")
+            if p:
+                try:
+                    Path(p).unlink(missing_ok=True)
+                except Exception:
+                    pass
+
+        logger.info(
+            "[数字人封面] 智能选帧完成: candidates=%d best_score=%s",
+            len(candidates),
+            best.get("score") if best else "n/a",
+        )
+        return best_url
+
+    except Exception:
+        logger.warning("[数字人封面] 智能选帧失败", exc_info=True)
+        return ""
+
+
+def persist_cover_to_oss(frame_url: str, *, job_id: str = "", prefix: str = "ai-avatar/covers") -> str:
+    """下载帧图并转存到 OSS，返回公网封面 URL.
+
+    Args:
+        frame_url: MediaKit 返回的临时帧图 URL
+        job_id: 关联任务 ID（用于 OSS key 命名）
+        prefix: OSS key 前缀
+
+    Returns:
+        OSS 公网 URL；失败回退原始 frame_url
+    """
+    if not frame_url:
+        return ""
+    tmp_path: Optional[str] = None
+    try:
+        import httpx
+
+        resp = httpx.get(frame_url, timeout=30, follow_redirects=True)
+        resp.raise_for_status()
+        if not resp.content:
+            return frame_url
+
+        with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp:
+            tmp.write(resp.content)
+            tmp_path = tmp.name
+
+        from packages.shared.storage import get_shared_storage_service
+
+        storage = get_shared_storage_service()
+        token = job_id or uuid.uuid4().hex[:12]
+        cover_key = f"{prefix}/{token}/cover_{uuid.uuid4().hex[:8]}.jpg"
+        public_url = storage.upload_file(
+            file_or_path=tmp_path,
+            storage_key=cover_key,
+            content_type="image/jpeg",
+        )
+        logger.info("[数字人封面] 封面已转存 OSS: key=%s", cover_key)
+        return public_url or frame_url
+    except Exception:
+        logger.warning("[数字人封面] 封面转存 OSS 失败，返回原始 URL", exc_info=True)
+        return frame_url
+    finally:
+        if tmp_path:
+            try:
+                Path(tmp_path).unlink(missing_ok=True)
+            except Exception:
+                pass
+
+
+def generate_smart_cover(video_url: str, *, job_id: str = "", max_frames: int = 5) -> str:
+    """一站式：MediaKit 智能抽帧选最佳 → 转存 OSS，返回封面公网 URL.
+
+    供独立封面接口与渲染管线复用。失败返回空字符串。
+    """
+    best_frame = select_best_cover_frame(video_url, max_frames=max_frames)
+    if not best_frame:
+        return ""
+    return persist_cover_to_oss(best_frame, job_id=job_id)

apps/api/app/services/ai_avatar_render_service.py

@@ -0,0 +1,394 @@
+"""AI数字人渲染合成 Service — #1798.
+
+职责：
+- 创建/查询/取消渲染任务
+- 调用 Celery 异步任务执行渲染
+- B-roll 合成 + 标题叠加 + 封面提取
+- 用户隔离
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+import tempfile
+import uuid
+from datetime import datetime, timezone
+from typing import Any, Optional
+
+from sqlalchemy.orm import Session
+
+from packages.adapters.sqlalchemy_impl.models import (
+    AiAvatarRenderJob,
+    LipsyncJobModel,
+    ScriptModel,
+)
+from packages.domain.video_filter_builder import (
+    build_cover_extract_command,
+    build_title_drawtext_filter,
+)
+from packages.shared.storage import get_shared_storage_service
+
+logger = logging.getLogger(__name__)
+
+
+class AiAvatarRenderError(Exception):
+    """渲染服务异常."""
+
+    def __init__(self, message: str, code: str = "RenderError"):
+        self.code = code
+        super().__init__(message)
+
+
+class AiAvatarRenderService:
+    """AI数字人渲染合成 Service."""
+
+    def __init__(self, db: Session):
+        self.db = db
+
+    # ── 创建任务 ──────────────────────────────────────────────────────────
+
+    def create_render_job(
+        self,
+        *,
+        user_id: str,
+        lipsync_job_id: str,
+        script_id: str = "",
+        b_roll_segments: list[dict[str, Any]] | None = None,
+        title_config: dict[str, Any],
+        cover_config: dict[str, Any],
+        project_id: str = "",
+    ) -> AiAvatarRenderJob:
+        """创建渲染任务.
+
+        Raises:
+            AiAvatarRenderError: 校验失败
+        """
+        # 1. 验证对口型任务
+        lipsync_job = (
+            self.db.query(LipsyncJobModel)
+            .filter(
+                LipsyncJobModel.id == lipsync_job_id,
+                LipsyncJobModel.user_id == user_id,
+            )
+            .first()
+        )
+        if lipsync_job is None:
+            raise AiAvatarRenderError("对口型任务不存在", code="LipsyncJobNotFound")
+        if lipsync_job.status != "completed":
+            raise AiAvatarRenderError(
+                f"对口型任务状态为 {lipsync_job.status}，仅 completed 状态可渲染",
+                code="LipsyncJobNotCompleted",
+            )
+        if not lipsync_job.output_video_url:
+            raise AiAvatarRenderError("对口型任务输出视频 URL 为空", code="LipsyncJobNoOutput")
+
+        # 2. 验证文案归属（仅当选了文案库条目时；手动输入文案直生场景 script_id 可空）
+        script_id = (script_id or "").strip()
+        if script_id:
+            script = (
+                self.db.query(ScriptModel)
+                .filter(
+                    ScriptModel.id == script_id,
+                    ScriptModel.user_id == user_id,
+                )
+                .first()
+            )
+            if script is None:
+                raise AiAvatarRenderError("文案不存在或无权访问", code="ScriptNotFound")
+
+        # 3. 创建渲染任务
+        job_id = str(uuid.uuid4())
+        job = AiAvatarRenderJob(
+            id=job_id,
+            user_id=user_id,
+            project_id=project_id,
+            lipsync_job_id=lipsync_job_id,
+            script_id=script_id,
+            b_roll_segments=[s if isinstance(s, dict) else s.model_dump() for s in (b_roll_segments or [])],
+            title_config=title_config,
+            cover_config=cover_config,
+            status="pending",
+        )
+        self.db.add(job)
+        self.db.flush()
+
+        job.submitted_at = datetime.now(timezone.utc)
+        self.db.commit()
+        self.db.refresh(job)
+        return job
+
+    # ── 查询任务 ──────────────────────────────────────────────────────────
+
+    def get_render_job(self, job_id: str, user_id: str) -> Optional[AiAvatarRenderJob]:
+        """获取渲染任务详情（用户隔离）."""
+        return (
+            self.db.query(AiAvatarRenderJob)
+            .filter(
+                AiAvatarRenderJob.id == job_id,
+                AiAvatarRenderJob.user_id == user_id,
+            )
+            .first()
+        )
+
+    def list_render_jobs(
+        self,
+        *,
+        user_id: str,
+        project_id: str = "",
+        status: str = "",
+        offset: int = 0,
+        limit: int = 20,
+    ) -> tuple[list[AiAvatarRenderJob], int]:
+        """获取渲染任务列表（分页 + 用户隔离）."""
+        query = self.db.query(AiAvatarRenderJob).filter(AiAvatarRenderJob.user_id == user_id)
+        if project_id:
+            query = query.filter(AiAvatarRenderJob.project_id == project_id)
+        if status:
+            query = query.filter(AiAvatarRenderJob.status == status)
+
+        total = query.count()
+        items = query.order_by(AiAvatarRenderJob.created_at.desc()).offset(offset).limit(limit).all()
+        return items, total
+
+    # ── 取消任务 ──────────────────────────────────────────────────────────
+
+    def cancel_render_job(self, job_id: str, user_id: str) -> Optional[AiAvatarRenderJob]:
+        """取消渲染任务（仅 pending 状态可取消）."""
+        job = self.get_render_job(job_id, user_id)
+        if job is None:
+            return None
+        if job.status in ("pending", "submitted"):
+            job.status = "cancelled"
+            job.updated_at = datetime.now(timezone.utc)
+            self.db.commit()
+            self.db.refresh(job)
+        return job
+
+    # ── 重试任务 ──────────────────────────────────────────────────────────
+
+    def retry_render_job(self, job_id: str, user_id: str) -> Optional[AiAvatarRenderJob]:
+        """重试失败的渲染任务."""
+        job = self.get_render_job(job_id, user_id)
+        if job is None:
+            return None
+        if job.status != "failed":
+            return None
+        job.status = "pending"
+        job.progress = 0
+        job.error_message = ""
+        job.output_video_url = ""
+        job.output_cover_url = ""
+        job.output_duration = 0.0
+        job.started_at = None
+        job.completed_at = None
+        job.updated_at = datetime.now(timezone.utc)
+        self.db.commit()
+        self.db.refresh(job)
+        return job
+
+    # ── 执行渲染（Celery 异步调用） ──────────────────────────────────────
+
+    def execute_render(self, job_id: str) -> None:
+        """执行渲染管线.
+
+        由 Celery 异步任务调用，流程:
+        1. 下载对口型输出视频 (20%)
+        2. 构建 FFmpeg 滤镜链 (40%)
+        3. 执行 FFmpeg 渲染 (80%)
+        4. 提取封面 (90%)
+        5. 上传到 OSS (95%)
+        6. 更新任务状态 (100%)
+        """
+        job = self.db.query(AiAvatarRenderJob).filter(AiAvatarRenderJob.id == job_id).first()
+        if job is None:
+            logger.error("渲染任务不存在: %s", job_id)
+            return
+
+        if job.status == "cancelled":
+            logger.info("渲染任务已取消: %s", job_id)
+            return
+
+        try:
+            # 更新状态为 processing
+            job.status = "processing"
+            job.started_at = datetime.now(timezone.utc)
+            job.progress = 5
+            job.updated_at = datetime.now(timezone.utc)
+            self.db.commit()
+
+            # 获取对口型任务信息
+            lipsync_job = self.db.query(LipsyncJobModel).filter(LipsyncJobModel.id == job.lipsync_job_id).first()
+            if lipsync_job is None:
+                raise AiAvatarRenderError("关联的对口型任务不存在", code="LipsyncJobNotFound")
+
+            # 1. 下载对口型输出视频 (20%)
+            input_video_path = self._download_video(lipsync_job.output_video_url)
+            job.progress = 20
+            self.db.commit()
+
+            # 2. 构建 FFmpeg 滤镜链 (40%)
+            from packages.domain.video_filter_builder import build_broll_overlay_filter
+
+            filter_complex = build_broll_overlay_filter(
+                b_roll_segments=job.b_roll_segments,
+                video_duration=lipsync_job.output_duration,
+            )
+
+            # 标题叠加
+            title_filter = build_title_drawtext_filter(job.title_config)
+            if title_filter:
+                if filter_complex:
+                    filter_complex += f"[vout]{title_filter}[vout_titled];"
+                else:
+                    filter_complex = f"[0:v]{title_filter}[vout_titled];"
+
+            # 清理末尾分号
+            if filter_complex.endswith(";"):
+                filter_complex = filter_complex[:-1]
+
+            # 最终输出标签
+            final_label = "vout_titled" if title_filter else ("vout" if filter_complex else None)
+
+            job.progress = 40
+            self.db.commit()
+
+            # 3. 执行 FFmpeg 渲染 (80%)
+            with tempfile.TemporaryDirectory() as tmpdir:
+                output_video_path = os.path.join(tmpdir, "output.mp4")
+
+                cmd = self._build_ffmpeg_command(
+                    input_video=input_video_path,
+                    b_roll_segments=job.b_roll_segments,
+                    filter_complex=filter_complex,
+                    final_label=final_label,
+                    output_path=output_video_path,
+                )
+
+                exit_code = os.system(cmd)
+                if exit_code != 0:
+                    raise AiAvatarRenderError(f"FFmpeg 渲染失败，退出码: {exit_code}", code="FFmpegFailed")
+
+                job.progress = 80
+                self.db.commit()
+
+                # 4. 提取封面 (90%)
+                cover_path = ""
+                if job.cover_config:
+                    cover_path = os.path.join(tmpdir, "cover.jpg")
+                    cover_cmd = build_cover_extract_command(job.cover_config, cover_path)
+                    cover_cmd = cover_cmd.replace("INPUT_VIDEO", output_video_path)
+                    cover_exit = os.system(cover_cmd)
+                    if cover_exit != 0:
+                        logger.warning("封面提取失败，跳过: %s", cover_cmd)
+                        cover_path = ""
+
+                job.progress = 90
+                self.db.commit()
+
+                # 5. 上传到 OSS (95%)
+                output_video_url = self._upload_to_oss(output_video_path, f"ai-avatar/{job_id}/output.mp4")
+                job.output_video_url = output_video_url
+
+                # 封面：优先复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧；
+                # MediaKit 不可用时回退到 FFmpeg 已按 cover_config 抽取的 cover_path
+                smart_cover_url = ""
+                if output_video_url:
+                    try:
+                        from app.services.ai_avatar_cover_service import (
+                            generate_smart_cover,
+                        )
+
+                        smart_cover_url = generate_smart_cover(output_video_url, job_id=job_id, max_frames=5)
+                    except Exception:
+                        logger.warning("智能封面(MediaKit)失败，回退 FFmpeg 封面 job_id=%s", job_id, exc_info=True)
+
+                if smart_cover_url:
+                    job.output_cover_url = smart_cover_url
+                elif cover_path:
+                    output_cover_url = self._upload_to_oss(cover_path, f"ai-avatar/{job_id}/cover.jpg")
+                    job.output_cover_url = output_cover_url
+
+                # 获取输出视频时长
+                job.output_duration = lipsync_job.output_duration
+                job.progress = 95
+                self.db.commit()
+
+            # 6. 完成
+            job.status = "completed"
+            job.progress = 100
+            job.completed_at = datetime.now(timezone.utc)
+            job.updated_at = datetime.now(timezone.utc)
+            self.db.commit()
+            logger.info("渲染任务完成: %s", job_id)
+
+        except AiAvatarRenderError as exc:
+            job.status = "failed"
+            job.error_message = str(exc)
+            job.updated_at = datetime.now(timezone.utc)
+            self.db.commit()
+            logger.error("渲染任务失败 [%s]: %s", job_id, exc)
+        except Exception as exc:
+            job.status = "failed"
+            job.error_message = f"渲染异常: {str(exc)}"
+            job.updated_at = datetime.now(timezone.utc)
+            self.db.commit()
+            logger.exception("渲染任务异常 [%s]", job_id)
+
+    def _download_video(self, url: str) -> str:
+        """下载视频到临时文件."""
+        import httpx
+
+        tmp = tempfile.NamedTemporaryFile(suffix=".mp4", delete=False)
+        try:
+            with httpx.Client(timeout=120) as client:
+                resp = client.get(url)
+                resp.raise_for_status()
+                tmp.write(resp.content)
+            return tmp.name
+        except Exception:
+            if os.path.exists(tmp.name):
+                os.unlink(tmp.name)
+            raise
+
+    def _build_ffmpeg_command(
+        self,
+        *,
+        input_video: str,
+        b_roll_segments: list[dict[str, Any]],
+        filter_complex: str,
+        final_label: Optional[str],
+        output_path: str,
+    ) -> str:
+        """构建 FFmpeg 命令."""
+        # 输入文件
+        inputs = f"-i {input_video}"
+        for seg in b_roll_segments:
+            asset_url = seg.get("asset_url", "")
+            if asset_url:
+                inputs += f" -i {asset_url}"
+
+        # 滤镜
+        if filter_complex and final_label:
+            filter_arg = f'-filter_complex "{filter_complex}" -map "[{final_label}]"'
+        elif filter_complex:
+            filter_arg = f'-filter_complex "{filter_complex}"'
+        else:
+            filter_arg = ""
+
+        return f"ffmpeg {inputs} {filter_arg} -c:v libx264 -preset fast -crf 23 -y {output_path}"
+
+    def _upload_to_oss(self, local_path: str, oss_key: str) -> str:
+        """上传文件到 OSS，返回 URL.
+
+        使用 SharedStorageService 统一存储服务。
+        """
+        storage = get_shared_storage_service()
+        url = storage.upload_file_smart(local_path, oss_key)
+        if url is None:
+            raise AiAvatarRenderError(
+                f"上传文件到 OSS 失败: {oss_key}",
+                code="OSSUploadFailed",
+            )
+        logger.info("上传文件到 OSS 成功: %s -> %s", local_path, url)
+        return url

apps/api/app/services/edit_plan_service.py

@@ -409,8 +409,13 @@ class EditPlanService:
                     clip_type=clip_item.get("clip_type", "main"),
                     order=order,
                     asset_id=clip_item.get("asset_id", ""),
+                    text_content=clip_item.get("text_content", ""),
                     start_time=clip_item.get("start_time", 0.0),
                     duration=clip_item.get("duration", 0.0),
+                    transition_effect=clip_item.get("transition_effect", "cut"),
+                    transition_duration=clip_item.get("transition_duration", 0.0),
+                    playback_speed=clip_item.get("playback_speed", 1.0),
+                    config=clip_item.get("config") or None,
                 )
                 model = EditPlanClipModel(
                     id=clip.id,
@@ -459,6 +464,159 @@ class EditPlanService:
             logger.exception("事务性替换片段失败: plan_id=%s", plan_id)
             raise
 
+    def reselect_plan_for_variant(
+        self,
+        source_plan_id: str,
+        candidate_asset_ids: list[str],
+        *,
+        created_by_user_id: str = "",
+        name_suffix: str = "变体",
+        voice_duration: float = 0.0,
+        rng=None,
+    ) -> EditPlan:
+        """为批量变体生成独立 plan：完整重跑单视频选片流程（#1743）。
+
+        与 clone_plan_for_variant（只重算起点、素材/顺序不变）不同，本方法：
+        - 源 plan 片段骨架（clip_type/order/duration/文案/转场）保留；
+        - 素材池 shuffle 随机分配 + main 片段顺序洗牌；
+        - 起点走场景镜头洗牌/随机起点/历史区间避让（与单视频同一入口）；
+        - 批次内同素材区间重叠 >20% 自动重选起点；
+        - 新片段区间 record_used_segments 写回素材 metadata（跨变体/跨任务避让）。
+
+        Args:
+            source_plan_id: 源 plan（任务 0 / 预览源）。
+            candidate_asset_ids: 素材池（源 plan 素材 ∪ 批次素材）。
+            created_by_user_id: 新 plan 归属用户。
+            name_suffix: plan 名后缀。
+            rng: 可选随机数（测试注入种子）。
+
+        Raises:
+            ValueError: 源 plan 不存在/无片段、素材池为空或时长全未知。
+        """
+        from packages.adapters.sqlalchemy_impl.models import AssetModel
+        from packages.domain.plan_generator_utils import extract_scene_points_from_metadata
+        from packages.domain.variant_plan_selector import reselect_clips_for_variant
+
+        source = self.get_plan_or_raise(source_plan_id)
+
+        # 分页读取源 plan 全部片段
+        clips: List[EditPlanClip] = []
+        skip, page = 0, 500
+        while True:
+            batch = self._clip_repo.list_by_plan(source_plan_id, skip=skip, limit=page)
+            if not batch:
+                break
+            clips.extend(batch)
+            if len(batch) < page:
+                break
+            skip += page
+        if not clips:
+            raise ValueError(f"源 plan 无片段，无法生成变体: {source_plan_id}")
+
+        source_clips_data: list[dict[str, Any]] = [
+            {
+                "order": c.order if c.order is not None else i,
+                "asset_id": c.asset_id,
+                "start_time": float(c.start_time or 0.0),
+                "duration": float(c.duration or 0.0),
+                "clip_type": c.clip_type,
+                "playback_speed": float(c.playback_speed or 1.0),
+                "transition_effect": c.transition_effect,
+                "transition_duration": float(c.transition_duration or 0.0),
+                "text_content": c.text_content or "",
+                "config": c.config or {},
+            }
+            for i, c in enumerate(clips)
+        ]
+
+        db = self._clip_repo.session
+
+        # #1749：配音时长 → 每段目标段长（片段数=模板片段数定死；素材不足由渲染末帧冻结铺满）
+        target_durations: list[float] | None = None
+        try:
+            voice = float(voice_duration or 0.0)
+        except (TypeError, ValueError):
+            voice = 0.0
+        if voice > 0 and source_clips_data:
+            from packages.domain.voice_duration_planner import plan_clip_durations
+
+            _effects: list[str | None] = [c.get("transition_effect") for c in source_clips_data]
+            _tdurs: list[float] = [float(c.get("transition_duration") or 0.0) for c in source_clips_data]
+            target_durations = plan_clip_durations(
+                len(source_clips_data),
+                voice,
+                transition_effects=_effects,
+                transition_durations=_tdurs,
+            )
+            if target_durations:
+                for _c, _d in zip(source_clips_data, target_durations, strict=False):
+                    _c["duration"] = _d
+
+        # 素材池 = 源 plan 素材 ∪ 调用方传入素材（去重保序）
+        pool_ids: list[str] = []
+        seen = set()
+        for aid in [c.asset_id for c in clips if c.asset_id] + list(candidate_asset_ids or []):
+            if aid and aid not in seen:
+                seen.add(aid)
+                pool_ids.append(aid)
+
+        # 时长 + 场景点
+        durations: dict[str, float] = {}
+        scene_points: dict[str, list[float]] = {}
+        if pool_ids:
+            for m in db.query(AssetModel).filter(AssetModel.id.in_(pool_ids)).all():
+                durations[m.id] = float(getattr(m, "duration", 0.0) or 0.0)
+                pts = extract_scene_points_from_metadata(getattr(m, "metadata", None))
+                if pts:
+                    scene_points[m.id] = pts
+
+        historical = get_used_segments(db, pool_ids)
+
+        # 创建新 plan（复制模板归属与 config）
+        new_plan = self.create_plan(
+            template_id=source.template_id,
+            name=f"{source.name or '剪辑计划'} · {name_suffix}",
+            config=dict(source.config or {}),
+            total_duration=source.total_duration,
+            project_id=source.project_id or "",
+            created_by_user_id=created_by_user_id or (source.created_by_user_id or ""),
+        )
+
+        # 批次内区间：以源 plan（变体 0）片段为初始避让对象
+        batch_segments: dict[str, list[tuple[float, float]]] = {}
+        for c in clips:
+            if c.asset_id and float(c.duration or 0) > 0:
+                st = float(c.start_time or 0.0)
+                batch_segments.setdefault(c.asset_id, []).append((st, st + float(c.duration)))
+
+        clips_data = reselect_clips_for_variant(
+            source_clips_data,
+            pool_ids,
+            asset_durations=durations,
+            asset_scene_points=scene_points,
+            historical_used_segments=historical,
+            batch_segments=batch_segments,
+            target_durations=target_durations,
+            rng=rng,
+        )
+
+        # 片段区间写回素材 metadata（与落库同事务；replace_all_clips_transactional 内 commit）
+        for item in clips_data:
+            aid = item.get("asset_id", "")
+            if aid:
+                st = float(item.get("start_time", 0.0))
+                record_used_segments(db, aid, st, st + float(item.get("duration", 0.0)), new_plan.id)
+
+        self.replace_all_clips_transactional(new_plan.id, clips_data)
+        logger.info(
+            "变体独立选片完成: source=%s new=%s clips=%d assets=%d",
+            source_plan_id,
+            new_plan.id,
+            len(clips_data),
+            len(pool_ids),
+        )
+        return new_plan
+
     def clone_plan_for_variant(
         self,
         source_plan_id: str,
@@ -569,6 +727,268 @@ class EditPlanService:
         )
         return new_plan
 
+    # ── #1749 配音时长分配 / 素材时长查询 / 批量变体 plan 确保 ──────────────
+
+    def get_asset_durations(self, asset_ids: list[str]) -> dict[str, float]:
+        """批量查询素材时长（秒），O(N) 单查；缺失/异常返回 0.0。"""
+        from packages.adapters.sqlalchemy_impl.models import AssetModel
+
+        ids = [a for a in dict.fromkeys(asset_ids or []) if a]
+        if not ids:
+            return {}
+        db = self._clip_repo.session
+        out: dict[str, float] = {}
+        for m in db.query(AssetModel).filter(AssetModel.id.in_(ids)).all():
+            try:
+                out[m.id] = float(getattr(m, "duration", 0.0) or 0.0)
+            except (TypeError, ValueError):
+                out[m.id] = 0.0
+        return out
+
+    def apply_voice_duration_to_plan(self, plan_id: str, voice_duration: float) -> Optional[EditPlan]:
+        """把配音时长分配到 plan 的每段（#1749）。
+
+        - 片段数保持不变（= 模板片段数，定死）；
+        - 每段 duration 按 voice_duration_planner 分配（含转场重叠扣减）；
+        - 素材短于段长 → start_time 钳制为 0（末帧冻结由渲染侧 tpad/apad 铺满）；
+        - plan.total_duration 回写为成片净时长（≈ 配音时长）；
+        - 幂等：配音时长相同则分配结果不变，可重复调用。
+
+        无配音（<=0）或无片段时直接返回 None，不报错。
+        """
+        try:
+            voice = float(voice_duration or 0.0)
+        except (TypeError, ValueError):
+            return None
+        if voice <= 0:
+            return None
+
+        plan = self.get_plan(plan_id)
+        if plan is None:
+            return None
+
+        clips: List[EditPlanClip] = []
+        skip, page = 0, 500
+        while True:
+            batch = self._clip_repo.list_by_plan(plan_id, skip=skip, limit=page)
+            if not batch:
+                break
+            clips.extend(batch)
+            if len(batch) < page:
+                break
+            skip += page
+        if not clips:
+            return None
+
+        clips.sort(key=lambda c: (c.order if c.order is not None else 0))
+
+        from packages.domain.voice_duration_planner import plan_clip_durations, total_output_duration
+
+        # #1764：从 plan config 读取节奏模板
+        rhythm_template = None
+        if plan and hasattr(plan, "config") and plan.config:
+            rhythm_template = plan.config.get("rhythm_template")
+
+        # #1768：先获取素材时长，传入 plan_clip_durations 用于最大片段钳制
+        asset_ids = [c.asset_id for c in clips if c.asset_id]
+        durations = self.get_asset_durations(asset_ids)
+        asset_durations_for_plan = [durations.get(c.asset_id, 0.0) for c in clips]
+
+        target = plan_clip_durations(
+            len(clips),
+            voice,
+            transition_effects=[c.transition_effect for c in clips],
+            transition_durations=[float(c.transition_duration or 0.0) for c in clips],
+            rhythm_template=rhythm_template,
+            asset_durations=asset_durations_for_plan,
+        )
+        if not target:
+            return None
+
+        clips_data: list[dict] = []
+        for i, c in enumerate(clips):
+            dur = float(target[i])
+            total = durations.get(c.asset_id, 0.0)
+            start = float(c.start_time or 0.0)
+            if c.asset_id and total > 0:
+                # 素材短于段长：起点钳 0，段长超出部分渲染侧末帧冻结
+                max_start = max(0.0, total - min(dur, total))
+                start = min(start, max_start)
+            clips_data.append(
+                {
+                    "order": c.order if c.order is not None else i,
+                    "asset_id": c.asset_id or "",
+                    "start_time": round(start, 3),
+                    "duration": dur,
+                    "clip_type": c.clip_type,
+                    "playback_speed": float(c.playback_speed or 1.0),
+                    "transition_effect": c.transition_effect,
+                    "transition_duration": float(c.transition_duration or 0.0),
+                    "text_content": c.text_content or "",
+                    "config": c.config or {},
+                }
+            )
+
+        self.replace_all_clips_transactional(plan_id, clips_data)
+
+        net = total_output_duration(
+            target,
+            transition_effects=[c.transition_effect for c in clips],
+            transition_durations=[float(c.transition_duration or 0.0) for c in clips],
+        )
+        try:
+            plan.total_duration = net
+            db = self._clip_repo.session
+            db.commit()
+        except Exception:
+            db.rollback()
+            logger.exception("回写 plan.total_duration 失败（不阻断）: plan_id=%s", plan_id)
+
+        logger.info(
+            "配音时长分配完成: plan=%s clips=%d voice=%.2fs 成片净时长=%.2fs",
+            plan_id,
+            len(clips),
+            voice,
+            net,
+        )
+        return plan
+
+    def ensure_variant_plans(
+        self,
+        source_plan_id: str,
+        count: int,
+        candidate_asset_ids: list[str],
+        *,
+        created_by_user_id: str = "",
+        voice_durations: Optional[list[float]] = None,
+        rng=None,
+    ) -> list[str]:
+        """确保批量 N 个变体各自拥有独立 plan（#1749 批量正式生成/预览共用）。
+
+        - 变体 0：clone 源 plan（不污染源 plan，片段独立可改），并按配音分配段长；
+        - 变体 1..N-1：reselect_plan_for_variant 完整重跑选片（素材级去重）；
+        - voice_durations：每个变体的配音时长（独立配音各自时长；统一配音同值）；
+          缺省/为 0 时不分配（段长保持骨架/模板值）。
+
+        Returns:
+            plan_id 列表，长度 == count，index 即 variant_index。
+        """
+        import random as _random
+
+        rng = rng or _random.Random()
+        plan_ids: list[str] = []
+
+        # 变体 0：clone（片段结构同源 plan，起点重算），不污染源 plan
+        plan0 = self.clone_plan_for_variant(
+            source_plan_id,
+            created_by_user_id=created_by_user_id,
+            name_suffix="变体1",
+        )
+        v0_voice = 0.0
+        if voice_durations and len(voice_durations) > 0:
+            try:
+                v0_voice = float(voice_durations[0] or 0.0)
+            except (TypeError, ValueError):
+                v0_voice = 0.0
+        if v0_voice > 0:
+            try:
+                self.apply_voice_duration_to_plan(plan0.id, v0_voice)
+            except Exception:
+                logger.exception("变体0 配音分配失败（不阻断）: plan=%s", plan0.id)
+        plan_ids.append(plan0.id)
+
+        # 变体 1..N-1：独立选片
+        for i in range(1, count):
+            voice = 0.0
+            if voice_durations and i < len(voice_durations):
+                try:
+                    voice = float(voice_durations[i] or 0.0)
+                except (TypeError, ValueError):
+                    voice = 0.0
+            variant = self.reselect_plan_for_variant(
+                source_plan_id,
+                candidate_asset_ids,
+                created_by_user_id=created_by_user_id,
+                name_suffix=f"变体{i + 1}",
+                voice_duration=voice,
+                rng=rng,
+            )
+            plan_ids.append(variant.id)
+
+        # #1764：为每个变体生成独立节奏模板（让批量视频片段时长分布不同）
+        from packages.domain.voice_duration_planner import RHYTHM_TEMPLATES, adapt_template_length
+
+        clip_count = 0
+        if voice_durations and len(voice_durations) > 0:
+            # 从源 plan 获取片段数
+            source_plan = self.get_plan(source_plan_id)
+            if source_plan and hasattr(source_plan, "clips"):
+                clip_count = len(list(source_plan.clips)) if source_plan.clips else 0
+
+        rhythm_templates_for_variants = []
+        if clip_count > 0:
+            for idx in range(len(plan_ids)):
+                # 每个变体用不同的 seed 选择节奏模板
+                variant_seed = rng.randint(0, 999999)
+                template = adapt_template_length(RHYTHM_TEMPLATES[variant_seed % len(RHYTHM_TEMPLATES)], clip_count)
+                rhythm_templates_for_variants.append(template)
+                logger.info("变体 %d 节奏模板: plan=%s template=%s", idx, plan_ids[idx], template)
+
+        # #1767：BGM 池差异化分配（让批量变体使用不同 BGM / 段落 / 音量）
+        from packages.domain.bgm_pool import allocate_bgm_pool_for_variants
+
+        source_bgm_config = {}
+        source_plan = self.get_plan(source_plan_id)
+        if source_plan and source_plan.config:
+            source_bgm_config = source_plan.config.get("bgm", {}) or {}
+
+        variant_seeds_for_bgm = [rng.randint(0, 999999) for _ in plan_ids]
+        bgm_pool_assignments = allocate_bgm_pool_for_variants(source_bgm_config, variant_seeds_for_bgm)
+
+        # 为每个变体生成独立视觉扰动参数（让批量视频画面本身更不同）
+        from packages.domain.variant_plan_selector import generate_visual_perturbation
+
+        for idx, pid in enumerate(plan_ids):
+            try:
+                perturbation = generate_visual_perturbation(rng)
+                # 变体 0 不做 hflip（保持预览 plan 原始画面方向）
+                if idx == 0:
+                    perturbation["hflip"] = False
+                config_update = {"visual_perturbation": perturbation}
+                # #1764：写入节奏模板
+                if idx < len(rhythm_templates_for_variants):
+                    config_update["rhythm_template"] = rhythm_templates_for_variants[idx]
+                # #1765：写入像素级扰动滤镜
+                from packages.domain.variant_plan_selector import generate_pixel_perturbation
+
+                pixel_pert = generate_pixel_perturbation(rng)
+                config_update["pixel_perturbation"] = pixel_pert
+                # #1767：写入 BGM 池分配（覆盖 bgm 配置中的 preset_id / audio_offset / volume_adjust_db）
+                if idx < len(bgm_pool_assignments):
+                    existing_bgm = dict((source_plan.config or {}).get("bgm", {}) or {})
+                    existing_bgm.update(bgm_pool_assignments[idx])
+                    config_update["bgm"] = existing_bgm
+                self.update_plan_config(pid, config_update)
+                logger.info(
+                    "变体 %d 视觉扰动+像素扰动+BGM池: plan=%s vis=%s pix=%s bgm=%s",
+                    idx,
+                    pid,
+                    perturbation,
+                    pixel_pert,
+                    bgm_pool_assignments[idx] if idx < len(bgm_pool_assignments) else None,
+                )
+            except Exception:
+                logger.exception("变体 %d 视觉扰动生成失败（不阻断）: plan=%s", idx, pid)
+
+        # 标记所有变体 plan 的 clips 为 ready（已分配素材+起点，语义上就是 ready）
+        for pid in plan_ids:
+            try:
+                self.mark_clips_ready(pid)
+            except Exception:
+                logger.exception("标记 clips ready 失败（不阻断）: plan=%s", pid)
+
+        return plan_ids
+
     # ── 片段分割与合并 ──────────────────────────────────────────────────────
 
     def split_clip(self, clip_id: str, split_time: float) -> Dict[str, Any]:
@@ -792,7 +1212,7 @@ class EditPlanService:
         config_asset_ids_count = len((plan.config or {}).get("asset_ids", []))
         clips_with_asset_count = sum(1 for c in clips if c.asset_id)
         logger.info(
-            "can_generate 诊断: plan=%s status=%s total_clips=%d " "clips_with_asset=%d config_asset_ids_count=%d",
+            "can_generate 诊断: plan=%s status=%s total_clips=%d clips_with_asset=%d config_asset_ids_count=%d",
             plan_id,
             plan.status,
             len(clips),
@@ -804,7 +1224,7 @@ class EditPlanService:
             config_asset_ids = (plan.config or {}).get("asset_ids", [])
             if config_asset_ids:
                 logger.warning(
-                    "can_generate 最后防线触发: plan=%s clips=%d 均无素材，" "从 config.asset_ids(%d个) 自动分配",
+                    "can_generate 最后防线触发: plan=%s clips=%d 均无素材，从 config.asset_ids(%d个) 自动分配",
                     plan_id,
                     len(clips),
                     len(config_asset_ids),
@@ -836,7 +1256,7 @@ class EditPlanService:
                     return False, "没有可渲染的就绪片段，自动修复后仍未分配素材"
             else:
                 logger.warning(
-                    "can_generate 失败: plan=%s clips=%d 均无素材，" "且 config.asset_ids 为空，无法自动修复",
+                    "can_generate 失败: plan=%s clips=%d 均无素材，且 config.asset_ids 为空，无法自动修复",
                     plan_id,
                     len(clips),
                 )

apps/api/app/services/edit_template_service.py

@@ -34,6 +34,17 @@ from packages.domain.template_clip_converter import (
 logger = logging.getLogger(__name__)
 
 
+class TemplateNotFoundError(Exception):
+    """模板不存在、已删除或当前用户无权访问.
+
+    与"模板存在但无片段配置"区分：路由层应映射为 HTTP 404。
+    """
+
+    def __init__(self, template_id: str) -> None:
+        self.template_id = template_id
+        super().__init__(f"模板不存在: {template_id}")
+
+
 class EditTemplateService:
     """模板管理服务
 
@@ -217,7 +228,14 @@ class EditTemplateService:
         skip: int = 0,
         limit: int = 100,
     ) -> List[TemplateClipConfig]:
-        """列出模板的片段配置"""
+        """列出模板的片段配置
+
+        注意：本方法要求模板存在于新表 ``edit_templates``（全局模板库），
+        主要服务于新模板系统的写入/发布路径。用户自建模板存放在旧表
+        ``templates``，不在 ``edit_templates`` 中，读取其片段配置请改用
+        :meth:`list_clip_configs_for_editor`，后者直接读取片段配置主表
+        ``template_clip_configs``，不依赖新模板主表、也不靠异常降级。
+        """
         # 确保模板存在
         self.get_template_or_raise(template_id)
         return self._clip_config_repo.list_by_template(
@@ -227,6 +245,52 @@ class EditTemplateService:
             limit=limit,
         )
 
+    def list_clip_configs_for_editor(
+        self,
+        template_id: str,
+        user_id: str,
+        *,
+        clip_type: Optional[ClipType] = None,
+        skip: int = 0,
+        limit: int = 100,
+    ) -> List[TemplateClipConfig]:
+        """编辑器读取模板片段配置的单一数据源入口.
+
+        片段配置主表是 ``template_clip_configs``（直接读取，不抛异常、不降级）。
+        模板主表按双表现状显式判定，不使用 try/except 控制流：
+
+        1. 用户自建模板在旧表 ``templates``（归属 user_id）→ 校验归属与未删除后直接读；
+        2. 全局模板在新表 ``edit_templates``（无 user_id，全局可读）→ 直接读；
+        3. 两者都没有 → 模板不存在/无权限，抛 :class:`TemplateNotFoundError`。
+
+        Args:
+            template_id: 模板 ID
+            user_id: 当前登录用户 ID（用于旧表模板归属校验）
+
+        Raises:
+            TemplateNotFoundError: 模板不存在、已删除或不归属于当前用户。
+        """
+        # 1) 用户自建模板（旧表 templates，归属 user_id）
+        if self._clip_config_repo.template_owned_by(template_id, user_id):
+            return self._clip_config_repo.list_by_template(
+                template_id,
+                clip_type=clip_type,
+                skip=skip,
+                limit=limit,
+            )
+
+        # 2) 全局模板（新表 edit_templates，无 user_id，全局可读）
+        if self._template_repo.get(template_id) is not None:
+            return self._clip_config_repo.list_by_template(
+                template_id,
+                clip_type=clip_type,
+                skip=skip,
+                limit=limit,
+            )
+
+        # 3) 两表都没有：不存在 / 已删除 / 无权限
+        raise TemplateNotFoundError(template_id)
+
     def get_clip_config(self, config_id: str) -> Optional[TemplateClipConfig]:
         """获取片段配置详情"""
         return self._clip_config_repo.get(config_id)

apps/api/app/services/lipsync_service.py

@@ -0,0 +1,338 @@
+"""对口型 Service — #1796 MediaKit 对口型业务逻辑, #1809 参数调整.
+
+职责：
+- 创建/查询对口型任务
+- 双输入模式：TTS 直生（voice_id + script_text，内部先合成音频转存 OSS）或直接音频（audio_url）
+- 调用 MediaKit 客户端提交异步任务
+- 轮询更新任务状态（中间状态同步 DB，成片转存自家 OSS）
+- 用户隔离（每个用户只能操作自己的任务）
+"""
+
+from __future__ import annotations
+
+import io
+import logging
+import uuid
+from datetime import datetime, timezone
+from typing import Optional
+
+from app.services.mediakit_client import (
+    STATUS_COMPLETED,
+    STATUS_FAILED,
+    STATUS_RUNNING,
+    MediaKitClient,
+    MediaKitError,
+    get_mediakit_client,
+)
+from sqlalchemy.orm import Session
+
+from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
+from packages.application.cosyvoice_service import CosyVoiceError, normalize_emotion
+from packages.shared.storage import get_shared_storage_service
+from packages.shared.url_security import ALLOWED_AUDIO_MIME_TYPES, safe_download_bytes
+
+logger = logging.getLogger(__name__)
+
+
+class LipsyncService:
+    """对口型任务 Service."""
+
+    def __init__(
+        self,
+        db: Session,
+        client: Optional[MediaKitClient] = None,
+        cosyvoice_service=None,
+        voice_clone_repo=None,
+    ):
+        self.db = db
+        self.client = client or get_mediakit_client()
+        self._cosyvoice = cosyvoice_service
+        self._voice_clone_repo = voice_clone_repo
+
+    def _get_cosyvoice(self):
+        """延迟获取 CosyVoiceService（与 tts 路由一致，含 OSS 预签名配置）."""
+        if self._cosyvoice is None:
+            from app.dependencies import get_cosyvoice_service
+
+            self._cosyvoice = get_cosyvoice_service()
+        return self._cosyvoice
+
+    def _resolve_voice_id(self, voice_id: str, user_id: str) -> str:
+        """将克隆音色 profile UUID 解析为 CosyVoice voice_id。
+
+        与 /tts/synthesize 保持一致：命中 profile → 校验归属 → 返回其 voice_id；
+        未命中（预置音色 ID 或克隆 CosyVoice voice_id）原样返回。
+        """
+        if not voice_id:
+            return ""
+        if self._voice_clone_repo is None:
+            try:
+                from app.dependencies import get_voice_clone_profile_repository
+
+                self._voice_clone_repo = get_voice_clone_profile_repository(self.db)
+            except Exception:
+                return voice_id
+        try:
+            profile = self._voice_clone_repo.get(voice_id)
+        except Exception:
+            return voice_id
+        if profile is None:
+            return voice_id
+        if getattr(profile, "user_id", "") != user_id:
+            raise MediaKitError("无权访问该音色", code="VoiceForbidden")
+        if not getattr(profile, "voice_id", ""):
+            raise MediaKitError("音色克隆尚未完成，请稍后再试", code="VoiceNotReady")
+        return profile.voice_id
+
+    def _synthesize_and_persist_audio(
+        self,
+        *,
+        user_id: str,
+        job_id: str,
+        voice_id: str,
+        script_text: str,
+        speed: float,
+        emotion: str,
+    ) -> str:
+        """TTS 直生：调 CosyVoice 合成音频并转存 OSS，返回可公网访问的音频 URL.
+
+        Raises:
+            MediaKitError: 合成失败
+        """
+        actual_voice_id = self._resolve_voice_id(voice_id, user_id)
+        cosyvoice = self._get_cosyvoice()
+        try:
+            result = cosyvoice.submit_synthesize_task(
+                text=script_text,
+                voice_id=actual_voice_id,
+                speed=speed,
+                emotion=normalize_emotion(emotion),
+            )
+        except CosyVoiceError as exc:
+            raise MediaKitError(f"TTS 合成失败: {exc}", code="TTSSynthesisFailed") from exc
+        except ValueError as exc:
+            raise MediaKitError(f"TTS 参数错误: {exc}", code="TTSInvalidParam") from exc
+
+        temp_url = result.get("audio_url", "")
+        if not temp_url:
+            raise MediaKitError("TTS 未返回音频 URL", code="TTSNoAudio")
+
+        # 转存到自家 OSS，避免临时 URL 过期导致 MediaKit 拉取失败
+        try:
+            audio_data = safe_download_bytes(
+                temp_url,
+                purpose="lipsync_tts_audio",
+                allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
+                timeout=60.0,
+            )
+            storage = get_shared_storage_service()
+            storage_key = f"lipsync-tts/{user_id}/{job_id}.mp3"
+            permanent_url = storage.upload_file(io.BytesIO(audio_data), storage_key, content_type="audio/mpeg")
+            logger.info("对口型 TTS 音频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
+            return permanent_url
+        except Exception as exc:
+            logger.warning("TTS 音频转存 OSS 失败，回退临时 URL: job_id=%s err=%s", job_id, exc)
+            return temp_url
+
+    # ── 创建任务 ──────────────────────────────────────────────────────────
+
+    def create_job(
+        self,
+        *,
+        user_id: str,
+        video_url: str,
+        audio_url: str = "",
+        voice_id: str = "",
+        script_text: str = "",
+        speed: float = 1.0,
+        emotion: str = "",
+        enable_video_loop: bool = False,
+        project_id: str = "",
+    ) -> LipsyncJobModel:
+        """创建对口型任务并提交到 MediaKit.
+
+        两种输入模式：
+        - TTS 直生：voice_id + script_text（audio_url 留空），后端先合成音频
+        - 直接音频：提供 audio_url
+
+        Raises:
+            MediaKitError: TTS 合成或 MediaKit 提交失败
+        """
+        # 0. TTS 直生模式：先合成音频（在创建 DB 记录之前完成，失败直接抛出）
+        if not audio_url:
+            if not (voice_id and script_text):
+                raise MediaKitError(
+                    "必须提供 audio_url 或 voice_id+script_text",
+                    code="InvalidInput",
+                )
+            # 预合成：用临时 job_id 命名 OSS 对象
+            pre_job_id = str(uuid.uuid4())
+            audio_url = self._synthesize_and_persist_audio(
+                user_id=user_id,
+                job_id=pre_job_id,
+                voice_id=voice_id,
+                script_text=script_text,
+                speed=speed,
+                emotion=emotion,
+            )
+
+        # 1. 创建数据库记录
+        job_id = str(uuid.uuid4())
+        job = LipsyncJobModel(
+            id=job_id,
+            user_id=user_id,
+            project_id=project_id,
+            video_url=video_url,
+            audio_url=audio_url,
+            enable_video_loop=enable_video_loop,
+            voice_id=voice_id or "",
+            script_text=script_text or "",
+            speed=speed,
+            emotion=normalize_emotion(emotion),
+            status="pending",
+        )
+        self.db.add(job)
+        self.db.flush()
+
+        # 3. 提交到 MediaKit
+        try:
+            result = self.client.submit_lipsync(
+                video_url=video_url,
+                audio_url=audio_url,
+                enable_video_loop=enable_video_loop,
+                client_token=job_id,  # 幂等控制
+            )
+            job.mediakit_task_id = result["task_id"]
+            job.status = "submitted"
+            job.submitted_at = datetime.now(timezone.utc)
+        except MediaKitError as exc:
+            job.status = "failed"
+            job.error_message = str(exc)
+            job.error_code = exc.code
+            logger.error("提交对口型任务失败: %s", exc)
+            raise
+
+        self.db.commit()
+        self.db.refresh(job)
+        return job
+
+    # ── 查询任务 ──────────────────────────────────────────────────────────
+
+    def get_job(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
+        """获取任务详情（用户隔离）."""
+        return (
+            self.db.query(LipsyncJobModel)
+            .filter(LipsyncJobModel.id == job_id, LipsyncJobModel.user_id == user_id)
+            .first()
+        )
+
+    def list_jobs(
+        self,
+        *,
+        user_id: str,
+        project_id: str = "",
+        status: str = "",
+        offset: int = 0,
+        limit: int = 20,
+    ) -> tuple[list[LipsyncJobModel], int]:
+        """获取任务列表（分页 + 用户隔离）."""
+        query = self.db.query(LipsyncJobModel).filter(LipsyncJobModel.user_id == user_id)
+        if project_id:
+            query = query.filter(LipsyncJobModel.project_id == project_id)
+        if status:
+            query = query.filter(LipsyncJobModel.status == status)
+
+        total = query.count()
+        items = query.order_by(LipsyncJobModel.created_at.desc()).offset(offset).limit(limit).all()
+        return items, total
+
+    # ── 更新任务状态（轮询） ──────────────────────────────────────────────
+
+    def refresh_job_status(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
+        """从 MediaKit 拉取最新状态并更新本地记录.
+
+        Returns:
+            更新后的 Job，或 None（任务不存在/不属于该用户）
+        """
+        job = self.get_job(job_id, user_id)
+        if job is None:
+            return None
+
+        # 终态不需要再轮询
+        if job.status in (STATUS_COMPLETED, "failed"):
+            return job
+
+        # 未提交的任务不轮询
+        if not job.mediakit_task_id:
+            return job
+
+        try:
+            status_data = self.client.get_task_status(job.mediakit_task_id)
+        except MediaKitError as exc:
+            logger.error("轮询对口型任务状态失败 [%s]: %s", job_id, exc)
+            return job
+
+        mk_status = status_data.get("status", STATUS_RUNNING)
+        logger.info("MediaKit 对口型状态 [%s]: %s", job_id, mk_status)
+
+        if mk_status == STATUS_COMPLETED:
+            result = status_data.get("result", {})
+            job.status = STATUS_COMPLETED
+            output_url = result.get("video_url", "")
+            # MediaKit 输出为临时 URL，转存自家 OSS 防止过期（失败则回退临时 URL）
+            job.output_video_url = self._persist_output_video(output_url, job_id, user_id)
+            job.output_duration = result.get("duration", 0.0)
+            job.completed_at = datetime.now(timezone.utc)
+        elif mk_status == STATUS_FAILED:
+            error = status_data.get("error", {})
+            job.status = "failed"
+            job.error_message = error.get("message", "任务执行失败")
+            job.error_code = error.get("code", "TaskFailed")
+            job.completed_at = datetime.now(timezone.utc)
+        else:
+            # 中间状态（running/processing/queued 等）同步到 DB，避免前端永远卡在 submitted
+            if isinstance(mk_status, str) and mk_status:
+                job.status = mk_status
+        job.updated_at = datetime.now(timezone.utc)
+        self.db.commit()
+        self.db.refresh(job)
+        return job
+
+    def _persist_output_video(self, temp_url: str, job_id: str, user_id: str) -> str:
+        """将 MediaKit 输出的临时视频 URL 转存到自家 OSS.
+
+        失败时回退返回原始临时 URL，不影响任务完成。
+        """
+        if not temp_url:
+            return ""
+        try:
+            import httpx
+
+            with httpx.Client(timeout=180.0, follow_redirects=True) as client:
+                resp = client.get(temp_url)
+                resp.raise_for_status()
+                data = resp.content
+            storage = get_shared_storage_service()
+            storage_key = f"lipsync-outputs/{user_id}/{job_id}.mp4"
+            permanent_url = storage.upload_file(io.BytesIO(data), storage_key, content_type="video/mp4")
+            logger.info("对口型输出视频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
+            return permanent_url or temp_url
+        except Exception as exc:
+            logger.warning("对口型输出视频转存 OSS 失败，回退临时 URL: job_id=%s err=%s", job_id, exc)
+            return temp_url
+
+    # ── 取消任务 ──────────────────────────────────────────────────────────
+
+    def cancel_job(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
+        """取消任务（仅 pending/submitted 状态可取消）."""
+        job = self.get_job(job_id, user_id)
+        if job is None:
+            return None
+
+        if job.status in ("pending", "submitted"):
+            job.status = "cancelled"
+            job.updated_at = datetime.now(timezone.utc)
+            self.db.commit()
+            self.db.refresh(job)
+
+        return job

apps/api/app/services/mediakit_client.py

@@ -0,0 +1,243 @@
+"""MediaKit 客户端 — 封装火山引擎 AI MediaKit 对口型 API.
+
+接口文档：https://docs.volcengine.com/docs/6448/2656064
+
+异步任务流程：
+1. POST /api/v1/tools/lip-sync  提交对口型任务 → 返回 task_id
+2. GET  /api/v1/tasks/{task_id} 轮询任务状态 → running/completed/failed
+3. completed 时 result.video_url 为口型对齐视频（临时链接 24h 有效）
+
+设计原则：
+- API Key 从配置读取（settings.mediakit_api_key）
+- 未配置 API Key 时所有方法返回降级响应，不阻塞主流程
+- HTTP 超时/网络异常统一包装为 MediaKitError
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any, Optional
+
+import httpx
+
+from packages.config import get_api_settings
+
+logger = logging.getLogger(__name__)
+
+# ── 任务状态常量 ──────────────────────────────────────────────────────────
+STATUS_RUNNING = "running"
+STATUS_COMPLETED = "completed"
+STATUS_FAILED = "failed"
+
+
+class MediaKitError(Exception):
+    """MediaKit API 调用异常."""
+
+    def __init__(self, message: str, code: str = "", request_id: str = ""):
+        self.code = code
+        self.request_id = request_id
+        super().__init__(message)
+
+
+class MediaKitClient:
+    """火山引擎 AI MediaKit 对口型 API 客户端.
+
+    用法：
+        client = get_mediakit_client()
+        result = client.submit_lipsync(video_url="...", audio_url="...")
+        task_id = result["task_id"]
+
+        status = client.get_task_status(task_id)
+        # {"status": "completed", "result": {"video_url": "...", "duration": 60.5}}
+    """
+
+    def __init__(self) -> None:
+        settings = get_api_settings()
+        self._api_key = settings.mediakit_api_key
+        self._base_url = settings.mediakit_base_url.rstrip("/")
+        self._timeout = settings.mediakit_timeout
+
+    @property
+    def is_available(self) -> bool:
+        """是否已配置 API Key（未配置时自动降级）."""
+        return bool(self._api_key)
+
+    def _headers(self) -> dict[str, str]:
+        return {
+            "Authorization": f"Bearer {self._api_key}",
+            "Content-Type": "application/json",
+        }
+
+    # ── 提交对口型任务 ────────────────────────────────────────────────────
+
+    def submit_lipsync(
+        self,
+        *,
+        video_url: str,
+        audio_url: str,
+        enable_video_loop: bool = False,
+        callback_url: Optional[str] = None,
+        callback_args: Optional[str] = None,
+        client_token: Optional[str] = None,
+    ) -> dict[str, Any]:
+        """提交视频口型对齐任务.
+
+        Args:
+            video_url: 人物视频 URL（MP4，≤30min，单人真人）
+            audio_url: 驱动音频 URL（mp3/aac/wav/m4a/flac）
+            enable_video_loop: 音频长于视频时是否循环画面
+            callback_url: 任务完成回调 URL
+            callback_args: 回调时原样返回的自定义参数
+            client_token: 幂等控制 token
+
+        Returns:
+            {"success": True, "task_id": "...", "request_id": "..."}
+
+        Raises:
+            MediaKitError: API 调用失败
+        """
+        if not self.is_available:
+            raise MediaKitError("MediaKit API Key 未配置", code="NotConfigured")
+
+        payload: dict[str, Any] = {
+            "video_url": video_url,
+            "audio_url": audio_url,
+        }
+        if enable_video_loop:
+            payload["enable_video_loop"] = True
+        if callback_url:
+            payload["callback_url"] = callback_url
+        if callback_args:
+            payload["callback_args"] = callback_args[:512]  # API 限制 512 字节
+        if client_token:
+            payload["client_token"] = client_token[:64]  # API 限制 64 字符
+
+        try:
+            with httpx.Client(timeout=self._timeout) as client:
+                resp = client.post(
+                    f"{self._base_url}/tools/lip-sync",
+                    headers=self._headers(),
+                    json=payload,
+                )
+                resp.raise_for_status()
+                data = resp.json()
+        except httpx.TimeoutException as exc:
+            raise MediaKitError(f"MediaKit API 超时 ({self._timeout}s)", code="Timeout") from exc
+        except httpx.HTTPStatusError as exc:
+            body = exc.response.text[:500]
+            raise MediaKitError(
+                f"MediaKit API HTTP {exc.response.status_code}: {body}",
+                code="HttpError",
+            ) from exc
+        except httpx.RequestError as exc:
+            raise MediaKitError(f"MediaKit API 网络错误: {exc}", code="NetworkError") from exc
+        except Exception as exc:
+            raise MediaKitError(f"MediaKit API 未知错误: {exc}", code="UnknownError") from exc
+
+        if not data.get("success"):
+            error = data.get("error", {})
+            raise MediaKitError(
+                error.get("message", "提交任务失败"),
+                code=error.get("code", "SubmitFailed"),
+                request_id=data.get("request_id", ""),
+            )
+
+        return {
+            "success": True,
+            "task_id": data["task_id"],
+            "request_id": data.get("request_id", ""),
+        }
+
+    # ── 查询任务状态 ──────────────────────────────────────────────────────
+
+    def get_task_status(self, task_id: str) -> dict[str, Any]:
+        """查询异步任务状态和结果.
+
+        Args:
+            task_id: 提交任务时返回的任务 ID
+
+        Returns:
+            {
+                "success": True,
+                "task_id": "...",
+                "status": "running" | "completed" | "failed",
+                "result": {"video_url": "...", "duration": 60.5} | None,
+                "error": {"code": "...", "message": "..."} | None,
+                "created_at": 1777291767,
+                "finished_at": 1777291851 | None,
+                "expires_at": 1777464650 | None,
+            }
+
+        Raises:
+            MediaKitError: API 调用失败
+        """
+        if not self.is_available:
+            raise MediaKitError("MediaKit API Key 未配置", code="NotConfigured")
+
+        try:
+            with httpx.Client(timeout=self._timeout) as client:
+                resp = client.get(
+                    f"{self._base_url}/tasks/{task_id}",
+                    headers=self._headers(),
+                )
+                resp.raise_for_status()
+                data = resp.json()
+        except httpx.TimeoutException as exc:
+            raise MediaKitError(f"MediaKit API 超时 ({self._timeout}s)", code="Timeout") from exc
+        except httpx.HTTPStatusError as exc:
+            body = exc.response.text[:500]
+            raise MediaKitError(
+                f"MediaKit API HTTP {exc.response.status_code}: {body}",
+                code="HttpError",
+            ) from exc
+        except httpx.RequestError as exc:
+            raise MediaKitError(f"MediaKit API 网络错误: {exc}", code="NetworkError") from exc
+        except Exception as exc:
+            raise MediaKitError(f"MediaKit API 未知错误: {exc}", code="UnknownError") from exc
+
+        if not data.get("success"):
+            error = data.get("error", {})
+            raise MediaKitError(
+                error.get("message", "查询任务失败"),
+                code=error.get("code", "QueryFailed"),
+                request_id=data.get("request_id", ""),
+            )
+
+        result: dict[str, Any] = {
+            "success": True,
+            "task_id": data.get("task_id", task_id),
+            "status": data.get("status", STATUS_RUNNING),
+            "result": data.get("result"),
+            "created_at": data.get("created_at"),
+            "finished_at": data.get("finished_at"),
+            "expires_at": data.get("expires_at"),
+        }
+
+        # 失败时提取错误信息
+        if data.get("status") == STATUS_FAILED:
+            error_obj = data.get("error", {})
+            result["error"] = {
+                "code": error_obj.get("code", "TaskFailed"),
+                "message": error_obj.get("message", "任务执行失败"),
+            }
+
+        return result
+
+
+# ── 单例 ──────────────────────────────────────────────────────────────────
+
+_client: Optional[MediaKitClient] = None
+
+
+def get_mediakit_client() -> MediaKitClient:
+    """获取 MediaKit 客户端单例."""
+    global _client
+    if _client is None:
+        _client = MediaKitClient()
+    return _client
+
+
+def reset_mediakit_client() -> None:
+    """重置客户端（测试用）."""
+    global _client
+    _client = None

apps/api/app/services/script_service.py

@@ -0,0 +1,109 @@
+"""ScriptService — Issue #1795 口播文案库 CRUD.
+
+纯 Service 层封装，routes 直接调用。
+"""
+
+from __future__ import annotations
+
+import uuid
+from datetime import datetime, timezone
+from typing import Optional
+
+from sqlalchemy.orm import Session
+
+from packages.adapters.sqlalchemy_impl.models import ScriptModel
+
+
+class ScriptNotFoundError(Exception):
+    """文案不存在或不属于当前用户."""
+
+
+class ScriptService:
+    """口播文案 CRUD."""
+
+    def __init__(self, db: Session) -> None:
+        self.db = db
+
+    # ── list ──────────────────────────────────────────────────────────────
+
+    def list_scripts(
+        self,
+        user_id: str,
+        skip: int = 0,
+        limit: int = 50,
+        tag: Optional[str] = None,
+    ) -> tuple[list[ScriptModel], int]:
+        """返回 (items, total)."""
+        q = self.db.query(ScriptModel).filter(ScriptModel.user_id == user_id)
+        if tag:
+            # JSON 数组包含查询
+            q = q.filter(ScriptModel.tags.contains([tag]))
+        total = q.count()
+        items = q.order_by(ScriptModel.created_at.desc()).offset(skip).limit(limit).all()
+        return items, total
+
+    # ── create ────────────────────────────────────────────────────────────
+
+    def create_script(
+        self,
+        user_id: str,
+        title: str,
+        content: str = "",
+        segments: list | None = None,
+        tags: list | None = None,
+    ) -> ScriptModel:
+        script = ScriptModel(
+            id=str(uuid.uuid4()),
+            user_id=user_id,
+            title=title,
+            content=content,
+            segments=segments if segments is not None else [],
+            tags=tags if tags is not None else [],
+        )
+        self.db.add(script)
+        self.db.commit()
+        self.db.refresh(script)
+        return script
+
+    # ── get ───────────────────────────────────────────────────────────────
+
+    def get_script(self, script_id: str, user_id: str) -> ScriptModel:
+        script = self.db.query(ScriptModel).filter(ScriptModel.id == script_id, ScriptModel.user_id == user_id).first()
+        if script is None:
+            raise ScriptNotFoundError(f"Script {script_id} not found")
+        return script
+
+    # ── update ────────────────────────────────────────────────────────────
+
+    def update_script(
+        self,
+        script_id: str,
+        user_id: str,
+        title: Optional[str] = None,
+        content: Optional[str] = None,
+        segments: Optional[list] = None,
+        tags: Optional[list] = None,
+    ) -> ScriptModel:
+        script = self.get_script(script_id, user_id)
+        if title is not None:
+            script.title = title
+        if content is not None:
+            script.content = content
+        if segments is not None:
+            script.segments = segments
+        if tags is not None:
+            script.tags = tags
+        script.updated_at = datetime.now(timezone.utc)
+        self.db.commit()
+        self.db.refresh(script)
+        return script
+
+    # ── delete ────────────────────────────────────────────────────────────
+
+    def delete_script(self, script_id: str, user_id: str) -> bool:
+        script = self.db.query(ScriptModel).filter(ScriptModel.id == script_id, ScriptModel.user_id == user_id).first()
+        if script is None:
+            return False
+        self.db.delete(script)
+        self.db.commit()
+        return True

apps/api/app/services/video_compose_service.py

@@ -39,6 +39,9 @@ from packages.domain.video_filter_builder import (
 )
 from packages.domain.video_filter_builder import build_concat_filter as _build_concat_filter_func
 from packages.domain.video_filter_builder import build_filter_complex as _build_filter_complex
+from packages.domain.video_filter_builder import (
+    build_title_drawtext_filter,
+)
 from packages.domain.video_filter_builder import build_xfade_filter as _build_xfade_filter_func
 from packages.domain.video_filter_builder import chain_filters as _chain_filters_func
 from packages.domain.video_filter_builder import has_audio as _has_audio_func
@@ -248,6 +251,27 @@ class VideoComposeService:
             transitions=[c.transition_effect for c in ready_clips],
         )
 
+        # ── #1789 标题 drawtext 滤镜叠加 ──
+        # 从 plan.config 读取 title_config，生成 drawtext 滤镜链入 filter_complex
+        title_cfg = (plan.config or {}).get("title", {}) or {}
+        if not isinstance(title_cfg, dict):
+            title_cfg = {}
+        # 同时兼容 plan.config["title_config"]（API 回写路径）
+        if not title_cfg.get("text") and not title_cfg.get("content"):
+            title_cfg_alt = (plan.config or {}).get("title_config", {}) or {}
+            if isinstance(title_cfg_alt, dict) and (title_cfg_alt.get("text") or title_cfg_alt.get("content")):
+                title_cfg = title_cfg_alt
+        drawtext_filter = build_title_drawtext_filter(title_cfg, output_width, output_height)
+        if drawtext_filter:
+            # 将最终输出标签从 [outv] 改为 [composed]，再链入 drawtext → [outv]
+            filter_complex = filter_complex.replace("[outv]", "[composed]")
+            filter_complex += f";[composed]{drawtext_filter}[outv]"
+            logger.info(
+                "[#1789] 标题 drawtext 滤镜已注入: plan_id=%s text=%s",
+                plan_id,
+                (title_cfg.get("text") or title_cfg.get("content") or "")[:30],
+            )
+
         # 构建完整命令
         command: list[str] = ["ffmpeg", "-y"]
 

apps/api/app/tasks/__init__.py

@@ -0,0 +1 @@
+"""Celery 异步任务模块."""

apps/api/app/tasks/ai_avatar_render.py

@@ -0,0 +1,48 @@
+"""AI数字人渲染 Celery 异步任务 — #1798."""
+
+from __future__ import annotations
+
+import logging
+
+from app.core.celery_app import celery_app
+from app.dependencies import get_db_session
+
+logger = logging.getLogger(__name__)
+
+
+@celery_app.task(bind=True, name="ai_avatar_render.execute", max_retries=2)
+def execute_ai_avatar_render(self, job_id: str) -> dict:
+    """执行 AI 数字人渲染管线.
+
+    进度更新:
+    - 0%: 任务开始
+    - 20%: 下载对口型视频完成
+    - 40%: 滤镜链构建完成
+    - 80%: FFmpeg 渲染完成
+    - 95%: 上传 OSS 完成
+    - 100%: 任务完成
+    """
+    logger.info("开始执行渲染任务: %s", job_id)
+    self.update_state(state="PROCESSING", meta={"progress": 0, "job_id": job_id})
+
+    try:
+        # 获取数据库 session
+        db_gen = get_db_session()
+        db = next(db_gen)
+        try:
+            from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+            service = AiAvatarRenderService(db)
+            service.execute_render(job_id)
+        finally:
+            try:
+                next(db_gen)
+            except StopIteration:
+                pass
+
+        return {"status": "completed", "job_id": job_id}
+
+    except Exception as exc:
+        logger.exception("渲染任务执行异常 [%s]: %s", job_id, exc)
+        self.update_state(state="FAILED", meta={"progress": 0, "error": str(exc)})
+        raise

apps/web/e2e/core-generation.spec.ts

@@ -185,6 +185,13 @@ test.describe("Core generation flow", () => {
     await expect(page.locator(".xx-choice-item.selected")).toBeVisible()
     await page.getByRole("button", { name: "下一步" }).click()
 
+    // Step1 下一步弹出数量选择弹窗（Issue #1677 固定6步：模板→素材→配音→标题→确认生成→封面）
+    // 单视频流程：默认 1 个，点击「生成 1 个视频」进入步骤2
+    await expect(page.getByRole("heading", { name: "要生成几个视频？" })).toBeVisible({
+      timeout: 10_000,
+    })
+    await page.getByRole("button", { name: "生成 1 个视频" }).click()
+
     // Step 2: select material (card grid UI)
     await expect(page.getByRole("heading", { name: /选择素材/ })).toBeVisible()
     const librarySelect = page.locator("select").first()
@@ -255,32 +262,19 @@ test.describe("Core generation flow", () => {
       expect(genData.items.length).toBeGreaterThan(0)
       expect(genData.items[0].id).toBeTruthy()
 
-      // Step 5: 确认生成页 — 任务创建成功后自动跳转，展示渲染进度
-      await expect(page.getByRole("heading", { name: /确认生成/ })).toBeVisible({
-        timeout: 15_000,
+      // 单视频（N=1）：点击「确认生成视频」后跳 Step 5「确认生成」，展示实时渲染进度
+      await expect(page.getByRole("heading", { name: "🎬 确认生成" })).toBeVisible({
+        timeout: 30_000,
       })
 
-      // Step 5 → Step 6：等待渲染终态
-      // - 完成：页面出现「视频生成完成」，步骤5「下一步」按钮解锁，点击进入封面
-      // - 失败：出现「生成失败」，停在确认生成页也算向导流程走通
-      // - 超时未终态（测试环境 worker 可能不处理任务）：进度仍在轮询，同样算走通
-      const renderSucceeded = await page
-        .getByText("视频生成完成", { exact: false })
-        .waitFor({ timeout: 180_000 })
-        .then(() => true)
-        .catch(() => false)
-      if (renderSucceeded) {
-        // 渲染完成：手动点「下一步」进入封面步骤（渲染完不自动跳转）
-        await page.getByRole("button", { name: "下一步" }).click()
-        // Step 6: 封面（最后一步，无主按钮），仅验证页面渲染
-        await expect(page.getByRole("heading", { name: /选择封面/ })).toBeVisible({
-          timeout: 15_000,
-        })
-      } else {
-        // 失败或超时：仍在确认生成页（进度展示或失败提示），向导流程已完整走通
-        await expect(page.getByRole("heading", { name: /确认生成/ })).toBeVisible()
-        console.log("[E2E] 渲染任务失败或未在 180s 内完成，冒烟测试仍通过（已达确认生成页）")
-      }
+      // 等待渲染完成：进度卡变为「视频生成完成」（最长等待 3 分钟）
+      await expect(page.getByText("视频生成完成")).toBeVisible({ timeout: 180_000 })
+
+      // 全部完成后「下一步：选择封面」解锁，点击进入 Step 6
+      await page.getByRole("button", { name: /下一步：选择封面/ }).click()
+      await expect(page.getByRole("heading", { name: /选择封面/ })).toBeVisible({
+        timeout: 30_000,
+      })
     } else {
       console.log(`[E2E] Generate API returned ${genResp.status()}, wizard flow test still passes`)
       // 创建失败时停留在标题页并展示错误提示

apps/web/public/vite.svg

@@ -0,0 +1,4 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64">
+  <rect width="64" height="64" rx="14" fill="#3b82f6"/>
+  <text x="32" y="44" font-size="34" text-anchor="middle">🦐</text>
+</svg>

apps/web/src/api/assets/libraries.ts

@@ -5,10 +5,22 @@ import apiClient from "../client"
 import { getOrCreateDefaultProject } from "../projects"
 import type { AssetLibraryItem } from "./types"
 
-/** 获取当前用户的所有素材库 */
-export const getAssetLibraries = async (): Promise<AssetLibraryItem[]> => {
-  const response = await apiClient.get("/asset-libraries")
-  return response.data.items || []
+/**
+ * 获取当前用户的素材库
+ *
+ * @param kind 可选，按素材库类型过滤（video/voice/image）。
+ *   后端 GET /asset-libraries 支持 kind 查询参数；这里同时在前端再按返回数据的
+ *   kind 字段兜底过滤一次，保证旧后端（忽略未知 query 参数）也不会把其他类型的库
+ *   混进来（#1777：视频选择器只展示视频库）。
+ */
+export const getAssetLibraries = async (
+  kind?: AssetLibraryItem["kind"],
+): Promise<AssetLibraryItem[]> => {
+  const response = await apiClient.get<{ items?: AssetLibraryItem[] }>("/asset-libraries", {
+    params: kind ? { kind } : undefined,
+  })
+  const items = response.data.items || []
+  return kind ? items.filter((lib) => lib.kind === kind) : items
 }
 
 /** 创建素材库（自动获取或创建默认项目以提供 project_id） */

apps/web/src/api/assets/types.ts

@@ -139,6 +139,17 @@ export interface DirectUploadPrepareResult {
    * 旧后端不返回该字段，前端降级为无预建卡片的原有行为。
    */
   asset_id?: string
+  /**
+   * 后端 file_hash 命中素材库已有相同文件时为 true，前端应跳过 transfer + complete 阶段
+   * 直接按「去重命中」处理（不调 transfer、不调 complete、立即刷新素材列表）。
+   * 旧后端不返回该字段，前端降级为走老流程。
+   */
+  duplicated?: boolean
+  /**
+   * 与 duplicated 语义一致：true 表示跳过传输，前端据此短路。
+   * 两个字段是同一语义的别名（后端可能只返回其一），前端任意为 true 即视为命中去重。
+   */
+  skip_transfer?: boolean
 }
 
 /** 直传完成确认返回 */

apps/web/src/api/assets/upload.ts

@@ -4,6 +4,7 @@
 import apiClient from "../client"
 import { getOrCreateDefaultProject } from "../projects"
 import type { DirectUploadPrepareResult, DirectUploadCompleteResult } from "./types"
+import { computeFileHash, makeClientUploadId } from "./uploadDedup"
 
 /** 预签名直传准备 */
 export const prepareDirectUpload = async (data: {
@@ -12,8 +13,13 @@ export const prepareDirectUpload = async (data: {
   filename: string
   content_type: string
   file_size: number
+  /** 前端算好的文件内容哈希（SHA-256 hex），打开后端 file_hash 去重闸门 */
+  file_hash?: string
+  /** 前端生成的上传幂等 token，同一次逻辑上传（含重试）保持不变 */
+  client_upload_id?: string
 }): Promise<DirectUploadPrepareResult> => {
-  const response = await apiClient.post("/upload/direct/prepare", data)
+  // prepare 单独放宽到 30s（全局 axios 实例只有 10s，staging 抖动时易超时）
+  const response = await apiClient.post("/upload/direct/prepare", data, { timeout: 30_000 })
   return response.data
 }
 
@@ -22,8 +28,16 @@ export const completeDirectUpload = async (data: {
   project_id: string
   library_id: string
   storage_key: string
+  /** 前端算好的文件内容哈希（与 prepare 一致），后端按 hash 幂等去重 */
+  file_hash?: string
+  /** 前端上传幂等 token（与 prepare 一致），同一次上传重发 complete 不重复建记录 */
+  client_upload_id?: string
+  /** 文件字节数；后端同名兜底去重需用它做大小校验，缺失（=0）时同名记录一律不判重 */
+  file_size?: number
 }): Promise<DirectUploadCompleteResult> => {
-  const response = await apiClient.post("/upload/direct/complete", data)
+  // complete 内含 OSS 存在性检查 + 建库 + 派单，放宽到 60s；
+  // 超时不代表失败（记录可能已建成），调用方禁止超时后盲目重传整个文件
+  const response = await apiClient.post("/upload/direct/complete", data, { timeout: 60_000 })
   return response.data
 }
 
@@ -109,8 +123,20 @@ export interface DirectUploadHandle {
 export const prepareDirectUploadHandle = async (data: {
   file: File
   library_id: string
+  /** 前端算好的文件内容哈希（SHA-256 hex），prepare/complete 均携带 */
+  fileHash?: string
+  /** 本次逻辑上传的幂等 token，prepare/complete 一致、重试复用 */
+  clientUploadId?: string
 }): Promise<DirectUploadHandle> => {
-  const project = await getOrCreateDefaultProject()
+  // 默认项目初始化失败（项目列表接口异常/自动创建失败）给出独立、明确的提示，
+  // 不与 prepare 的签名接口错误混在一起
+  let project: Awaited<ReturnType<typeof getOrCreateDefaultProject>>
+  try {
+    project = await getOrCreateDefaultProject()
+  } catch (err) {
+    const reason = err instanceof Error ? err.message : "网络异常"
+    throw new Error(`初始化默认项目失败，无法开始上传：${reason}`)
+  }
 
   const prepared = await prepareDirectUpload({
     project_id: project.id,
@@ -118,6 +144,8 @@ export const prepareDirectUploadHandle = async (data: {
     filename: data.file.name,
     content_type: data.file.type || "application/octet-stream",
     file_size: data.file.size,
+    file_hash: data.fileHash,
+    client_upload_id: data.clientUploadId,
   })
 
   return {
@@ -128,6 +156,10 @@ export const prepareDirectUploadHandle = async (data: {
         project_id: project.id,
         library_id: data.library_id,
         storage_key: prepared.storage_key,
+        file_hash: data.fileHash,
+        client_upload_id: data.clientUploadId,
+        // 透传文件字节数：后端同名兜底去重依赖大小校验，缺省会导致同名新视频被误判重复
+        file_size: data.file.size,
       }),
   }
 }
@@ -137,8 +169,30 @@ export const uploadAssetDirect = async (data: {
   file: File
   library_id: string
   onProgress?: (percent: number) => void
+  /** 文件内容哈希；未传时自动补算（配音/封面/克隆等非队列链路统一受益） */
+  fileHash?: string
+  /** 幂等 token；未传时自动生成 */
+  clientUploadId?: string
 }): Promise<DirectUploadCompleteResult> => {
-  const handle = await prepareDirectUploadHandle({ file: data.file, library_id: data.library_id })
+  // 自动补算哈希与幂等 token：确保 file_hash 去重闸门对所有上传链路生效
+  const fileHash = data.fileHash ?? (await computeFileHash(data.file))
+  const clientUploadId = data.clientUploadId ?? makeClientUploadId()
+  const handle = await prepareDirectUploadHandle({
+    file: data.file,
+    library_id: data.library_id,
+    fileHash,
+    clientUploadId,
+  })
+  // prepare 阶段后端 file_hash 命中素材库已有相同文件：跳过 transfer + complete
+  if (handle.prepared.skip_transfer || handle.prepared.duplicated) {
+    return {
+      storage_key: handle.prepared.storage_key,
+      ingest_job_id: "",
+      url: "",
+      duplicated: true,
+      asset_id: handle.prepared.asset_id,
+    }
+  }
   await handle.transfer(data.onProgress)
   return handle.complete()
 }

apps/web/src/api/assets/uploadDedup.ts

@@ -0,0 +1,143 @@
+/**
+ * 上传去重 / 幂等工具（Issue #1714）
+ *
+ * 背景：同一文件被反复入队、complete 超时后盲目重传，导致后端创建大量重复
+ * PROCESSING 素材记录。本模块提供两类纯函数：
+ *
+ * 1. 文件指纹：
+ *    - makeFileFingerprint()：文件名+大小+lastModified，入队去重用（同步、零开销）
+ *    - computeFileHash()：SHA-256 内容哈希（小文件全量、大文件抽样头尾），
+ *      prepare/complete 时发给后端打开 file_hash 去重闸门
+ * 2. 队列去重：findDuplicateInQueue() 判断文件是否已在队列中
+ * 3. 幂等 token：makeClientUploadId() 生成上传幂等 ID（每次"一次逻辑上传"一个，
+ *    重试复用同一 ID，重新入队才生成新 ID）
+ */
+
+/** 全量哈希阈值：≤64MB 全量读入计算；超过即走头尾抽样，避免 100~256MB 视频被整文件读进内存卡死页面 */
+export const HASH_FULL_READ_LIMIT = 64 * 1024 * 1024 // 64MB
+/** 抽样读取的头尾片段大小（各 16MB） */
+export const HASH_SAMPLE_CHUNK = 16 * 1024 * 1024
+
+/** 计算指纹时，文件在队列中已存在的状态（已失败的可以重试，不算重复） */
+export type DedupExcludeStatus = "error" | "done"
+
+/**
+ * 文件入队指纹：同库 + 文件名 + 大小 + 修改时间。
+ * 同一文件（File 对象由 <input> 重选或拖拽重复触发时三个字段均一致）稳定复现；
+ * 不同文件极小概率碰撞时可由后端 file_hash 内容去重兜底。
+ */
+export function makeFileFingerprint(file: Pick<File, "name" | "size" | "lastModified">): string {
+  return `${file.name}::${file.size}::${file.lastModified}`
+}
+
+/**
+ * 在现有队列项中查找同一文件的在途记录。
+ * 已失败（error）的项允许重试路径复用、已完成（done）的可跳过；
+ * 处于 preparing/uploading/ingesting 的在途项一律视为重复，禁止重复入队。
+ *
+ * 返回命中的队列项 id（tempId），未命中返回 null。
+ */
+export function findDuplicateInQueue<T extends { fileKey: string; status: string }>(
+  queue: T[],
+  fileKey: string,
+  excludeStatuses: DedupExcludeStatus[] = [],
+): T | null {
+  const exclude = new Set<string>(excludeStatuses)
+  return queue.find((it) => it.fileKey === fileKey && !exclude.has(it.status)) ?? null
+}
+
+/** 生成上传幂等 token：一次"逻辑上传"一个，重试复用、重新入队换新 */
+export function makeClientUploadId(): string {
+  const rand =
+    typeof crypto !== "undefined" && "randomUUID" in crypto
+      ? crypto.randomUUID()
+      : `${Date.now()}-${Math.random().toString(36).slice(2, 10)}-${Math.random()
+          .toString(36)
+          .slice(2, 10)}`
+  return `up_${Date.now().toString(36)}_${rand.replace(/-/g, "").slice(0, 16)}`
+}
+
+/** 读取 Blob/File 片段为 ArrayBuffer：优先 Blob.arrayBuffer()，老环境回退 FileReader */
+function readAsArrayBuffer(blob: Blob): Promise<ArrayBuffer> {
+  if (typeof blob.arrayBuffer === "function") {
+    return blob.arrayBuffer()
+  }
+  return new Promise<ArrayBuffer>((resolve, reject) => {
+    const reader = new FileReader()
+    reader.onload = () => resolve(reader.result as ArrayBuffer)
+    reader.onerror = () => reject(reader.error ?? new Error("FileReader read failed"))
+    reader.readAsArrayBuffer(blob)
+  })
+}
+
+/**
+ * 把 buffer 复制到当前 JS realm 的 Uint8Array 再哈希。
+ * jsdom/测试环境中 Blob.arrayBuffer() 可能返回另一 realm 的 ArrayBuffer，
+ * Node WebCrypto 的 WebIDL instanceof 校验会拒绝跨 realm 参数。
+ */
+async function digestSha256(buffer: ArrayBuffer): Promise<ArrayBuffer> {
+  const subtle =
+    typeof globalThis !== "undefined" && globalThis.crypto ? globalThis.crypto.subtle : null
+  if (!subtle) throw new Error("crypto.subtle unavailable")
+  const local = new Uint8Array(buffer.byteLength)
+  local.set(new Uint8Array(buffer))
+  return subtle.digest("SHA-256", local)
+}
+
+function toHex(buffer: ArrayBuffer): string {
+  const bytes = new Uint8Array(buffer)
+  let hex = ""
+  for (let i = 0; i < bytes.length; i += 1) {
+    hex += bytes[i].toString(16).padStart(2, "0")
+  }
+  return hex
+}
+
+/**
+ * 计算文件内容 SHA-256（hex，64 字符，与后端 file_hash 字段长度一致）。
+ * - ≤64MB：全量哈希，内容一致必然一致
+ * - >64MB：哈希「头部 16MB + 尾部 16MB + 文件大小」，视频素材体积大、
+ *   头部含 moov 元数据、尾部含 mdat 结尾，抽样碰撞概率可忽略，
+ *   且避免 100~256MB 视频被整文件读进内存导致页面卡死/崩溃
+ *
+ * 运行环境不支持 crypto.subtle（非安全上下文/老浏览器）时返回空字符串，
+ * 调用方据此降级为不传 hash（后端仍有幂等 token + 同文件名兜底去重）。
+ */
+export async function computeFileHash(file: File): Promise<string> {
+  try {
+    const subtle =
+      typeof globalThis !== "undefined" &&
+      globalThis.crypto &&
+      typeof globalThis.crypto.subtle?.digest === "function"
+        ? globalThis.crypto.subtle
+        : null
+    if (!subtle) return ""
+
+    if (file.size <= HASH_FULL_READ_LIMIT) {
+      const data = await readAsArrayBuffer(file.slice(0, file.size))
+      return toHex(await digestSha256(data))
+    }
+
+    // 大文件：头 8MB + 尾 8MB + 大小，拼成一段后哈希
+    const head = await readAsArrayBuffer(file.slice(0, HASH_SAMPLE_CHUNK))
+    const tail =
+      file.size > HASH_SAMPLE_CHUNK
+        ? await readAsArrayBuffer(file.slice(Math.max(0, file.size - HASH_SAMPLE_CHUNK), file.size))
+        : new ArrayBuffer(0)
+    const merged = new Uint8Array(head.byteLength + tail.byteLength + 8)
+    merged.set(new Uint8Array(head), 0)
+    merged.set(new Uint8Array(tail), head.byteLength)
+    const sizeView = new DataView(merged.buffer, head.byteLength + tail.byteLength, 8)
+    // 文件大小以 64 位大端写入（BigInt 最稳；不支持 BigInt64 时手算高低位）
+    if (typeof sizeView.setBigUint64 === "function") {
+      sizeView.setBigUint64(0, BigInt(file.size), false)
+    } else {
+      sizeView.setUint32(0, Math.floor(file.size / 0x100000000), false)
+      sizeView.setUint32(4, file.size >>> 0, false)
+    }
+    return toHex(await digestSha256(merged.buffer))
+  } catch (err) {
+    console.warn("[uploadDedup] 计算文件哈希失败，降级为不传 file_hash:", err)
+    return ""
+  }
+}

apps/web/src/api/auth/index.ts

@@ -12,13 +12,18 @@ export type {
   UserResponse,
   WechatAuthUrlResponse,
   WechatCallbackResponse,
+  WechatBindUrlResponse,
+  WechatBindCompleteResponse,
+  WechatUnbindResponse,
+  UpdateProfileRequest,
+  UpdateProfileResponse,
   SendVerificationCodeRequest,
   BindContactRequest,
   BindContactResponse,
 } from "./types"
 
 // 用户工具函数
-export { normalizeUser } from "./user"
+export { normalizeUser, updateProfile } from "./user"
 
 // 登录/注册/登出/刷新
 export { login, refreshAccessToken, register, logout } from "./login"
@@ -32,8 +37,14 @@ export { requestPasswordReset, resetPassword } from "./password"
 // 邮箱验证
 export { verifyEmail } from "./email"
 
-// 微信登录
-export { getWechatAuthUrl, wechatCallback } from "./wechat"
+// 微信登录 / 绑定
+export {
+  getWechatAuthUrl,
+  wechatCallback,
+  getWechatBindUrl,
+  bindWechat,
+  unbindWechat,
+} from "./wechat"
 
 // 联系方式
 export { sendVerificationCode, bindContact } from "./contact"

apps/web/src/api/auth/types.ts

@@ -34,6 +34,17 @@ export interface User {
   is_email_verified: boolean
   email_verified: boolean
   created_at?: string
+  /** 微信是否已绑定 */
+  wechat_bound?: boolean
+  /** 微信昵称（绑定后展示） */
+  wechat_nickname?: string
+  /** 头像 URL（微信头像等） */
+  avatar_url?: string
+  /** 手机号 */
+  phone?: string
+  phone_verified?: boolean
+  /** 资料是否完善（微信新用户首次登录为 false，需填昵称引导） */
+  profile_completed?: boolean
 }
 
 export interface UserResponse {
@@ -45,6 +56,12 @@ export interface UserResponse {
   is_email_verified?: boolean
   email_verified?: boolean
   created_at?: string
+  wechat_bound?: boolean
+  wechat_nickname?: string
+  avatar_url?: string
+  phone?: string
+  phone_verified?: boolean
+  profile_completed?: boolean
 }
 
 export interface WechatAuthUrlResponse {
@@ -80,3 +97,30 @@ export interface BindContactResponse {
   success: boolean
   user: User
 }
+
+/** 更新个人资料请求 */
+export interface UpdateProfileRequest {
+  display_name?: string
+}
+
+/** 更新个人资料响应（返回最新用户信息） */
+export interface UpdateProfileResponse {
+  user: UserResponse
+}
+
+/** 微信绑定授权链接响应 */
+export interface WechatBindUrlResponse {
+  auth_url: string
+  state: string
+}
+
+/** 微信绑定完成响应 */
+export interface WechatBindCompleteResponse {
+  success: boolean
+  user: UserResponse
+}
+
+/** 微信解绑响应 */
+export interface WechatUnbindResponse {
+  success: boolean
+}

apps/web/src/api/auth/user.ts

@@ -1,4 +1,5 @@
-import type { User, UserResponse } from "./types"
+import apiClient from "../client"
+import type { User, UserResponse, UpdateProfileRequest, UpdateProfileResponse } from "./types"
 
 /**
  * 规范化用户数据，兼容不同后端返回格式
@@ -16,5 +17,19 @@ export const normalizeUser = (data: UserResponse): User => {
     is_email_verified: emailVerified,
     email_verified: emailVerified,
     created_at: data.created_at,
+    wechat_bound: data.wechat_bound,
+    wechat_nickname: data.wechat_nickname,
+    avatar_url: data.avatar_url,
+    phone: data.phone,
+    phone_verified: data.phone_verified,
+    profile_completed: data.profile_completed,
   }
 }
+
+/**
+ * 更新个人资料（昵称等）
+ */
+export const updateProfile = async (data: UpdateProfileRequest): Promise<User> => {
+  const response = await apiClient.patch<UpdateProfileResponse>("/auth/me", data)
+  return normalizeUser(response.data.user)
+}

apps/web/src/api/auth/wechat.ts

@@ -1,8 +1,14 @@
 import apiClient from "../client"
-import type { WechatAuthUrlResponse, WechatCallbackResponse } from "./types"
+import type {
+  WechatAuthUrlResponse,
+  WechatCallbackResponse,
+  WechatBindUrlResponse,
+  WechatBindCompleteResponse,
+  WechatUnbindResponse,
+} from "./types"
 
 /**
- * 获取微信授权链接
+ * 获取微信授权链接（登录场景）
  */
 export const getWechatAuthUrl = async (): Promise<WechatAuthUrlResponse> => {
   const response = await apiClient.get("/auth/wechat/url")
@@ -19,3 +25,30 @@ export const wechatCallback = async (
   const response = await apiClient.post("/auth/wechat/callback", { code, state })
   return response.data
 }
+
+/**
+ * 获取微信绑定授权链接（已登录用户绑定场景）
+ */
+export const getWechatBindUrl = async (): Promise<WechatBindUrlResponse> => {
+  const response = await apiClient.get("/auth/wechat/bind/url")
+  return response.data
+}
+
+/**
+ * 微信绑定完成（扫码回调后用 code 绑定到当前登录账号）
+ */
+export const bindWechat = async (
+  code: string,
+  state: string,
+): Promise<WechatBindCompleteResponse> => {
+  const response = await apiClient.post("/auth/wechat/bind", { code, state })
+  return response.data
+}
+
+/**
+ * 解绑微信
+ */
+export const unbindWechat = async (): Promise<WechatUnbindResponse> => {
+  const response = await apiClient.delete("/auth/wechat/bind")
+  return response.data
+}

apps/web/src/api/auth/wxLogin.ts

@@ -0,0 +1,112 @@
+/**
+ * 微信扫码登录 WxLogin JS-SDK 动态加载与授权参数解析
+ *
+ * 微信官网嵌入式二维码方案：页面引入 https://res.wx.qq.com/connect/zh_CN/htmledition/js/wxLogin.js
+ * 后挂载全局 window.WxLogin，new WxLogin({...}) 会在指定容器内渲染二维码 iframe。
+ * 本模块负责：动态加载该脚本（带超时/失败检测）、从后端返回的 auth_url 中解析
+ * WxLogin 所需的 appid / redirect_uri / state。
+ */
+
+const WX_LOGIN_SRC = "https://res.wx.qq.com/connect/zh_CN/htmledition/js/wxLogin.js"
+/** 脚本加载超时（毫秒）：超时视为加载失败，调用方回退整页跳转 */
+const WX_LOGIN_LOAD_TIMEOUT = 8000
+
+/** WxLogin 构造参数（微信官方字段，保持原名） */
+export interface WxLoginOptions {
+  /** 是否内嵌二维码（回调在 iframe 内完成） */
+  self_redirect: boolean
+  /** 二维码容器元素 id */
+  id: string
+  /** 微信开放平台 AppID */
+  appid: string
+  /** 应用授权作用域，网站应用固定 snsapi_login */
+  scope: "snsapi_login"
+  /** 回调地址（需与微信开放平台配置一致，WxLogin 内部会 encodeURIComponent） */
+  redirect_uri: string
+  /** 防 CSRF 随机串，由后端 state store 生成并在回调时一次性消费 */
+  state: string
+  /** 二维码样式：black / white */
+  style?: "black" | "white"
+  /** 自定义样式链接（可选） */
+  href?: string
+}
+
+/** 微信脚本挂载到 window 上的全局构造函数类型 */
+export interface WxLoginConstructor {
+  new (options: WxLoginOptions): unknown
+}
+
+declare global {
+  interface Window {
+    WxLogin?: WxLoginConstructor
+  }
+}
+
+let loadPromise: Promise<WxLoginConstructor> | null = null
+
+/**
+ * 动态加载微信 WxLogin JS（单例：并发调用复用同一个 promise）。
+ * 加载失败或超时会 reject，调用方应回退到整页跳转授权方式。
+ */
+export function loadWxLoginScript(): Promise<WxLoginConstructor> {
+  if (window.WxLogin) return Promise.resolve(window.WxLogin)
+  if (loadPromise) return loadPromise
+
+  loadPromise = new Promise<WxLoginConstructor>((resolve, reject) => {
+    const script = document.createElement("script")
+    script.src = WX_LOGIN_SRC
+    script.async = true
+    script.onload = () => {
+      if (window.WxLogin) {
+        resolve(window.WxLogin)
+      } else {
+        loadPromise = null
+        reject(new Error("微信登录脚本加载完成但 WxLogin 未挂载"))
+      }
+    }
+    script.onerror = () => {
+      loadPromise = null
+      script.remove()
+      reject(new Error("微信登录脚本加载失败"))
+    }
+    document.head.appendChild(script)
+
+    // 超时兜底：部分网络环境下脚本既不 onload 也不 onerror
+    window.setTimeout(() => {
+      if (window.WxLogin) {
+        resolve(window.WxLogin)
+        return
+      }
+      loadPromise = null
+      script.remove()
+      reject(new Error("微信登录脚本加载超时"))
+    }, WX_LOGIN_LOAD_TIMEOUT)
+  })
+
+  return loadPromise
+}
+
+/** 从微信授权链接 query 中解析出的 WxLogin 所需参数 */
+export interface ParsedWxAuthParams {
+  appid: string
+  /** 已 URL 解码的回调地址（传给 WxLogin 时由其内部再次编码） */
+  redirect_uri: string
+  state: string
+}
+
+/**
+ * 从后端返回的微信授权链接（https://open.weixin.qq.com/connect/qrconnect?appid=...&redirect_uri=...&state=...）
+ * 中解析 appid / redirect_uri / state。解析失败时返回 null，由调用方回退整页跳转。
+ */
+export function parseWxAuthUrl(authUrl: string, stateFallback?: string): ParsedWxAuthParams | null {
+  try {
+    const url = new URL(authUrl)
+    const appid = url.searchParams.get("appid")
+    const redirectUri = url.searchParams.get("redirect_uri")
+    const state = url.searchParams.get("state") || stateFallback || ""
+    if (!appid || !redirectUri || !state) return null
+    return { appid, redirect_uri: redirectUri, state }
+  } catch {
+    return null
+  }
+}

apps/web/src/api/client.ts

@@ -55,6 +55,19 @@ apiClient.interceptors.response.use(
   async (error: AxiosError<{ detail?: string; message?: string; msg?: string }>) => {
     const originalRequest = error.config as InternalAxiosRequestConfig & {
       _retry?: boolean
+      /**
+       * 调用方自行处理错误提示时置 true：拦截器跳过全局 message 弹窗（#1777）。
+       * 例如失效模板自动回退时，调用方会弹「原模板已失效，已自动切换」，
+       * 不再叠加后端原始错误文案。错误仍会 reject，不影响 catch 逻辑。
+       */
+      _silentErrorToast?: boolean
+    }
+
+    // 调用方声明自行处理提示：标记为已展示，跳过下面所有全局 message 弹窗
+    if (originalRequest?._silentErrorToast) {
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      ;(error as any).__msgShown = true
+      return Promise.reject(error)
     }
 
     // 401 → 尝试刷新 Token

apps/web/src/api/editing-planner/templates.ts

@@ -12,17 +12,25 @@ import type {
   ListCategoriesResponse,
 } from "./types"
 
-/** 获取模板列表 */
+/** 获取模板列表
+ *
+ * valid_only=true 时请求后端仅返回已配置片段的模板（剪辑页选模板使用，
+ * 避免选中无片段配置的模板导致 from-assets 400，#1769/#1772）；
+ * 后端尚未支持该参数时会忽略未知 query 字段，前端再按 segments/is_active 兜底过滤。
+ * 模板编辑器/我的模板不传，可查看全部模板（含未配置片段的草稿）。
+ */
 export const getEditingTemplates = async (params?: {
   category?: string
   tag?: string
   skip?: number
   limit?: number
+  validOnly?: boolean
 }): Promise<EditingTemplate[]> => {
   const response = await apiClient.get<ListTemplatesResponse>("/templates", {
     params: {
       skip: params?.skip ?? 0,
       limit: params?.limit ?? 50,
+      ...(params?.validOnly ? { valid_only: true } : {}),
     },
   })
   let list = response.data.items

apps/web/src/api/editing-planner/types.ts

@@ -44,8 +44,10 @@ export interface BgmConfig {
 export interface TemplateSegment {
   id?: string
   segment_order: number
-  duration_min: number
-  duration_max: number
+  /** @deprecated 模板无时长概念（#1750 基线）：字段保留仅为兼容旧数据读取，新模板可不传 */
+  duration_min?: number
+  /** @deprecated 同上 */
+  duration_max?: number
   material_type: string | null
 }
 
@@ -59,7 +61,8 @@ export interface EditingTemplate {
   title_config: TitleConfig
   subtitle_config: SubtitleConfig
   bgm_config: BgmConfig
-  estimated_duration: number
+  /** @deprecated 模板无时长概念（#1750 基线）：成片时长由配音时长决定；字段保留兼容旧数据 */
+  estimated_duration?: number
   segments: TemplateSegment[]
   watermark_config?: WatermarkConfig
   intro_outro_config?: IntroOutroConfig
@@ -89,7 +92,8 @@ export interface SaveTemplatePayload {
   title_config: TitleConfig
   subtitle_config: SubtitleConfig
   bgm_config: BgmConfig
-  estimated_duration: number
+  /** @deprecated 模板无时长概念（#1750 基线）：保留兼容旧数据 */
+  estimated_duration?: number
   segments: Omit<TemplateSegment, "id">[]
   watermark_config?: WatermarkConfig
   intro_outro_config?: IntroOutroConfig

apps/web/src/api/errors.ts

@@ -0,0 +1,132 @@
+/**
+ * 统一错误信息提取
+ * 把 axios 错误（后端 detail / FastAPI 校验错误 / HTTP 状态码）、XHR/OSS 错误、
+ * 网络/超时错误、普通 Error 统一转成「可直接展示给用户」的中文信息。
+ *
+ * 与 api/client.ts 响应拦截器的提示口径保持一致；拦截器负责全局 toast，
+ * 页面/队列卡片用本工具把真实原因展示在持久位置（回调页、失败卡片等）。
+ */
+import type { AxiosError } from "axios"
+
+/** 后端错误响应体可能出现的字段（FastAPI：detail；历史接口：message/msg） */
+interface ErrorBody {
+  detail?: unknown
+  message?: unknown
+  msg?: unknown
+}
+
+/** FastAPI 422 校验错误单项 */
+interface ValidationItem {
+  loc?: (string | number)[]
+  msg?: string
+}
+
+/** 从后端响应体提取人类可读信息（detail 可能是字符串、对象、422 数组） */
+function extractBodyMessage(data: unknown): string {
+  if (!data || typeof data !== "object") return ""
+  const body = data as ErrorBody
+
+  const walk = (val: unknown): string => {
+    if (typeof val === "string") return val
+    if (Array.isArray(val)) {
+      // FastAPI 422: [{loc, msg, type}, ...] → 取每条 msg 拼接
+      const parts = val
+        .map((item) => {
+          if (typeof item === "string") return item
+          if (item && typeof item === "object") {
+            const v = item as ValidationItem
+            if (typeof v.msg === "string") {
+              const field = Array.isArray(v.loc) ? v.loc.filter((x) => x !== "body").join(".") : ""
+              return field ? `${field}: ${v.msg}` : v.msg
+            }
+            return walk(item)
+          }
+          return ""
+        })
+        .filter(Boolean)
+      return parts.join("；")
+    }
+    if (val && typeof val === "object") {
+      const obj = val as Record<string, unknown>
+      if (typeof obj.message === "string") return obj.message
+      if (typeof obj.msg === "string") return obj.msg
+      if (typeof obj.detail === "string") return obj.detail
+      if (obj.message && typeof obj.message === "object") return walk(obj.message)
+      if (obj.msg && typeof obj.msg === "object") return walk(obj.msg)
+      try {
+        return JSON.stringify(val)
+      } catch {
+        return ""
+      }
+    }
+    return ""
+  }
+
+  return walk(body.detail) || walk(body.message) || walk(body.msg)
+}
+
+/** 无响应体时按 HTTP 状态码给出兜底提示（与 client.ts 拦截器口径一致） */
+function statusFallback(status: number): string {
+  switch (status) {
+    case 400:
+      return "请求参数有误（HTTP 400）"
+    case 401:
+      return "登录状态已失效，请重新登录（HTTP 401）"
+    case 403:
+      return "没有权限执行该操作（HTTP 403）"
+    case 404:
+      return "请求的资源不存在（HTTP 404）"
+    case 409:
+      return "操作冲突，资源状态已变化（HTTP 409）"
+    case 413:
+      return "文件过大，请缩小后重试（HTTP 413）"
+    case 415:
+      return "不支持的文件格式（HTTP 415）"
+    case 429:
+      return "操作过于频繁，请稍后再试（HTTP 429）"
+    case 503:
+      return "服务暂不可用，请稍后再试（HTTP 503）"
+    default:
+      if (status >= 500) return `服务器繁忙，请稍后再试（HTTP ${status}）`
+      return `请求失败（HTTP ${status}）`
+  }
+}
+
+/**
+ * 从任意抛出值提取可展示的错误信息。
+ * @param fallback 全部提取失败时的兜底文案
+ */
+export function getErrorMessage(err: unknown, fallback = "操作失败，请稍后重试"): string {
+  if (!err) return fallback
+
+  // axios 错误（后端 JSON 响应 / HTTP 错误状态）
+  const ax = err as AxiosError<ErrorBody>
+  if (ax.isAxiosError || (typeof ax === "object" && "response" in (ax as object))) {
+    // 超时
+    if (ax.code === "ECONNABORTED" || /timeout/i.test(ax.message || "")) {
+      return "请求超时，请检查网络后重试"
+    }
+    const resp = ax.response
+    if (resp) {
+      const bodyMsg = extractBodyMessage(resp.data)
+      if (bodyMsg) return bodyMsg
+      return statusFallback(resp.status)
+    }
+    // 请求已发出但无响应（断网/CORS/DNS）
+    if (ax.request) return "网络连接异常，请检查网络设置"
+    return ax.message || fallback
+  }
+
+  if (err instanceof Error) {
+    // XHR 直传 OSS 失败等场景自带详细 message（含 HTTP 状态 + OSS Code/Message）
+    if (err.message) return err.message
+  }
+  if (typeof err === "string") return err
+
+  return fallback
+}
+
+/** client.ts 拦截器是否已对该错误弹过全局 toast（__msgShown 标记） */
+export function isErrorMsgShown(err: unknown): boolean {
+  return Boolean((err as { __msgShown?: boolean } | null)?.__msgShown)
+}

apps/web/src/api/generation/types.ts

@@ -33,15 +33,36 @@ export interface CreatePreviewRequest {
     preset_id?: string
     volume?: number
   }
+  /** 批量预览数量（1~10），默认1。N>1 时返回 N 个独立变体任务 */
+  preview_count?: number
+  /** 各变体独立标题文字：长度1=共用，长度=preview_count=独立，空数组=使用 title_config.text */
+  titles?: string[]
+  /** 各变体独立配音素材库ID：长度1=共用，长度=preview_count=独立，空数组=回退 voice_library_id */
+  voice_library_ids?: string[]
+  /** 各变体独立封面URL：长度1=共用，长度=preview_count=独立（预览阶段通常为空） */
+  cover_urls?: string[]
 }
 
-/** 创建预览任务响应 */
-export interface CreatePreviewResponse {
+/** 单个预览变体任务 */
+export interface PreviewVariantItem {
   task_id: string
-  status: PreviewStatus
+  status: string
+  progress: number
   is_preview: boolean
+  variant_index: number
   resolution: string
-  created_at: string
+  video_url: string
+  duration: number
+  error_message: string
+  title_text: string
+  voice_library_id: string
+  created_at?: string | null
+}
+
+/** 创建预览任务响应（单变体，preview_count=1 时 items 长度为1） */
+export interface CreatePreviewResponse {
+  items: PreviewVariantItem[]
+  total: number
   /** 后端自动关联的编辑计划 ID（用于 fallback 路径传递 source_edit_plan_id） */
   source_edit_plan_id?: string
 }

apps/web/src/api/generation/variantPlans.ts

@@ -0,0 +1,61 @@
+/**
+ * 批量变体剪辑计划 API（#1744）
+ *
+ * 批量预览时向后端申请 N 个变体的「独立剪辑计划片段」：
+ * - 变体 0 保留源 plan（用户在编辑器/智能选片产出的片段，含标题样式编辑结果）；
+ * - 变体 1..N-1 由后端 reselect_plan_for_variant 完整重跑单视频选片流程
+ *   （素材洗牌 + main 片段顺序洗牌 + 镜头/起点随机 + 跨变体 20% 区间避让 +
+ *   素材使用区间写回 metadata），与正式批量生成 POST /generation/tasks?count=N
+ *   使用同一套选片逻辑；
+ * - 正式生成时把 variant_plan_ids 原样回传，后端直接关联这些 plan 渲染，
+ *   不再重新选片 —— 预览所见即成片。
+ *
+ * 该接口只做选片/建 plan（秒级），不触发视频渲染，无渲染成本。
+ * 后端端点未上线（404）或选片失败（素材不足等）时前端降级为本地 variantSeed
+ * 模拟预览，不阻塞用户流程。
+ */
+import apiClient from "../client"
+import type { EditPlanClip } from "../template-editor"
+
+/** 批量变体计划请求体 */
+export interface BatchVariantPlansRequest {
+  template_id: string
+  /** 本批次素材池（手动选择或智能匹配结果） */
+  asset_ids: string[]
+  /** 变体数量（≥1）；=1 时只返回源 plan 片段 */
+  count: number
+  /** 源剪辑计划 ID：优先取预览/草稿关联的 plan；不传由后端按 template_id+user 兜底最新 plan */
+  source_edit_plan_id?: string
+}
+
+/** 单个变体的计划片段 */
+export interface VariantPlan {
+  /** 变体序号，从 0 开始 */
+  variant_index: number
+  /** 该变体关联的剪辑计划 ID（正式生成时回传，实现预览即成片） */
+  plan_id: string
+  /** 该变体的真实片段（顺序/素材/起点与正式成片一致） */
+  clips: EditPlanClip[]
+}
+
+/** 批量变体计划响应 */
+export interface BatchVariantPlansResponse {
+  items: VariantPlan[]
+  total: number
+}
+
+/**
+ * 创建批量变体剪辑计划并返回各变体片段。
+ *
+ * 注意：端点 404（后端未上线）/ 400（素材不足）等失败由调用方 catch 后降级，
+ * 不要抛 unhandled rejection。
+ */
+export async function createBatchVariantPlans(
+  params: BatchVariantPlansRequest,
+): Promise<BatchVariantPlansResponse> {
+  const response = await apiClient.post<BatchVariantPlansResponse>(
+    "/generation/variant-plans",
+    params,
+  )
+  return response.data
+}

apps/web/src/api/scripts/index.ts

@@ -0,0 +1,2 @@
+export * from "./scripts"
+export * from "./types"

apps/web/src/api/scripts/scripts.ts

@@ -0,0 +1,37 @@
+/**
+ * 文案库 API
+ * 对接后端 /api/v1/scripts（CRUD + 列表解包）
+ */
+import apiClient from "../client"
+import type {
+  ScriptItem,
+  ScriptListResponse,
+  CreateScriptRequest,
+  UpdateScriptRequest,
+} from "./types"
+
+/** 获取文案列表 — 必须解包 items（后端返回 {items,total}）*/
+export const getScripts = async (): Promise<ScriptItem[]> => {
+  const response = await apiClient.get<ScriptListResponse | ScriptItem[]>("/scripts")
+  const data = response.data as unknown
+  if (Array.isArray(data)) return data
+  const items = (data as { items?: ScriptItem[] })?.items
+  return Array.isArray(items) ? items : []
+}
+
+/** 新建文案 */
+export const createScript = async (data: CreateScriptRequest): Promise<ScriptItem> => {
+  const response = await apiClient.post<ScriptItem>("/scripts", data)
+  return response.data
+}
+
+/** 更新文案 */
+export const updateScript = async (id: string, data: UpdateScriptRequest): Promise<ScriptItem> => {
+  const response = await apiClient.put<ScriptItem>(`/scripts/${id}`, data)
+  return response.data
+}
+
+/** 删除文案 */
+export const deleteScript = async (id: string): Promise<void> => {
+  await apiClient.delete(`/scripts/${id}`)
+}

apps/web/src/api/scripts/types.ts

@@ -0,0 +1,24 @@
+/**
+ * 文案库 API — 类型定义
+ * 对接后端 /api/v1/scripts
+ */
+export interface ScriptItem {
+  id: string
+  title: string
+  content: string
+  char_count: number
+  created_at: string
+  updated_at?: string
+}
+
+export interface ScriptListResponse {
+  items: ScriptItem[]
+  total: number
+}
+
+export interface CreateScriptRequest {
+  title: string
+  content: string
+}
+
+export type UpdateScriptRequest = Partial<CreateScriptRequest>

apps/web/src/api/tasks/types.ts

@@ -92,6 +92,20 @@ export interface CreateGenerationTaskRequest {
     preset_id?: string
     volume?: number
   }
+  /** 批量生成数量（1~10），默认1。不传=单条旧逻辑 */
+  count?: number
+  /** 各变体独立标题文字：长度1=共用，长度=count=独立，空数组=使用 title_config/custom_title */
+  titles?: string[]
+  /** 各变体独立配音素材库ID：长度1=共用，长度=count=独立，空数组=回退 voice_library_id */
+  voice_library_ids?: string[]
+  /** 各变体独立封面URL：长度1=共用，长度=count=独立，空数组=回退 cover_url */
+  cover_urls?: string[]
+  /**
+   * 批量变体剪辑计划 ID（#1744）：预览阶段后端独立选片产出的 plan id 列表
+   * （按变体全量索引，长度=previewCount）。正式生成回传后后端直接关联这些
+   * plan 渲染、不再重新选片，保证预览所见即成片。后端未支持时忽略该字段。
+   */
+  variant_plan_ids?: string[]
 }
 
 /** 单个生成任务详情（对齐后端 GenerationTaskResponse） */

apps/web/src/api/template-editor/clips.ts

@@ -91,7 +91,7 @@ export async function createClipsFromAssets(
   assetIds: string[],
   clipType = "main",
   requiredClipsCount?: number,
-  opts?: { signal?: AbortSignal },
+  opts?: { signal?: AbortSignal; silentErrorToast?: boolean },
 ): Promise<ClipsFromAssetsResponse> {
   const body: Record<string, unknown> = {
     asset_ids: assetIds,
@@ -104,7 +104,12 @@ export async function createClipsFromAssets(
   const response = await apiClient.post<ClipsFromAssetsResponse>(
     `/templates/${templateId}/editor/clips/from-assets`,
     body,
-    { timeout: 60000, signal: opts?.signal },
+    {
+      timeout: 60000,
+      signal: opts?.signal,
+      // _silentErrorToast 由 api/client.ts 响应拦截器读取（抑制全局错误 toast，#1777）
+      ...(opts?.silentErrorToast ? ({ _silentErrorToast: true } as Record<string, unknown>) : {}),
+    },
   )
   return response.data
 }

apps/web/src/api/template-editor/editPlans.ts

@@ -43,11 +43,17 @@ export async function updateEditPlanClips(
   templateId: string,
   clips: EditPlanClipInput[],
   signal?: AbortSignal,
+  /** 为 true 时抑制全局错误 toast（调用方自行提示，如失效模板回退 #1777） */
+  silentErrorToast?: boolean,
 ): Promise<{ count: number }> {
   const response = await apiClient.put(
     `/templates/${templateId}/editor/clips`,
     { clips },
-    { signal },
+    {
+      signal,
+      // _silentErrorToast 由 api/client.ts 响应拦截器读取（抑制全局错误 toast）
+      ...(silentErrorToast ? ({ _silentErrorToast: true } as Record<string, unknown>) : {}),
+    },
   )
   return response.data
 }

apps/web/src/api/templates/types.ts

@@ -42,8 +42,10 @@ export interface TemplateItem {
 export interface TemplateSegment {
   id?: string
   segment_order: number
-  duration_min: number
-  duration_max: number
+  /** @deprecated 模板无时长概念（#1750 基线）：字段保留仅为兼容旧数据读取 */
+  duration_min?: number
+  /** @deprecated 同上 */
+  duration_max?: number
   material_type: string | null
   description?: string
 }
@@ -55,8 +57,6 @@ export interface TemplateListParams {
   category?: string
   tags?: string
   keyword?: string
-  /** 时长筛选（秒）：short < 30, medium 30-120, long > 120 */
-  duration_range?: "short" | "medium" | "long"
 }
 
 /** 模板列表分页响应 */

apps/web/src/api/tts/types.ts

@@ -103,6 +103,7 @@ export interface TTSPreviewRequest {
   voice_id: string
   speed?: number
   pitch?: number
+  emotion?: string // 情绪参数：natural/excited/calm/friendly
 }
 
 /** TTS 试听响应 */

apps/web/src/components/auth/WechatQrModal/WechatQrModal.css

@@ -0,0 +1,74 @@
+.xx-wechat-qr-modal {
+  position: relative;
+  padding: 8px 0 4px;
+  min-height: 320px;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  justify-content: center;
+}
+
+/* 常驻二维码容器（WxLogin 渲染目标） */
+.xx-wechat-qr-container {
+  display: flex;
+  justify-content: center;
+  min-height: 260px;
+}
+
+/* loading / error 遮罩层，覆盖在二维码容器之上 */
+.xx-wechat-qr-overlay {
+  position: absolute;
+  inset: 0;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  justify-content: center;
+  background: #fff;
+  text-align: center;
+  color: #666;
+}
+
+.xx-wechat-qr-overlay p {
+  margin-top: 16px;
+  margin-bottom: 0;
+}
+
+.xx-wechat-qr-container iframe {
+  border: none;
+}
+
+.xx-wechat-qr-tip {
+  margin: 12px 0 0;
+  color: #666;
+  font-size: 14px;
+}
+
+.xx-wechat-qr-error {
+  text-align: center;
+  width: 100%;
+}
+
+.xx-wechat-qr-error-msg {
+  color: #ef4444;
+  font-size: 14px;
+  line-height: 1.6;
+  margin: 0 0 16px;
+  word-break: break-word;
+}
+
+.xx-wechat-qr-error-actions {
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  gap: 12px;
+}
+
+.xx-wechat-qr-fallback {
+  background: none;
+  border: none;
+  color: var(--primary-color, #3b82f6);
+  cursor: pointer;
+  font-size: 13px;
+  padding: 0;
+  text-decoration: underline;
+}

apps/web/src/components/auth/WechatQrModal/index.tsx

@@ -0,0 +1,265 @@
+/**
+ * 微信扫码二维码弹窗（登录 / 绑定复用）
+ *
+ * 微信官方嵌入式二维码方案：弹窗内用 new WxLogin({ self_redirect: true }) 渲染二维码，
+ * 扫码后微信重定向到本站回调页（在二维码 iframe 内加载），回调页通过 postMessage
+ * 把成功/失败结果通知本弹窗（消息协议见 ./messages）。
+ *
+ * 兜底：获取授权链接成功但 WxLogin JS 加载失败/超时时，自动回退整页跳转授权
+ * （与旧流程一致）；获取授权链接本身失败时在弹窗内展示错误并提供重试。
+ */
+import React, { useEffect, useRef, useState } from "react"
+import { Spin } from "antd"
+import Modal from "@/components/ui/Modal"
+import Button from "@/components/ui/Button"
+import {
+  getWechatAuthUrl,
+  getWechatBindUrl,
+  getCurrentUser,
+  normalizeUser,
+  type User,
+} from "@/api/auth"
+import { useAuthStore } from "@/store/authStore"
+import { scheduleProactiveRefresh } from "@/api/auth/tokenRefresh"
+import { getErrorMessage } from "@/api/errors"
+import { loadWxLoginScript, parseWxAuthUrl } from "@/api/auth/wxLogin"
+import { isWechatQrMessage, type WechatQrScene } from "./messages"
+import "./WechatQrModal.css"
+
+export interface WechatQrModalProps {
+  open: boolean
+  scene: WechatQrScene
+  onClose: () => void
+  /** 登录场景成功回调（needOnboarding=true 时调用方应跳昵称引导页） */
+  onLoginSuccess?: (needOnboarding: boolean) => void
+  /** 绑定场景成功回调（调用方刷新用户信息/提示） */
+  onBindSuccess?: () => void
+}
+
+type QrStatus = "loading" | "qrcode" | "error"
+
+const CONTAINER_ID: Record<WechatQrScene, string> = {
+  login: "wechat-qr-login-container",
+  bind: "wechat-qr-bind-container",
+}
+
+const STATE_STORAGE_KEY: Record<WechatQrScene, string> = {
+  login: "wechat_state",
+  bind: "wechat_bind_state",
+}
+
+/**
+ * 等待二维码容器挂载到 DOM。antd Modal 内容通过 portal 渲染且带进场动画，
+ * 父组件 effect 首次执行时容器可能尚未出现在 document 中。
+ */
+function waitForContainer(id: string, timeoutMs = 3000): Promise<HTMLElement | null> {
+  return new Promise((resolve) => {
+    const start = Date.now()
+    const check = () => {
+      const el = document.getElementById(id)
+      if (el) {
+        resolve(el)
+        return
+      }
+      if (Date.now() - start > timeoutMs) {
+        resolve(null)
+        return
+      }
+      setTimeout(check, 50)
+    }
+    check()
+  })
+}
+
+const WechatQrModal: React.FC<WechatQrModalProps> = ({
+  open,
+  scene,
+  onClose,
+  onLoginSuccess,
+  onBindSuccess,
+}) => {
+  const setAuth = useAuthStore((state) => state.setAuth)
+  const setUser = useAuthStore((state) => state.setUser)
+  const [status, setStatus] = useState<QrStatus>("loading")
+  const [errorMsg, setErrorMsg] = useState("")
+  /** 刷新二维码计数：变化时重新请求授权链接并重渲染 */
+  const [renderSeq, setRenderSeq] = useState(0)
+  /** 最新授权链接，用于"整页打开"兜底 */
+  const authUrlRef = useRef<string | null>(null)
+
+  const isLogin = scene === "login"
+
+  // 初始化：获取授权链接 → 加载 WxLogin JS → 内嵌渲染二维码
+  useEffect(() => {
+    if (!open) return
+    let cancelled = false
+    authUrlRef.current = null
+    setStatus("loading")
+    setErrorMsg("")
+
+    const init = async () => {
+      try {
+        const fetchUrl = isLogin ? getWechatAuthUrl : getWechatBindUrl
+        const result = await fetchUrl()
+        if (cancelled) return
+        // 写 state（整页跳转兜底路径的回调页也会清理它）
+        localStorage.setItem(STATE_STORAGE_KEY[scene], result.state)
+        authUrlRef.current = result.auth_url
+
+        const params = parseWxAuthUrl(result.auth_url, result.state)
+        if (!params) {
+          // 授权链接格式异常：直接整页跳转，由微信侧/回调页兜底
+          window.location.href = result.auth_url
+          return
+        }
+
+        const WxLogin = await loadWxLoginScript()
+        if (cancelled) return
+        // 等 Modal portal 中的容器挂载完成
+        const container = await waitForContainer(CONTAINER_ID[scene])
+        if (cancelled) return
+        if (!container) {
+          window.location.href = result.auth_url
+          return
+        }
+        container.innerHTML = ""
+        new WxLogin({
+          self_redirect: true,
+          id: CONTAINER_ID[scene],
+          appid: params.appid,
+          scope: "snsapi_login",
+          redirect_uri: params.redirect_uri,
+          state: params.state,
+          style: "black",
+        })
+        if (!cancelled) setStatus("qrcode")
+      } catch (err) {
+        if (cancelled) return
+        if (authUrlRef.current) {
+          // 授权链接已拿到但二维码脚本加载失败/超时：回退整页跳转
+          window.location.href = authUrlRef.current
+          return
+        }
+        // 授权链接接口本身失败：弹窗内展示真实原因，允许重试
+        setErrorMsg(getErrorMessage(err, "微信服务暂不可用，请稍后重试"))
+        setStatus("error")
+      }
+    }
+
+    init()
+    return () => {
+      cancelled = true
+    }
+  }, [open, scene, isLogin, renderSeq])
+
+  // 监听 iframe 内回调页 postMessage 回来的扫码结果
+  useEffect(() => {
+    if (!open) return
+
+    const handleMessage = async (event: MessageEvent) => {
+      // 只接受同源消息
+      if (event.origin !== window.location.origin) return
+      if (!isWechatQrMessage(event.data, scene)) return
+      const msg = event.data
+
+      if (msg.success) {
+        if (isLogin) {
+          // iframe 内回调页已把 token 写入 localStorage（同源共享），
+          // 父窗口同步内存登录态后交给调用方跳转
+          try {
+            const userData = await getCurrentUser()
+            const user = normalizeUser(userData) as User
+            setAuth(
+              user,
+              localStorage.getItem("access_token") || "",
+              localStorage.getItem("refresh_token"),
+            )
+            scheduleProactiveRefresh()
+          } catch {
+            // token 已持久化，即使这里失败路由守卫/刷新也能恢复登录态
+          }
+          onLoginSuccess?.(msg.payload?.needOnboarding ?? false)
+        } else {
+          try {
+            const userData = await getCurrentUser()
+            setUser(normalizeUser(userData) as User)
+          } catch {
+            // 绑定结果以后端为准，调用方 invalidateQueries 会兜底刷新
+          }
+          onBindSuccess?.()
+        }
+        return
+      }
+
+      // 失败：弹窗内展示回调页透传的真实原因，提供刷新/整页跳转
+      setErrorMsg(msg.detail || "微信授权失败，请重试")
+      setStatus("error")
+    }
+
+    window.addEventListener("message", handleMessage)
+    return () => window.removeEventListener("message", handleMessage)
+  }, [open, scene, isLogin, onLoginSuccess, onBindSuccess, setAuth, setUser])
+
+  const handleRefresh = () => setRenderSeq((seq) => seq + 1)
+
+  const handleFullPageRedirect = () => {
+    if (authUrlRef.current) {
+      window.location.href = authUrlRef.current
+    }
+  }
+
+  return (
+    <Modal
+      title={isLogin ? "微信扫码登录" : "绑定微信"}
+      open={open}
+      onCancel={onClose}
+      footer={null}
+      width={380}
+      maskClosable={false}
+      destroyOnHidden
+    >
+      <div className="xx-wechat-qr-modal">
+        {/* 二维码容器常驻：WxLogin 在 loading 阶段就会把 iframe 渲染进来，
+            不能按 status 条件渲染，否则 effect 里永远找不到容器 */}
+        <div
+          id={CONTAINER_ID[scene]}
+          className="xx-wechat-qr-container"
+          style={{ visibility: status === "qrcode" ? "visible" : "hidden" }}
+        />
+
+        {status === "loading" && (
+          <div className="xx-wechat-qr-overlay">
+            <Spin size="large" />
+            <p>正在生成微信二维码...</p>
+          </div>
+        )}
+
+        {status === "qrcode" && (
+          <p className="xx-wechat-qr-tip">请使用微信扫描二维码{isLogin ? "登录" : "绑定账号"}</p>
+        )}
+
+        {status === "error" && (
+          <div className="xx-wechat-qr-overlay xx-wechat-qr-error">
+            <p className="xx-wechat-qr-error-msg">{errorMsg}</p>
+            <div className="xx-wechat-qr-error-actions">
+              <Button buttonType="primary" buttonSize="md" onClick={handleRefresh}>
+                刷新二维码
+              </Button>
+              {authUrlRef.current && (
+                <button
+                  type="button"
+                  className="xx-wechat-qr-fallback"
+                  onClick={handleFullPageRedirect}
+                >
+                  使用整页方式打开
+                </button>
+              )}
+            </div>
+          </div>
+        )}
+      </div>
+    </Modal>
+  )
+}
+
+export default WechatQrModal

apps/web/src/components/auth/WechatQrModal/messages.ts

@@ -0,0 +1,71 @@
+/**
+ * 微信扫码弹窗与 iframe 内回调页之间的 postMessage 消息协议
+ *
+ * 流程：弹窗内 WxLogin(self_redirect:true) 渲染的二维码 iframe 扫码后，
+ * 微信重定向到本站回调页（同源，在 iframe 内加载）；回调页完成换 token/绑定后，
+ * 通过 window.parent.postMessage 把结果通知弹窗，弹窗负责关闭/展示错误/同步登录态。
+ */
+
+/** 扫码场景：登录 / 绑定 */
+export type WechatQrScene = "login" | "bind"
+
+export interface WechatQrSuccessPayload {
+  /** 登录场景：是否需要昵称引导（新用户或资料未完善） */
+  needOnboarding?: boolean
+}
+
+export interface WechatQrMessageData {
+  /** 固定协议标识，父窗口只认该 source */
+  source: "xiaoxia-wechat-qr"
+  /** 场景，需与弹窗发起时一致（login/bind），父窗口据此过滤 */
+  scene: WechatQrScene
+  /** 成功 / 失败 */
+  success: boolean
+  /** 失败时的真实原因（已在回调页拼好，含后端 detail） */
+  detail?: string
+  payload?: WechatQrSuccessPayload
+}
+
+export const WECHAT_QR_MESSAGE_SOURCE = "xiaoxia-wechat-qr"
+
+/** 判断收到的 message 是否为本协议消息（且场景匹配） */
+export function isWechatQrMessage(
+  data: unknown,
+  scene: WechatQrScene,
+): data is WechatQrMessageData {
+  if (!data || typeof data !== "object") return false
+  const msg = data as Partial<WechatQrMessageData>
+  return msg.source === WECHAT_QR_MESSAGE_SOURCE && msg.scene === scene
+}
+
+/** 当前页面是否运行在 iframe（弹窗内嵌二维码）中 */
+export function isInIframe(): boolean {
+  try {
+    return window.parent !== window
+  } catch {
+    // 跨域访问 window.parent 可能抛异常，按非 iframe 处理
+    return false
+  }
+}
+
+/**
+ * iframe 内回调页向父窗口上报扫码结果。同源回调页加载，targetOrigin 限定本站 origin。
+ */
+export function postWechatQrResult(
+  scene: WechatQrScene,
+  success: boolean,
+  options?: { detail?: string; needOnboarding?: boolean },
+): void {
+  if (!isInIframe()) return
+  const data: WechatQrMessageData = {
+    source: WECHAT_QR_MESSAGE_SOURCE,
+    scene,
+    success,
+    detail: options?.detail,
+    payload:
+      success && options?.needOnboarding !== undefined
+        ? { needOnboarding: options.needOnboarding }
+        : undefined,
+  }
+  window.parent.postMessage(data, window.location.origin)
+}

apps/web/src/components/common/ChunkErrorBoundary.tsx

@@ -0,0 +1,85 @@
+/**
+ * 全局错误边界：专门兜底"发版后旧标签页懒加载 chunk 失效"导致的白屏，
+ * 同时兜住页面级渲染崩溃，避免任何未捕获错误导致整页白屏无反馈。
+ *
+ * 捕获到 ChunkLoadError / Failed to fetch dynamically imported module：
+ *  1. 首次：自动整页刷新一次（sessionStorage 标记，刷新后 index.html 重新拉取，
+ *     拿到新 chunk 引用，白屏自愈）
+ *  2. 刷新后仍失败（标记未过期）：不再自动刷新，显示"系统已更新，请点击刷新"
+ *     兜底界面，由用户手动点击
+ *
+ * 其他非 chunk 错误：显示通用错误页 + "返回首页"按钮（跳首页而非刷新当前 URL，
+ * 避免刷新后再次命中同一路由崩溃形成死循环）。
+ */
+import React from "react"
+import { Button, Result } from "antd"
+import {
+  getChunkReloadedAt,
+  goHomeRecover,
+  isChunkLoadError,
+  reloadForChunkError,
+} from "@/utils/chunkLoadError"
+
+interface Props {
+  children: React.ReactNode
+}
+
+interface State {
+  error: Error | null
+  isChunkError: boolean
+  /** 捕获错误时是否已经自动刷新过（决定显示自动刷新中还是手动兜底） */
+  alreadyReloaded: boolean
+}
+
+class ChunkErrorBoundary extends React.Component<Props, State> {
+  state: State = { error: null, isChunkError: false, alreadyReloaded: false }
+
+  static getDerivedStateFromError(error: Error): State {
+    const chunk = isChunkLoadError(error)
+    return {
+      error,
+      isChunkError: chunk,
+      alreadyReloaded: chunk ? getChunkReloadedAt() !== null : false,
+    }
+  }
+
+  componentDidCatch(error: Error): void {
+    // 仅 chunk 错误且本次会话没自动刷新过 → 打标记并整页刷新（自愈）
+    if (isChunkLoadError(error) && getChunkReloadedAt() === null) {
+      reloadForChunkError()
+    }
+  }
+
+  render(): React.ReactNode {
+    const { error, isChunkError, alreadyReloaded } = this.state
+    if (!error) return this.props.children
+
+    if (isChunkError && !alreadyReloaded) {
+      // 已打标记、componentDidCatch 里已触发 reload；极短瞬间展示加载中
+      return (
+        <Result status="info" title="系统正在更新" subTitle="检测到新版本，正在自动刷新页面…" />
+      )
+    }
+
+    // 手动兜底统一跳首页（整页导航）：chunk 失效时脱离旧 chunk 引用；
+    // 业务崩溃时绕开当前报错路由，避免刷新-再崩死循环
+    return (
+      <Result
+        status="warning"
+        title={isChunkError ? "系统已更新" : "页面出现异常"}
+        subTitle={
+          isChunkError
+            ? "检测到新版本，请点击下方按钮回到首页加载最新内容。"
+            : "页面加载遇到问题，点击返回首页通常可以恢复，未保存的内容可能丢失。"
+        }
+        extra={
+          <Button type="primary" onClick={goHomeRecover}>
+            {isChunkError ? "刷新并返回首页" : "返回首页"}
+          </Button>
+        }
+      />
+    )
+  }
+}
+
+export default ChunkErrorBoundary

apps/web/src/components/layout/MainLayout.css

@@ -5,7 +5,7 @@
      .xx-app-shell          全屏 flex 容器
      ├── header (xx-top-nav) 顶部导航（Header.tsx 管理）
      └── .xx-app-body       水平 flex 行
-         ├── .xx-app-sidebar  左侧侧边栏（240px / 64px 折叠）
+         ├── .xx-app-sidebar  左侧侧边栏（128px / 64px 折叠）
          └── .xx-app-content  主内容区（自适应）
 
    所有尺寸/颜色均使用 global.css 设计系统变量
@@ -31,7 +31,7 @@
 
 /* ── 侧边栏 ───────────────────────────────────────────────── */
 .xx-app-sidebar {
-  width: 240px;
+  width: 128px;
   flex-shrink: 0;
   position: sticky;
   top: 0;
@@ -103,6 +103,12 @@
   padding: var(--space-sm);
 }
 
+/* 展开态（侧边栏 128px）水平 padding 收窄，为菜单文字留出完整一行空间 */
+.xx-app-sidebar:not(.xx-collapsed) .xx-sidebar-content {
+  padding-left: var(--space-xs);
+  padding-right: var(--space-xs);
+}
+
 /* ── 主内容区 ─────────────────────────────────────────────── */
 .xx-app-content {
   flex: 1;
@@ -136,7 +142,7 @@
 
   /* 展开态恢复完整宽度 */
   .xx-app-sidebar:not(.xx-collapsed) {
-    width: 240px;
+    width: 128px;
   }
 
   .xx-app-sidebar:not(.xx-collapsed) .xx-sidebar-toggle {
@@ -159,7 +165,7 @@
     top: 56px; /* 移动端 Header 高度 */
     left: 0;
     bottom: 0;
-    width: 240px;
+    width: 128px;
     transform: translateX(-100%);
     transition: transform var(--transition-slow);
     box-shadow: none;

apps/web/src/components/layout/MainLayout.tsx

@@ -2,7 +2,7 @@
  * MainLayout - 主布局组件（Task 1.2）
  *
  * 三栏布局：左侧侧边栏 + 顶部导航栏 + 主内容区
- * - 侧边栏：240px 固定宽度，可折叠至 64px 图标栏
+ * - 侧边栏：128px 固定宽度，可折叠至 64px 图标栏
  * - 顶部导航：复用 Header 组件（68px 固定高度）
  * - 主内容区：自适应填充剩余空间
  * - 响应式：移动端（<768px）隐藏侧边栏

apps/web/src/components/layout/Sidebar.css

@@ -30,7 +30,7 @@
 
 /* 分组标题 */
 .xx-sidebar-group-title {
-  padding: var(--space-sm) var(--space-md) var(--space-xs);
+  padding: var(--space-sm) var(--space-sm) var(--space-xs);
   font-size: var(--font-size-xs);
   font-weight: var(--font-weight-semibold);
   color: var(--text-tertiary);
@@ -56,9 +56,9 @@
 .xx-sidebar-menu-item {
   display: flex;
   align-items: center;
-  gap: var(--space-sm);
-  padding: var(--space-sm) var(--space-md);
-  margin: 0 var(--space-xs);
+  gap: var(--space-xs);
+  padding: var(--space-sm) var(--space-xs);
+  margin: 0 var(--space-xxs);
   border-radius: var(--radius-sm);
   cursor: pointer;
   color: var(--text-secondary);
@@ -97,12 +97,12 @@
   align-items: center;
   justify-content: center;
   flex-shrink: 0;
-  width: 36px;
-  height: 36px;
-  border-radius: 10px;
+  width: 28px;
+  height: 28px;
+  border-radius: 8px;
   background: #f1f5f9;
   color: var(--text-secondary);
-  font-size: 18px;
+  font-size: 16px;
   line-height: 1;
   transition: 0.15s ease;
 }
@@ -119,7 +119,9 @@
 /* ── 菜单项文字 ───────────────────────────────────────────── */
 .xx-sidebar-menu-label {
   flex: 1;
+  min-width: 0;
   overflow: hidden;
+  white-space: nowrap;
   text-overflow: ellipsis;
 }
 
@@ -133,8 +135,16 @@
 /* 折叠时菜单项居中，仅图标 */
 .xx-sidebar-nav--collapsed .xx-sidebar-menu-item {
   justify-content: center;
-  padding: var(--space-sm);
-  margin: 0 var(--space-xxs);
+  padding: var(--space-xs);
+  margin: 0;
+}
+
+/* 折叠态图标恢复更大尺寸居中 */
+.xx-sidebar-nav--collapsed .xx-sidebar-menu-icon {
+  width: 32px;
+  height: 32px;
+  border-radius: 8px;
+  font-size: 16px;
 }
 
 /* 折叠时隐藏分组标题 */

apps/web/src/config/navigation.ts

@@ -18,6 +18,7 @@ import {
   ControlOutlined,
   CrownOutlined,
   UnorderedListOutlined,
+  UserOutlined,
 } from "@ant-design/icons"
 
 /** 导航项类型 */
@@ -57,6 +58,12 @@ export const NAV_ITEMS: NavItem[] = [
     path: "/app/titles",
     icon: React.createElement(FileTextOutlined),
   },
+  {
+    key: "scripts",
+    label: "文案库",
+    path: "/app/scripts",
+    icon: React.createElement(EditOutlined),
+  },
   {
     key: "voices",
     label: "配音库",
@@ -88,6 +95,12 @@ export const NAV_ITEMS: NavItem[] = [
     path: "/app/generate",
     icon: React.createElement(VideoCameraOutlined),
   },
+  {
+    key: "ai-avatar",
+    label: "AI数字人",
+    path: "/app/ai-avatar",
+    icon: React.createElement(UserOutlined),
+  },
   {
     key: "history",
     label: "任务历史",
@@ -131,6 +144,12 @@ export const NAV_GROUPS: NavGroup[] = [
         path: "/app/generate",
         icon: React.createElement(VideoCameraOutlined),
       },
+      {
+        key: "ai-avatar",
+        label: "AI数字人",
+        path: "/app/ai-avatar",
+        icon: React.createElement(UserOutlined),
+      },
       {
         key: "editing-planner",
         label: "剪辑模板",
@@ -160,6 +179,12 @@ export const NAV_GROUPS: NavGroup[] = [
         path: "/app/titles",
         icon: React.createElement(FileTextOutlined),
       },
+      {
+        key: "scripts",
+        label: "文案库",
+        path: "/app/scripts",
+        icon: React.createElement(EditOutlined),
+      },
       {
         key: "products",
         label: "成品库",

apps/web/src/main.tsx

@@ -9,6 +9,7 @@ import { QueryClient, QueryClientProvider } from "@tanstack/react-query"
 import { ConfigProvider, App as AntApp } from "antd"
 import zhCN from "antd/locale/zh_CN"
 import router from "./router"
+import ChunkErrorBoundary from "./components/common/ChunkErrorBoundary"
 import { scheduleProactiveRefresh } from "./api/auth/tokenRefresh"
 
 // 应用启动时，如果用户已登录，立即调度主动 token 刷新
@@ -99,7 +100,9 @@ ReactDOM.createRoot(document.getElementById("root")!).render(
     <QueryClientProvider client={queryClient}>
       <ConfigProvider locale={zhCN} theme={theme}>
         <AntApp>
-          <RouterProvider router={router} />
+          <ChunkErrorBoundary>
+            <RouterProvider router={router} />
+          </ChunkErrorBoundary>
         </AntApp>
       </ConfigProvider>
     </QueryClientProvider>

apps/web/src/pages/admin/Admin.css

@@ -11,7 +11,7 @@
 
 .admin-coming-soon-page {
   padding: 32px;
-  max-width: 1400px;
+  max-width: 1680px;
   margin: 0 auto;
 }
 

apps/web/src/pages/ai-avatar/AiAvatarPage.tsx

@@ -0,0 +1,522 @@
+/**
+ * AI数字人 — 主页面（v3）
+ * 5列水平面板布局
+ */
+import React, { useState, useCallback, useEffect, useRef } from "react"
+import { message } from "antd"
+import "./AiAvatar.css"
+import { useAiAvatar } from "./hooks/useAiAvatar"
+import { PanelVideoSelector } from "./components/PanelVideoSelector"
+import PanelVoiceSelector from "./components/PanelVoiceSelector"
+import PanelScriptAndLipsync from "./components/PanelScriptAndLipsync"
+import PanelTitleConfig from "./components/PanelTitleConfig"
+import PanelCoverAndGenerate from "./components/PanelCoverAndGenerate"
+import { ModalAssetPicker } from "./components/ModalAssetPicker"
+import ModalBRollEditor from "./components/ModalBRollEditor"
+import {
+  getScripts,
+  getAssetById,
+  createLipsyncJob,
+  getLipsyncJob,
+  submitRender,
+  generateSmartCover,
+} from "./api/aiAvatar"
+import {
+  normalizeEmotion,
+  buildTitleConfigPayload,
+  buildCoverConfigPayload,
+} from "./utils/contract"
+
+/** 面板折叠状态 */
+type PanelKey = "video" | "voice" | "script" | "title" | "cover"
+
+const AiAvatarPage: React.FC = () => {
+  const state = useAiAvatar()
+  const [collapsed, setCollapsed] = useState<Record<PanelKey, boolean>>({
+    video: false,
+    voice: false,
+    script: false,
+    title: false,
+    cover: false,
+  })
+
+  /* ── 对口型生成弹窗 ── */
+  const [showLipsyncModal, setShowLipsyncModal] = useState(false)
+  const [lipsyncStatus, setLipsyncStatus] = useState<"generating" | "completed" | "failed">(
+    "generating",
+  )
+  const [lipsyncErrorMessage, setLipsyncErrorMessage] = useState("")
+  /* ── 智能封面加载态 ── */
+  const [smartCoverLoading, setSmartCoverLoading] = useState(false)
+
+  /* ── 对口型轮询 ── */
+  const lipsyncTimerRef = useRef<ReturnType<typeof setInterval> | null>(null)
+
+  const togglePanel = useCallback((key: PanelKey) => {
+    setCollapsed((prev) => ({ ...prev, [key]: !prev[key] }))
+  }, [])
+
+  /* ── 对口型 ── */
+  const handleGenerateLipsync = useCallback(async () => {
+    // ② 缺项明确提示（#1809）：不再静默 return
+    const video = state.selectedVideo
+    const voice = state.selectedVoice
+    const text = state.scriptText.trim()
+    const missing: string[] = []
+    if (!video) missing.push("出镜视频")
+    if (!voice) missing.push("音色")
+    if (!text) missing.push("文案")
+    if (missing.length > 0 || !video || !voice) {
+      message.warning(`请先选择${missing.join("、")}`)
+      return
+    }
+    try {
+      // 显示生成弹窗
+      setShowLipsyncModal(true)
+      setLipsyncStatus("generating")
+      setLipsyncErrorMessage("")
+
+      // ① 先按素材 id 拿 file_url（#1809 补充：对齐后端新参数 video_url）
+      console.log("[对口型] 开始生成:", {
+        videoId: video.id,
+        voiceId: voice.voice_id,
+        voiceType: voice.type,
+        textLen: state.scriptText.length,
+      })
+      const asset = await getAssetById(video.id)
+      console.log("[对口型] getAssetById 响应:", {
+        id: asset?.id,
+        file_url: asset?.file_url?.substring(0, 100),
+      })
+      const videoUrl = asset?.file_url
+      if (!videoUrl) {
+        console.error("[对口型] file_url 为空，asset:", asset)
+        setShowLipsyncModal(false)
+        message.error("获取出镜视频播放地址失败，请重新选择素材")
+        return
+      }
+      // ② 模式A TTS直生：video_url + voice_id + script_text，语速/情绪英文枚举透传（#1822）
+      const payload = {
+        voice_id: voice.voice_id,
+        script_text: state.scriptText,
+        video_url: videoUrl,
+        speed: state.speed, // 语速 0.5~2.0
+        emotion: normalizeEmotion(state.emotion), // natural/excited/calm/friendly
+      }
+      console.log("[对口型] createLipsyncJob 请求:", payload)
+      const job = await createLipsyncJob(payload)
+      console.log("[对口型] createLipsyncJob 响应:", { id: job.id, status: job.status })
+      state.setLipsyncJob(job)
+      // 开始轮询
+      if (lipsyncTimerRef.current) clearInterval(lipsyncTimerRef.current)
+      lipsyncTimerRef.current = setInterval(async () => {
+        try {
+          const updated = await getLipsyncJob(job.id)
+          state.setLipsyncJob(updated)
+          console.log("[对口型] 轮询状态:", {
+            id: updated.id,
+            status: updated.status,
+            error: updated.error_message,
+          })
+          if (updated.status === "completed") {
+            if (lipsyncTimerRef.current) clearInterval(lipsyncTimerRef.current)
+            setLipsyncStatus("completed")
+            setTimeout(() => {
+              setShowLipsyncModal(false)
+              message.success("对口型视频生成完成")
+            }, 1000)
+          } else if (updated.status === "failed") {
+            if (lipsyncTimerRef.current) clearInterval(lipsyncTimerRef.current)
+            setLipsyncStatus("failed")
+            setLipsyncErrorMessage(updated.error_message || "对口型生成失败")
+          }
+        } catch (err) {
+          console.error("[对口型] 轮询错误:", err)
+        }
+      }, 3000)
+    } catch (err) {
+      console.error("[对口型] 创建失败:", {
+        status: (err as { response?: { status?: number } })?.response?.status,
+        data: (err as { response?: { data?: unknown } })?.response?.data,
+        message: err instanceof Error ? err.message : String(err),
+      })
+      setShowLipsyncModal(false)
+      message.error(err instanceof Error ? err.message : "对口型任务提交失败，请重试")
+    }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [state.selectedVideo, state.selectedVoice, state.scriptText, state.speed, state.emotion])
+
+  // 取消对口型生成
+  const handleCancelLipsync = useCallback(() => {
+    if (lipsyncTimerRef.current) {
+      clearInterval(lipsyncTimerRef.current)
+      lipsyncTimerRef.current = null
+    }
+    setShowLipsyncModal(false)
+    setLipsyncStatus("generating")
+    setLipsyncErrorMessage("")
+  }, [])
+
+  // 清理轮询
+  useEffect(() => {
+    return () => {
+      if (lipsyncTimerRef.current) clearInterval(lipsyncTimerRef.current)
+    }
+  }, [])
+
+  /* ── 生成视频 ── */
+  const handleGenerate = useCallback(async () => {
+    // ② 前置条件提示（#1809）
+    if (!state.lipsyncJob || state.lipsyncJob.status !== "completed") {
+      message.warning("请先生成对口型视频，待对口型完成后再提交渲染")
+      return
+    }
+    state.setIsGenerating(true)
+    try {
+      await submitRender({
+        lipsync_job_id: state.lipsyncJob.id,
+        script_id: state.script?.id,
+        b_roll_segments: state.bRollSegments as never,
+        // 字段映射：build_title_drawtext_filter 真实口径 text/font_size/font_color/position/...
+        title_config: buildTitleConfigPayload(state.titleConfig),
+        // cover_config：智能封面 cover_url + 截帧 timestamp
+        cover_config: buildCoverConfigPayload(state.coverConfig, state.coverConfig.smart_cover_url),
+        resolution: state.resolution,
+      })
+      message.success("渲染任务已提交，可在视频管理中查看进度")
+    } catch (err) {
+      console.error("渲染任务提交失败:", err)
+      message.error(err instanceof Error ? err.message : "渲染任务提交失败，请重试")
+    } finally {
+      state.setIsGenerating(false)
+    }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [
+    state.lipsyncJob,
+    state.script,
+    state.bRollSegments,
+    state.titleConfig,
+    state.coverConfig,
+    state.resolution,
+  ])
+
+  /* ── 智能封面：调后端 MediaKit 选帧接口（#1822） ── */
+  const handleSmartCover = useCallback(async () => {
+    // 基于对口型成片抽帧，必须先完成对口型
+    const videoUrl = state.lipsyncJob?.output_video_url
+    if (state.lipsyncJob?.status !== "completed" || !videoUrl) {
+      message.warning("请先生成对口型视频，完成后再智能获取封面")
+      return
+    }
+    setSmartCoverLoading(true)
+    try {
+      const res = await generateSmartCover(videoUrl, 5)
+      if (res.cover_url) {
+        state.setCoverConfig((prev) => ({
+          ...prev,
+          mode: "auto_frame",
+          smart_cover_url: res.cover_url,
+          thumbnail_url: res.cover_url,
+        }))
+        message.success("智能封面已生成")
+      } else {
+        message.error(res.message || "智能封面生成失败，请稍后重试")
+      }
+    } catch (err) {
+      console.error("智能封面生成失败:", err)
+      message.error(err instanceof Error ? err.message : "智能封面生成失败，请重试")
+    } finally {
+      setSmartCoverLoading(false)
+    }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [state.lipsyncJob])
+
+  /* ── 配置汇总 ── */
+  const summary = {
+    videoName: state.selectedVideo?.name || null,
+    voiceName: state.selectedVoice?.name || null,
+    scriptLength: state.scriptText.length,
+    lipsyncStatus: state.lipsyncJob?.status || null,
+    brollCount: state.bRollSegments.length,
+    hasTitle: state.titleConfig.title.length > 0,
+    hasCover: state.coverConfig.enabled,
+  }
+
+  return (
+    <div className="aa-page">
+      <div className="aa-page-header">
+        <h1>AI数字人</h1>
+      </div>
+      <div className="aa-page-body">
+        {/* 面板1：出镜视频 */}
+        <div className={`aa-panel aa-panel--p1${collapsed.video ? " collapsed" : ""}`}>
+          <div className="aa-panel__header" onClick={() => togglePanel("video")}>
+            <span className="aa-panel__title">出镜视频</span>
+            <span className="aa-panel__toggle">▼</span>
+          </div>
+          <div className="aa-panel__body">
+            <PanelVideoSelector
+              selectedVideo={state.selectedVideo}
+              onSelectVideo={() => state.setShowAssetPicker(true)}
+              onRemoveVideo={state.removeVideo}
+              titleConfig={state.titleConfig}
+            />
+          </div>
+        </div>
+
+        {/* 面板2：配音库 */}
+        <div className={`aa-panel aa-panel--p2${collapsed.voice ? " collapsed" : ""}`}>
+          <div className="aa-panel__header" onClick={() => togglePanel("voice")}>
+            <span className="aa-panel__title">配音库</span>
+            <span className="aa-panel__toggle">▼</span>
+          </div>
+          <div className="aa-panel__body">
+            <PanelVoiceSelector
+              voiceSource={state.voiceSource}
+              onVoiceSourceChange={state.setVoiceSource}
+              selectedVoice={state.selectedVoice}
+              onSelectVoice={state.setSelectedVoice}
+              emotion={state.emotion}
+              onEmotionChange={state.setEmotion}
+              speed={state.speed}
+              onSpeedChange={state.setSpeed}
+              language={state.language}
+              onLanguageChange={state.setLanguage}
+            />
+          </div>
+        </div>
+
+        {/* 面板3：文案 & 对口型 */}
+        <div className={`aa-panel aa-panel--p3${collapsed.script ? " collapsed" : ""}`}>
+          <div className="aa-panel__header" onClick={() => togglePanel("script")}>
+            <span className="aa-panel__title">文案 & 对口型</span>
+            <span className="aa-panel__toggle">▼</span>
+          </div>
+          <div className="aa-panel__body">
+            <PanelScriptAndLipsync
+              scriptText={state.scriptText}
+              onScriptTextChange={state.setScriptText}
+              onOpenScriptModal={() => state.setShowScriptModal(true)}
+              lipsyncJob={state.lipsyncJob}
+              onGenerateLipsync={handleGenerateLipsync}
+              bRollSegments={state.bRollSegments}
+              onOpenBRollModal={() => state.setShowBRollModal(true)}
+              onRemoveBRoll={state.removeBRollSegment}
+            />
+          </div>
+        </div>
+
+        {/* 面板4：标题配置 */}
+        <div className={`aa-panel aa-panel--p4${collapsed.title ? " collapsed" : ""}`}>
+          <div className="aa-panel__header" onClick={() => togglePanel("title")}>
+            <span className="aa-panel__title">标题配置</span>
+            <span className="aa-panel__toggle">▼</span>
+          </div>
+          <div className="aa-panel__body">
+            <PanelTitleConfig titleConfig={state.titleConfig} onUpdate={state.updateTitleConfig} />
+          </div>
+        </div>
+
+        {/* 面板5：封面 & 生成 */}
+        <div className={`aa-panel aa-panel--p5${collapsed.cover ? " collapsed" : ""}`}>
+          <div className="aa-panel__header" onClick={() => togglePanel("cover")}>
+            <span className="aa-panel__title">封面 & 生成</span>
+            <span className="aa-panel__toggle">▼</span>
+          </div>
+          <div className="aa-panel__body">
+            <PanelCoverAndGenerate
+              coverConfig={state.coverConfig}
+              onCoverConfigChange={(partial) =>
+                state.setCoverConfig((prev) => ({ ...prev, ...partial }))
+              }
+              onSmartCover={handleSmartCover}
+              smartCoverLoading={smartCoverLoading}
+              canSmartCover={state.lipsyncJob?.status === "completed"}
+              resolution={state.resolution}
+              onResolutionChange={state.setResolution}
+              isGenerating={state.isGenerating}
+              onGenerate={handleGenerate}
+              summary={summary}
+            />
+          </div>
+        </div>
+      </div>
+
+      {/* 素材库弹窗 */}
+      {state.showAssetPicker && (
+        <ModalAssetPicker
+          open={state.showAssetPicker}
+          onClose={() => state.setShowAssetPicker(false)}
+          onSelect={state.selectVideo}
+          selectedId={state.selectedVideo?.id}
+        />
+      )}
+
+      {/* 文案选择弹窗 */}
+      {state.showScriptModal && (
+        <ScriptSelectModalLazy
+          open={state.showScriptModal}
+          onClose={() => state.setShowScriptModal(false)}
+          onSelect={state.selectScript}
+        />
+      )}
+
+      {/* B-roll 编辑器弹窗 */}
+      {state.showBRollModal && (
+        <ModalBRollEditor
+          open={state.showBRollModal}
+          onClose={() => state.setShowBRollModal(false)}
+          existingSegments={state.bRollSegments}
+          scriptText={state.scriptText}
+          outputDuration={state.lipsyncJob?.output_duration ?? 0}
+          onConfirm={state.addBRollSegment}
+          onRemove={state.removeBRollSegment}
+        />
+      )}
+
+      {/* 对口型生成弹窗 */}
+      {showLipsyncModal && (
+        <div className="aa-modal-overlay">
+          <div className="aa-modal" onClick={(e) => e.stopPropagation()}>
+            <div className="aa-modal__header">
+              <span className="aa-modal__title">对口型生成</span>
+              <button className="aa-modal__close" onClick={handleCancelLipsync}>
+                ✕
+              </button>
+            </div>
+            <div
+              className="aa-modal__body"
+              style={{
+                display: "flex",
+                flexDirection: "column",
+                alignItems: "center",
+                padding: "40px 20px",
+              }}
+            >
+              {lipsyncStatus === "generating" && (
+                <>
+                  <div className="aa-lipsync-spinner" />
+                  <div style={{ marginTop: 20, fontSize: 15, color: "#1a1a2e" }}>
+                    对口型视频生成中…
+                  </div>
+                  <div style={{ marginTop: 8, fontSize: 13, color: "#8c8ca1" }}>
+                    请勿关闭页面，完成后将自动提示
+                  </div>
+                </>
+              )}
+              {lipsyncStatus === "completed" && (
+                <>
+                  <div style={{ fontSize: 48 }}>✅</div>
+                  <div style={{ marginTop: 16, fontSize: 15, color: "#1a1a2e" }}>
+                    对口型视频生成完成
+                  </div>
+                </>
+              )}
+              {lipsyncStatus === "failed" && (
+                <>
+                  <div style={{ fontSize: 48 }}>❌</div>
+                  <div style={{ marginTop: 16, fontSize: 15, color: "#1a1a2e" }}>
+                    对口型生成失败
+                  </div>
+                  {lipsyncErrorMessage && (
+                    <div style={{ marginTop: 8, fontSize: 13, color: "#ff4d4f" }}>
+                      {lipsyncErrorMessage}
+                    </div>
+                  )}
+                </>
+              )}
+            </div>
+            <div className="aa-modal__footer">
+              {lipsyncStatus === "generating" && (
+                <button className="aa-btn aa-btn--danger" onClick={handleCancelLipsync}>
+                  取消生成
+                </button>
+              )}
+              {lipsyncStatus !== "generating" && (
+                <button className="aa-btn" onClick={handleCancelLipsync}>
+                  关闭
+                </button>
+              )}
+            </div>
+          </div>
+        </div>
+      )}
+    </div>
+  )
+}
+
+/** 文案选择弹窗（内联实现，轻量版） */
+const ScriptSelectModalLazy: React.FC<{
+  open: boolean
+  onClose: () => void
+  onSelect: (script: import("./types").Script) => void
+}> = ({ open, onClose, onSelect }) => {
+  const [scripts, setScripts] = useState<import("./types").Script[]>([])
+  const [search, setSearch] = useState("")
+  const [loading, setLoading] = useState(false)
+
+  useEffect(() => {
+    if (!open) return
+    setLoading(true)
+    getScripts()
+      .then((items) => setScripts(Array.isArray(items) ? items : []))
+      .catch(() => setScripts([]))
+      .finally(() => setLoading(false))
+  }, [open])
+
+  const filtered = scripts.filter(
+    (s) => !search || s.title.includes(search) || s.content.includes(search),
+  )
+
+  return (
+    <div className="aa-modal-overlay" onClick={onClose}>
+      <div className="aa-modal" onClick={(e) => e.stopPropagation()}>
+        <div className="aa-modal__header">
+          <span className="aa-modal__title">选择文案</span>
+          <button className="aa-modal__close" onClick={onClose}>
+            ✕
+          </button>
+        </div>
+        <div className="aa-modal__body">
+          <div className="aa-script-list-header">
+            <input
+              className="aa-input"
+              placeholder="搜索文案..."
+              value={search}
+              onChange={(e) => setSearch(e.target.value)}
+            />
+          </div>
+          {loading ? (
+            <div className="aa-empty">加载中...</div>
+          ) : filtered.length === 0 ? (
+            <div className="aa-empty">
+              <div className="aa-empty__icon">📝</div>
+              暂无文案，请手动输入或新建
+            </div>
+          ) : (
+            <div className="aa-script-list">
+              {filtered.map((s) => (
+                <div key={s.id} className="aa-script-item" onClick={() => onSelect(s)}>
+                  <span className="aa-script-item__icon">📄</span>
+                  <div className="aa-script-item__info">
+                    <div className="aa-script-item__title">{s.title}</div>
+                    <div className="aa-script-item__meta">
+                      {s.char_count}字 · {new Date(s.created_at).toLocaleDateString()}
+                    </div>
+                  </div>
+                </div>
+              ))}
+            </div>
+          )}
+        </div>
+        <div className="aa-modal__footer">
+          <button className="aa-btn" onClick={onClose}>
+            取消
+          </button>
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default AiAvatarPage

apps/web/src/pages/ai-avatar/api/aiAvatar.ts

@@ -0,0 +1,94 @@
+/**
+ * AI数字人 — API 封装（#1822 契约对齐）
+ */
+import apiClient from "@/api/client"
+import type { Script, LipsyncJob, RenderJob, BRollSegment } from "../types"
+
+/* ── 文案库 ── */
+export const getScripts = async (): Promise<Script[]> => {
+  const response = await apiClient.get<{ items?: Script[] } | Script[]>("/scripts")
+  // 后端列表返回 { items, total } 分页对象，做兼容解包 + 数组防御（#1809 白屏修复）
+  const data = response.data as unknown
+  if (Array.isArray(data)) return data
+  const items = (data as { items?: Script[] })?.items
+  return Array.isArray(items) ? items : []
+}
+
+export const getScriptById = async (id: string): Promise<Script> => {
+  const response = await apiClient.get<Script>(`/scripts/${id}`)
+  return response.data
+}
+
+export const createScript = async (data: { title: string; content: string }): Promise<Script> => {
+  const response = await apiClient.post<Script>("/scripts", data)
+  return response.data
+}
+
+export const deleteScript = async (id: string): Promise<void> => {
+  await apiClient.delete(`/scripts/${id}`)
+}
+
+/* ── 素材单查（拿到 file_url 作为对口型的 video_url） ── */
+export const getAssetById = async (id: string): Promise<{ file_url?: string; id: string }> => {
+  const response = await apiClient.get<{ file_url?: string; id: string }>(`/assets/${id}`)
+  return response.data
+}
+
+/* ── 对口型（模式A：TTS 直生，后端内部合成音频；不要先调 TTS 拿 audio_url） ── */
+export const createLipsyncJob = async (data: {
+  /** 人物视频 URL（MP4）；由素材 id 经 getAssetById 拿 file_url，禁止传 video_asset_id */
+  video_url: string
+  /** 音色 ID（预置音色 或 克隆音色 profile UUID，后端会解析） */
+  voice_id: string
+  /** 要合成的文案（手动输入或文案库内容） */
+  script_text: string
+  /** 语速 0.5~2.0，默认 1.0 */
+  speed?: number
+  /** 情绪英文枚举：natural/excited/calm/friendly */
+  emotion?: string
+  enable_video_loop?: boolean
+  project_id?: string
+}): Promise<LipsyncJob> => {
+  const response = await apiClient.post<LipsyncJob>("/lipsync/jobs", data)
+  return response.data
+}
+
+export const getLipsyncJob = async (id: string): Promise<LipsyncJob> => {
+  const response = await apiClient.get<LipsyncJob>(`/lipsync/jobs/${id}`)
+  return response.data
+}
+
+/* ── 智能封面（MediaKit 抽帧 + 质量评分选最佳帧，独立于渲染任务） ── */
+export const generateSmartCover = async (
+  video_url: string,
+  max_frames = 5,
+): Promise<{ cover_url: string; status: string; message: string }> => {
+  const response = await apiClient.post<{ cover_url: string; status: string; message: string }>(
+    "/ai-avatar/render/smart-cover",
+    { video_url, max_frames },
+  )
+  return response.data
+}
+
+/* ── 渲染 ── */
+export const submitRender = async (data: {
+  lipsync_job_id: string
+  script_id?: string
+  b_roll_segments?: BRollSegment[]
+  title_config?: Record<string, unknown>
+  cover_config?: Record<string, unknown>
+  project_id?: string
+  resolution?: string
+}): Promise<RenderJob> => {
+  const response = await apiClient.post<RenderJob>("/ai-avatar/render", data)
+  return response.data
+}
+
+export const getRenderJob = async (jobId: string): Promise<RenderJob> => {
+  const response = await apiClient.get<RenderJob>(`/ai-avatar/render/${jobId}`)
+  return response.data
+}
+
+export const cancelRenderJob = async (jobId: string): Promise<void> => {
+  await apiClient.post(`/ai-avatar/render/${jobId}/cancel`)
+}

apps/web/src/pages/ai-avatar/components/ModalAssetPicker.tsx

@@ -0,0 +1,209 @@
+/**
+ * AI数字人 — 出镜视频选择弹窗（#1809 ③）
+ * 交互对齐智能剪辑 Step2：先选素材库（video 库）→ 再选该库内视频。
+ * 搜索框 + 素材库下拉 + 竖屏 9:16 视频缩略图网格 + 底部确认选择。
+ */
+import { useEffect, useState } from "react"
+import { getAssets, getAssetLibraries, type AssetItem, type AssetLibraryItem } from "@/api/assets"
+
+export interface ModalAssetPickerProps {
+  open: boolean
+  onClose: () => void
+  onSelect: (asset: AssetItem) => void
+  /** 已选中的素材 ID（用于高亮） */
+  selectedId?: string
+}
+
+export function ModalAssetPicker({ open, onClose, onSelect, selectedId }: ModalAssetPickerProps) {
+  const [keyword, setKeyword] = useState("")
+  const [libraries, setLibraries] = useState<AssetLibraryItem[]>([])
+  const [libraryId, setLibraryId] = useState<string>("")
+  const [assets, setAssets] = useState<AssetItem[]>([])
+  const [pickedId, setPickedId] = useState<string | null>(null)
+  const [loadingLibs, setLoadingLibs] = useState(false)
+  const [loadingAssets, setLoadingAssets] = useState(false)
+  const [error, setError] = useState("")
+
+  /* 弹窗打开：重置状态 */
+  useEffect(() => {
+    if (!open) return
+    setKeyword("")
+    setLibraries([])
+    setLibraryId("")
+    setAssets([])
+    setError("")
+    setPickedId(selectedId ?? null)
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [open])
+
+  /* 第一步：加载视频素材库列表（仅 kind=video，对齐智能剪辑 #1777） */
+  useEffect(() => {
+    if (!open) return
+    let cancelled = false
+    setLoadingLibs(true)
+    getAssetLibraries("video")
+      .then((libs) => {
+        if (cancelled) return
+        const list = Array.isArray(libs) ? libs : []
+        setLibraries(list)
+        // 默认选中第一个视频库
+        if (list.length > 0) setLibraryId((prev) => prev || list[0].id)
+      })
+      .catch(() => {
+        if (!cancelled) setError("素材库加载失败，请重试")
+      })
+      .finally(() => {
+        if (!cancelled) setLoadingLibs(false)
+      })
+    return () => {
+      cancelled = true
+    }
+  }, [open])
+
+  /* 第二步：选中库后拉取该库视频素材（关键字防抖） */
+  useEffect(() => {
+    if (!open || !libraryId) return
+    let cancelled = false
+    setLoadingAssets(true)
+    const load = async () => {
+      try {
+        // getAssets 返回 { items, total }；拉满一页（数字人口播视频库通常不大）
+        const { items } = await getAssets(libraryId, { page_size: 100 })
+        if (cancelled) return
+        let list = Array.isArray(items) ? items : []
+        // 仅保留视频素材（出镜视频要求）
+        list = list.filter((a) => a.mime_type?.includes("video"))
+        const kw = keyword.trim()
+        if (kw) list = list.filter((a) => a.name?.includes(kw))
+        setAssets(list)
+        setError("")
+      } catch {
+        if (!cancelled) {
+          setError("素材加载失败，请重试")
+          setAssets([])
+        }
+      } finally {
+        if (!cancelled) setLoadingAssets(false)
+      }
+    }
+    const timer = window.setTimeout(load, 300)
+    return () => {
+      cancelled = true
+      window.clearTimeout(timer)
+    }
+  }, [open, libraryId, keyword])
+
+  if (!open) return null
+
+  const handleConfirm = () => {
+    if (!pickedId) return
+    const asset = assets.find((a) => a.id === pickedId)
+    if (asset) onSelect(asset)
+    onClose()
+  }
+
+  return (
+    <div className="aa-modal-overlay" onClick={onClose}>
+      <div className="aa-modal" onClick={(e) => e.stopPropagation()}>
+        {/* 头部 */}
+        <div className="aa-modal__header">
+          <span className="aa-modal__title">选择出镜视频</span>
+          <button type="button" className="aa-modal__close" onClick={onClose} aria-label="关闭">
+            ×
+          </button>
+        </div>
+
+        {/* 主体：素材库选择 + 搜索 + 网格 */}
+        <div className="aa-modal__body">
+          {/* 第一步：选素材库 */}
+          <div className="aa-asset-search">
+            <select
+              className="aa-select"
+              style={{ width: 160, flex: "0 0 auto" }}
+              value={libraryId}
+              onChange={(e) => setLibraryId(e.target.value)}
+              disabled={loadingLibs || libraries.length === 0}
+            >
+              {libraries.length === 0 ? (
+                <option value="">{loadingLibs ? "素材库加载中…" : "暂无视频素材库"}</option>
+              ) : (
+                libraries.map((lib) => (
+                  <option key={lib.id} value={lib.id}>
+                    📁 {lib.name}
+                  </option>
+                ))
+              )}
+            </select>
+            <input
+              className="aa-input"
+              type="text"
+              placeholder="搜索素材名称…"
+              value={keyword}
+              onChange={(e) => setKeyword(e.target.value)}
+            />
+          </div>
+
+          {libraries.length === 0 && !loadingLibs ? (
+            <div className="aa-empty">
+              <div className="aa-empty__icon">📁</div>
+              暂无视频素材库，请先在「素材库」中创建视频库并上传视频
+            </div>
+          ) : loadingAssets ? (
+            <div className="aa-empty">
+              <div className="aa-empty__icon">⏳</div>
+              素材加载中…
+            </div>
+          ) : error ? (
+            <div className="aa-empty">
+              <div className="aa-empty__icon">⚠️</div>
+              {error}
+            </div>
+          ) : assets.length === 0 ? (
+            <div className="aa-empty">
+              <div className="aa-empty__icon">🎬</div>
+              该素材库暂无视频素材
+            </div>
+          ) : (
+            <div className="aa-asset-grid">
+              {assets.map((asset) => {
+                const isActive = asset.id === pickedId
+                const thumb = asset.thumbnail_url || asset.file_url
+                const isVideo = asset.mime_type?.includes("video")
+                return (
+                  <div
+                    key={asset.id}
+                    className={`aa-asset-card${isActive ? " selected" : ""}`}
+                    onClick={() => setPickedId(asset.id)}
+                  >
+                    {isVideo && !asset.thumbnail_url ? (
+                      <video src={asset.file_url} muted preload="metadata" />
+                    ) : (
+                      <img src={thumb} alt={asset.name} />
+                    )}
+                    {isActive && <div className="aa-asset-card__check">✓</div>}
+                    <div className="aa-asset-card__name">{asset.name}</div>
+                  </div>
+                )
+              })}
+            </div>
+          )}
+        </div>
+
+        {/* 底部：取消 + 确认选择 */}
+        <div className="aa-modal__footer">
+          <button type="button" className="aa-btn aa-btn--ghost" onClick={onClose}>
+            取消
+          </button>
+          <button
+            type="button"
+            className="aa-btn aa-btn--primary"
+            onClick={handleConfirm}
+            disabled={!pickedId}
+          >
+            确认选择
+          </button>
+        </div>
+      </div>
+    </div>
+  )
+}

apps/web/src/pages/ai-avatar/components/ModalBRollEditor.tsx

@@ -0,0 +1,421 @@
+/**
+ * AI数字人 — B-roll 画面插入编辑器弹窗（#1809 ④⑤⑥）
+ *
+ * 布局：
+ * - 左侧：先选素材库（video 库）→ 再选该库视频素材（已被其他 segment 使用的素材
+ *   标灰 + "已选择" 遮罩，pointer-events:none 防重复选择）
+ * - 右侧：文案句子列表（点选对应段落，替代原数字索引框）/ 全屏 or 画中画 / 四角位置+大小
+ *   （开始/结束时间已删除，按句子字数占比 × 口播总时长自动估算）
+ * - 底部：已配置的画面插入列表（可删除）
+ */
+import React, { useEffect, useMemo, useState } from "react"
+import { getAssets, getAssetLibraries, type AssetItem, type AssetLibraryItem } from "@/api/assets"
+import type { BRollSegment, BRollInsertMode, PipPosition } from "../types"
+import { splitScriptIntoSentences, type ScriptSentence } from "../utils/sentences"
+
+interface ModalBRollEditorProps {
+  open: boolean
+  onClose: () => void
+  /** 当前已有的 B-roll segments（用于标灰已选素材） */
+  existingSegments: BRollSegment[]
+  /** 当前文案全文（用于分句） */
+  scriptText: string
+  /** 对口型成片总时长（秒），用于时间自动估算 */
+  outputDuration: number
+  onConfirm: (segment: BRollSegment) => void
+  onRemove: (id: string) => void
+}
+
+const PIP_POSITION_OPTIONS: { value: PipPosition; label: string }[] = [
+  { value: "top-left", label: "左上" },
+  { value: "top-right", label: "右上" },
+  { value: "bottom-left", label: "左下" },
+  { value: "bottom-right", label: "右下" },
+]
+
+const MODE_LABEL: Record<BRollInsertMode, string> = {
+  fullscreen: "全屏切换",
+  pip: "画中画",
+}
+
+const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
+  open,
+  onClose,
+  existingSegments,
+  scriptText,
+  outputDuration,
+  onConfirm,
+  onRemove,
+}) => {
+  /* ── 素材库（④ 先选库再选素材） ── */
+  const [libraries, setLibraries] = useState<AssetLibraryItem[]>([])
+  const [libraryId, setLibraryId] = useState<string>("")
+  const [availableAssets, setAvailableAssets] = useState<AssetItem[]>([])
+  const [loadingLibs, setLoadingLibs] = useState(false)
+  const [loadingAssets, setLoadingAssets] = useState(false)
+  const [assetError, setAssetError] = useState("")
+
+  /* ── 右侧设置本地状态 ── */
+  const [selectedAsset, setSelectedAsset] = useState<AssetItem | null>(null)
+  const [selectedSentence, setSelectedSentence] = useState<ScriptSentence | null>(null)
+  const [mode, setMode] = useState<BRollInsertMode>("fullscreen")
+  const [pipPosition, setPipPosition] = useState<PipPosition>("top-right")
+  const [pipScale, setPipScale] = useState(0.3)
+
+  /** 文案分句（⑤） */
+  const sentences = useMemo(
+    () => splitScriptIntoSentences(scriptText, outputDuration),
+    [scriptText, outputDuration],
+  )
+
+  /** 已被现有 segments 占用的素材 id 集合（标灰、禁止重复选择） */
+  const usedAssetIds = useMemo(
+    () => new Set(existingSegments.map((seg) => seg.asset.id)),
+    [existingSegments],
+  )
+
+  /* 弹窗打开：重置选择 + 加载视频库列表 */
+  useEffect(() => {
+    if (!open) return
+    setSelectedAsset(null)
+    setSelectedSentence(null)
+    setMode("fullscreen")
+    setPipPosition("top-right")
+    setPipScale(0.3)
+    setLibraries([])
+    setLibraryId("")
+    setAvailableAssets([])
+    setAssetError("")
+    setLoadingLibs(true)
+    let cancelled = false
+    getAssetLibraries("video")
+      .then((libs) => {
+        if (cancelled) return
+        const list = Array.isArray(libs) ? libs : []
+        setLibraries(list)
+        if (list.length > 0) setLibraryId(list[0].id)
+      })
+      .catch(() => {
+        if (!cancelled) setAssetError("素材库加载失败，请重试")
+      })
+      .finally(() => {
+        if (!cancelled) setLoadingLibs(false)
+      })
+    return () => {
+      cancelled = true
+    }
+  }, [open])
+
+  /* 选中库后拉取该库视频素材 */
+  useEffect(() => {
+    if (!open || !libraryId) return
+    let cancelled = false
+    setLoadingAssets(true)
+    getAssets(libraryId, { page_size: 100 })
+      .then(({ items }) => {
+        if (cancelled) return
+        const list = (Array.isArray(items) ? items : []).filter((a) =>
+          a.mime_type?.includes("video"),
+        )
+        setAvailableAssets(list)
+        setAssetError("")
+      })
+      .catch(() => {
+        if (!cancelled) {
+          setAssetError("素材加载失败，请重试")
+          setAvailableAssets([])
+        }
+      })
+      .finally(() => {
+        if (!cancelled) setLoadingAssets(false)
+      })
+    return () => {
+      cancelled = true
+    }
+  }, [open, libraryId])
+
+  if (!open) return null
+
+  /** 选择素材（已选素材因 pointer-events:none 不会触发） */
+  const handleSelectAsset = (asset: AssetItem) => {
+    if (usedAssetIds.has(asset.id)) return
+    setSelectedAsset(asset)
+  }
+
+  /** 确认添加一段 B-roll（⑥ 时间取所选句子的估算起止） */
+  const handleConfirm = () => {
+    if (!selectedAsset || !selectedSentence) return
+    const startTime = selectedSentence.startTime
+    const endTime = Math.max(selectedSentence.endTime, startTime + 0.5)
+    const segment: BRollSegment = {
+      id: crypto.randomUUID(),
+      asset: selectedAsset,
+      script_segment_index: selectedSentence.index,
+      start_time: startTime,
+      end_time: endTime,
+      mode,
+      pip_position: pipPosition,
+      pip_scale: mode === "pip" ? pipScale : 0.3,
+    }
+    onConfirm(segment)
+    // 重置素材/句子选择，保留模式设置便于连续添加
+    setSelectedAsset(null)
+    setSelectedSentence(null)
+  }
+
+  const canConfirm = selectedAsset !== null && selectedSentence !== null
+
+  return (
+    <div className="aa-modal-overlay" onClick={onClose}>
+      <div className="aa-modal aa-modal--wide" onClick={(e) => e.stopPropagation()}>
+        {/* 头部 */}
+        <div className="aa-modal__header">
+          <span className="aa-modal__title">🎞️ 画面插入（B-roll）</span>
+          <button type="button" className="aa-modal__close" onClick={onClose}>
+            ✕
+          </button>
+        </div>
+
+        {/* 主体：左素材 + 右设置 */}
+        <div className="aa-modal__body">
+          <div className="aa-broll-modal-body">
+            {/* 左侧：选库 + 素材网格 */}
+            <div className="aa-broll-left">
+              <div className="aa-broll-lib-row">
+                <select
+                  className="aa-select"
+                  value={libraryId}
+                  onChange={(e) => setLibraryId(e.target.value)}
+                  disabled={loadingLibs || libraries.length === 0}
+                >
+                  {libraries.length === 0 ? (
+                    <option value="">{loadingLibs ? "素材库加载中…" : "暂无视频素材库"}</option>
+                  ) : (
+                    libraries.map((lib) => (
+                      <option key={lib.id} value={lib.id}>
+                        📁 {lib.name}
+                      </option>
+                    ))
+                  )}
+                </select>
+              </div>
+              <div className="aa-broll-asset-grid">
+                {availableAssets.map((asset) => {
+                  const alreadySelected = usedAssetIds.has(asset.id)
+                  const isCurrent = selectedAsset?.id === asset.id
+                  const classNames = [
+                    "aa-broll-asset-thumb",
+                    isCurrent ? "selected" : "",
+                    alreadySelected ? "already-selected" : "",
+                  ]
+                    .filter(Boolean)
+                    .join(" ")
+                  return (
+                    <div
+                      key={asset.id}
+                      className={classNames}
+                      onClick={() => handleSelectAsset(asset)}
+                      title={asset.name}
+                    >
+                      {asset.thumbnail_url ? (
+                        <img src={asset.thumbnail_url} alt={asset.name} />
+                      ) : (
+                        <div className="aa-broll-asset-placeholder">🎬</div>
+                      )}
+                      <span className="aa-asset-card__name">{asset.name}</span>
+                    </div>
+                  )
+                })}
+                {loadingAssets ? (
+                  <div className="aa-empty" style={{ gridColumn: "1 / -1" }}>
+                    <div className="aa-empty__icon">⏳</div>
+                    素材加载中…
+                  </div>
+                ) : availableAssets.length === 0 ? (
+                  <div className="aa-empty" style={{ gridColumn: "1 / -1" }}>
+                    <div className="aa-empty__icon">🎬</div>
+                    {assetError || "该素材库暂无视频素材"}
+                  </div>
+                ) : null}
+              </div>
+            </div>
+
+            {/* 右侧：插入设置 */}
+            <div className="aa-broll-right">
+              <div className="aa-broll-settings">
+                {/* ⑤ 文案句子列表（替代段落索引数字框） */}
+                <div className="aa-form-field">
+                  <label className="aa-label">对应文案句子（点选）</label>
+                  {sentences.length === 0 ? (
+                    <div className="aa-sentence-empty">
+                      请先在「文案 & 对口型」面板填写或选择文案
+                    </div>
+                  ) : (
+                    <div className="aa-sentence-list">
+                      {sentences.map((sent) => {
+                        const active = selectedSentence?.index === sent.index
+                        return (
+                          <button
+                            key={sent.index}
+                            type="button"
+                            className={`aa-sentence-item${active ? " active" : ""}`}
+                            onClick={() => setSelectedSentence(sent)}
+                            title={sent.text}
+                          >
+                            <span className="aa-sentence-item__idx">{sent.index + 1}</span>
+                            <span className="aa-sentence-item__text">{sent.text}</span>
+                            {outputDuration > 0 && (
+                              <span className="aa-sentence-item__time">
+                                {sent.startTime.toFixed(1)}-{sent.endTime.toFixed(1)}s
+                              </span>
+                            )}
+                          </button>
+                        )
+                      })}
+                    </div>
+                  )}
+                </div>
+
+                {/* 插入模式 */}
+                <div className="aa-form-field">
+                  <label className="aa-label">插入模式</label>
+                  <div className="aa-broll-mode-toggle">
+                    <button
+                      type="button"
+                      className={`aa-broll-mode-btn${mode === "fullscreen" ? " active" : ""}`}
+                      onClick={() => setMode("fullscreen")}
+                    >
+                      全屏切换
+                    </button>
+                    <button
+                      type="button"
+                      className={`aa-broll-mode-btn${mode === "pip" ? " active" : ""}`}
+                      onClick={() => setMode("pip")}
+                    >
+                      画中画
+                    </button>
+                  </div>
+                </div>
+
+                {/* 画中画：四角位置 + 大小 */}
+                {mode === "pip" && (
+                  <>
+                    <div className="aa-form-field">
+                      <label className="aa-label">画中画位置</label>
+                      <div className="aa-pip-positions">
+                        {PIP_POSITION_OPTIONS.map((opt) => (
+                          <button
+                            key={opt.value}
+                            type="button"
+                            className={`aa-pip-pos-btn${
+                              pipPosition === opt.value ? " active" : ""
+                            }`}
+                            onClick={() => setPipPosition(opt.value)}
+                          >
+                            {opt.label}
+                          </button>
+                        ))}
+                      </div>
+                    </div>
+                    <div className="aa-form-field">
+                      <div className="aa-field-label-row">
+                        <label className="aa-label">画中画大小</label>
+                        <span style={{ fontSize: 12, color: "#8c8ca1" }}>
+                          {Math.round(pipScale * 100)}%
+                        </span>
+                      </div>
+                      <input
+                        type="range"
+                        min={0.1}
+                        max={0.6}
+                        step={0.05}
+                        value={pipScale}
+                        onChange={(e) => setPipScale(Number(e.target.value))}
+                        style={{ width: "100%" }}
+                      />
+                    </div>
+                  </>
+                )}
+
+                {/* 当前选择提示（⑥ 自动估算时间在这里展示） */}
+                <div className="aa-broll-hint">
+                  {selectedAsset && selectedSentence ? (
+                    <>
+                      <div>已选素材：{selectedAsset.name}</div>
+                      <div>
+                        对应第 {selectedSentence.index + 1} 句 · 时间段{" "}
+                        {selectedSentence.startTime.toFixed(1)}s -{" "}
+                        {Math.max(
+                          selectedSentence.endTime,
+                          selectedSentence.startTime + 0.5,
+                        ).toFixed(1)}
+                        s （按字数自动估算）
+                      </div>
+                    </>
+                  ) : (
+                    <div style={{ color: "#8c8ca1" }}>
+                      {!selectedAsset ? "请从左侧选择一段素材" : "请在上方点选对应的文案句子"}
+                    </div>
+                  )}
+                </div>
+              </div>
+            </div>
+          </div>
+
+          {/* 底部：已配置的画面插入列表 */}
+          <div className="aa-broll-list">
+            <div className="aa-broll-list__title">已配置画面插入（{existingSegments.length}）</div>
+            {existingSegments.length === 0 ? (
+              <div className="aa-empty" style={{ padding: 12 }}>
+                尚未配置画面插入
+              </div>
+            ) : (
+              existingSegments.map((seg) => (
+                <div key={seg.id} className="aa-broll-item">
+                  {seg.asset.thumbnail_url ? (
+                    <img className="aa-broll-item__thumb" src={seg.asset.thumbnail_url} alt="" />
+                  ) : (
+                    <div className="aa-broll-item__thumb" />
+                  )}
+                  <div className="aa-broll-item__info">
+                    <div style={{ fontWeight: 500, color: "#1a1a2e" }}>{seg.asset.name}</div>
+                    <div style={{ color: "#8c8ca1", fontSize: 11 }}>
+                      第 {seg.script_segment_index + 1} 句 · {MODE_LABEL[seg.mode]}
+                      {seg.mode === "pip" ? ` · ${seg.pip_position}` : ""} ·{" "}
+                      {seg.start_time.toFixed(1)}s - {seg.end_time.toFixed(1)}s
+                    </div>
+                  </div>
+                  <button
+                    type="button"
+                    className="aa-broll-item__remove"
+                    title="删除"
+                    onClick={() => onRemove(seg.id)}
+                  >
+                    🗑
+                  </button>
+                </div>
+              ))
+            )}
+          </div>
+        </div>
+
+        {/* 底部按钮 */}
+        <div className="aa-modal__footer">
+          <button type="button" className="aa-btn" onClick={onClose}>
+            关闭
+          </button>
+          <button
+            type="button"
+            className="aa-btn aa-btn--primary"
+            disabled={!canConfirm}
+            onClick={handleConfirm}
+          >
+            ✅ 添加画面插入
+          </button>
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default ModalBRollEditor

apps/web/src/pages/ai-avatar/components/PanelCoverAndGenerate.tsx

@@ -0,0 +1,220 @@
+/**
+ * AI数字人 — 面板5：封面 & 生成
+ * - 竖屏 9:16 封面预览（从视频截取 / 自定义上传）
+ * - 分辨率选择（720p / 1080p / 4K）
+ * - 配置汇总卡片（出镜视频/音色/文案/对口型/B-roll/标题/封面）
+ * - 渐变紫色生成按钮
+ *
+ * 注意：v3 已删除"画面插入模式"，本面板不包含该选项。
+ */
+import React, { useRef } from "react"
+import type { AiAvatarCoverConfig } from "../types"
+
+interface PanelCoverAndGenerateProps {
+  coverConfig: AiAvatarCoverConfig
+  onCoverConfigChange: (partial: Partial<AiAvatarCoverConfig>) => void
+  resolution: string
+  onResolutionChange: (r: string) => void
+  isGenerating: boolean
+  onGenerate: () => void
+  /** 智能获取封面（MediaKit 选帧） */
+  onSmartCover: () => void
+  smartCoverLoading: boolean
+  canSmartCover: boolean
+  /** 配置汇总信息 */
+  summary: {
+    videoName: string | null
+    voiceName: string | null
+    scriptLength: number
+    lipsyncStatus: string | null
+    brollCount: number
+    hasTitle: boolean
+    hasCover: boolean
+  }
+}
+
+const RESOLUTION_OPTIONS = [
+  { value: "720p", label: "720p（高清）" },
+  { value: "1080p", label: "1080p（全高清）" },
+  { value: "4k", label: "4K（超清）" },
+]
+
+const LIPSYNC_STATUS_LABEL: Record<string, { text: string; cls: string }> = {
+  idle: { text: "未开始", cls: "aa-status-badge--idle" },
+  pending: { text: "排队中", cls: "aa-status-badge--pending" },
+  processing: { text: "生成中", cls: "aa-status-badge--processing" },
+  completed: { text: "已完成", cls: "aa-status-badge--completed" },
+  failed: { text: "失败", cls: "aa-status-badge--failed" },
+}
+
+const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
+  coverConfig,
+  onCoverConfigChange,
+  resolution,
+  onResolutionChange,
+  isGenerating,
+  onGenerate,
+  onSmartCover,
+  smartCoverLoading,
+  canSmartCover,
+  summary,
+}) => {
+  const uploadInputRef = useRef<HTMLInputElement>(null)
+
+  /** 自定义上传封面 */
+  const handleUploadClick = () => {
+    uploadInputRef.current?.click()
+  }
+
+  const handleFileChange = (e: React.ChangeEvent<HTMLInputElement>) => {
+    const file = e.target.files?.[0]
+    if (!file) return
+    // 本地预览：生成 object URL（实际上传由父级/后端链路处理）
+    const url = URL.createObjectURL(file)
+    onCoverConfigChange({ mode: "upload", upload_url: url, thumbnail_url: url })
+    // 允许重复选择同一文件
+    e.target.value = ""
+  }
+
+  /** 智能获取封面（调后端 MediaKit 抽帧评分选最佳帧，#1822） */
+  const handleSmartCover = () => {
+    onCoverConfigChange({ mode: "auto_frame" })
+    onSmartCover()
+  }
+
+  const lipsync = summary.lipsyncStatus ? LIPSYNC_STATUS_LABEL[summary.lipsyncStatus] : null
+
+  const canGenerate = summary.lipsyncStatus === "completed" && !isGenerating
+
+  return (
+    <div className="aa-cover-generate">
+      {/* 封面预览（竖屏 9:16） */}
+      <div className="aa-cover-preview">
+        {coverConfig.thumbnail_url ? (
+          <img src={coverConfig.thumbnail_url} alt="封面预览" />
+        ) : (
+          <span className="aa-cover-preview__placeholder">暂无封面</span>
+        )}
+      </div>
+
+      <div className="aa-cover-actions">
+        <button
+          type="button"
+          className={`aa-btn aa-btn--ghost${coverConfig.mode === "auto_frame" ? " active" : ""}`}
+          onClick={handleSmartCover}
+          disabled={smartCoverLoading || !canSmartCover}
+          title={canSmartCover ? "基于对口型成片智能选帧" : "请先完成对口型生成"}
+        >
+          {smartCoverLoading ? "⏳ 智能选帧中…" : "🎬 智能获取封面"}
+        </button>
+        <button
+          type="button"
+          className={`aa-btn aa-btn--ghost${coverConfig.mode === "upload" ? " active" : ""}`}
+          onClick={handleUploadClick}
+        >
+          📷 自定义上传
+        </button>
+        <input
+          ref={uploadInputRef}
+          type="file"
+          accept="image/*"
+          style={{ display: "none" }}
+          onChange={handleFileChange}
+        />
+      </div>
+
+      {/* 分辨率选择 */}
+      <div className="aa-form-field">
+        <label className="aa-label">分辨率</label>
+        <select
+          className="aa-select"
+          value={resolution}
+          onChange={(e) => onResolutionChange(e.target.value)}
+        >
+          {RESOLUTION_OPTIONS.map((opt) => (
+            <option key={opt.value} value={opt.value}>
+              {opt.label}
+            </option>
+          ))}
+        </select>
+      </div>
+
+      {/* 配置汇总 */}
+      <div className="aa-generate-section">
+        <div className="aa-config-summary">
+          <div className="aa-config-summary__row">
+            <span>出镜视频</span>
+            {summary.videoName ? (
+              <span className="aa-config-summary__value">{summary.videoName}</span>
+            ) : (
+              <span className="aa-config-summary__empty">未配置</span>
+            )}
+          </div>
+          <div className="aa-config-summary__row">
+            <span>音色</span>
+            {summary.voiceName ? (
+              <span className="aa-config-summary__value">{summary.voiceName}</span>
+            ) : (
+              <span className="aa-config-summary__empty">未配置</span>
+            )}
+          </div>
+          <div className="aa-config-summary__row">
+            <span>文案</span>
+            {summary.scriptLength > 0 ? (
+              <span className="aa-config-summary__value">{summary.scriptLength} 字</span>
+            ) : (
+              <span className="aa-config-summary__empty">未配置</span>
+            )}
+          </div>
+          <div className="aa-config-summary__row">
+            <span>对口型</span>
+            {lipsync ? (
+              <span className={`aa-status-badge ${lipsync.cls}`}>{lipsync.text}</span>
+            ) : (
+              <span className="aa-config-summary__empty">未配置</span>
+            )}
+          </div>
+          <div className="aa-config-summary__row">
+            <span>B-roll 画面插入</span>
+            <span className="aa-config-summary__value">
+              {summary.brollCount > 0 ? `${summary.brollCount} 段` : "无"}
+            </span>
+          </div>
+          <div className="aa-config-summary__row">
+            <span>标题</span>
+            {summary.hasTitle ? (
+              <span className="aa-config-summary__value">已设置</span>
+            ) : (
+              <span className="aa-config-summary__empty">未配置</span>
+            )}
+          </div>
+          <div className="aa-config-summary__row">
+            <span>封面</span>
+            {summary.hasCover ? (
+              <span className="aa-config-summary__value">已开启</span>
+            ) : (
+              <span className="aa-config-summary__empty">未配置</span>
+            )}
+          </div>
+        </div>
+
+        {/* 生成按钮 */}
+        <button
+          type="button"
+          className="aa-btn aa-btn--generate aa-btn--full"
+          disabled={!canGenerate}
+          onClick={onGenerate}
+        >
+          {isGenerating ? "⏳ 生成中..." : "🚀 开始生成视频"}
+        </button>
+        {summary.lipsyncStatus !== "completed" && !isGenerating && (
+          <div style={{ marginTop: 8, fontSize: 11, color: "#8c8ca1", textAlign: "center" }}>
+            请先完成对口型生成
+          </div>
+        )}
+      </div>
+    </div>
+  )
+}
+
+export default PanelCoverAndGenerate

apps/web/src/pages/ai-avatar/components/PanelScriptAndLipsync.tsx

@@ -0,0 +1,222 @@
+/**
+ * AI数字人 — 文案 & 对口型面板（面板4）
+ * 上半区：文案（文案库选择 / 手动输入）；下半区：对口型视频预览（9:16）+ B-roll 画面
+ */
+import { useState } from "react"
+import type { LipsyncJob, BRollSegment } from "../types"
+
+interface PanelScriptAndLipsyncProps {
+  scriptText: string
+  onScriptTextChange: (text: string) => void
+  onOpenScriptModal: () => void
+  lipsyncJob: LipsyncJob | null
+  onGenerateLipsync: () => void
+  bRollSegments: BRollSegment[]
+  onOpenBRollModal: () => void
+  onRemoveBRoll: (id: string) => void
+}
+
+type ScriptTab = "library" | "manual"
+
+const BROLL_MODE_LABEL: Record<BRollSegment["mode"], string> = {
+  fullscreen: "全屏",
+  pip: "画中画",
+}
+
+function formatTime(seconds: number): string {
+  const m = Math.floor(seconds / 60)
+  const s = Math.round(seconds % 60)
+  return `${m}:${s.toString().padStart(2, "0")}`
+}
+
+export function PanelScriptAndLipsync({
+  scriptText,
+  onScriptTextChange,
+  onOpenScriptModal,
+  lipsyncJob,
+  onGenerateLipsync,
+  bRollSegments,
+  onOpenBRollModal,
+  onRemoveBRoll,
+}: PanelScriptAndLipsyncProps) {
+  const [scriptTab, setScriptTab] = useState<ScriptTab>("library")
+
+  /* 对口型状态判断 */
+  const isGenerating = lipsyncJob?.status === "pending" || lipsyncJob?.status === "processing"
+  const isDone = lipsyncJob?.status === "completed"
+  const isFailed = lipsyncJob?.status === "failed"
+
+  const statusText =
+    lipsyncJob?.status === "processing"
+      ? "对口型生成中…"
+      : lipsyncJob?.status === "pending"
+        ? "排队中…"
+        : "对口型生成中…"
+
+  return (
+    <div className="aa-script-lipsync">
+      {/* ── 上半区：文案 ── */}
+      <div className="aa-script-tabs">
+        <button
+          type="button"
+          className={`aa-script-tab${scriptTab === "library" ? " active" : ""}`}
+          onClick={() => setScriptTab("library")}
+        >
+          从文案库选择
+        </button>
+        <button
+          type="button"
+          className={`aa-script-tab${scriptTab === "manual" ? " active" : ""}`}
+          onClick={() => setScriptTab("manual")}
+        >
+          手动输入
+        </button>
+      </div>
+
+      {scriptTab === "library" && (
+        <button
+          type="button"
+          className="aa-btn aa-btn--ghost aa-btn--full"
+          style={{ marginBottom: 8 }}
+          onClick={onOpenScriptModal}
+        >
+          📚 从文案库选择文案
+        </button>
+      )}
+
+      <textarea
+        className="aa-textarea"
+        value={scriptText}
+        readOnly={scriptTab === "library"}
+        placeholder={
+          scriptTab === "library" ? "点击上方按钮，从文案库选择文案…" : "请输入数字人口播文案…"
+        }
+        onChange={(e) => onScriptTextChange(e.target.value)}
+      />
+      <div className="aa-char-count">{scriptText.length} 字</div>
+
+      {/* ── B-roll 画面 ── */}
+      <div className="aa-lipsync-section">
+        <div className="aa-lipsync-section__title">
+          <span style={{ marginRight: 8 }}>🎞️ 插入画面</span>
+          {bRollSegments.length > 0 && (
+            <span className="aa-broll-badge">🎬 {bRollSegments.length} 个画面</span>
+          )}
+        </div>
+        <div className="aa-lipsync-actions">
+          <button
+            type="button"
+            className="aa-btn aa-btn--primary aa-btn--full"
+            onClick={onOpenBRollModal}
+          >
+            🎬 插入画面
+          </button>
+        </div>
+
+        {bRollSegments.length > 0 && (
+          <div className="aa-broll-list">
+            {bRollSegments.map((seg) => (
+              <div key={seg.id} className="aa-broll-item">
+                {seg.asset.thumbnail_url || seg.asset.file_url ? (
+                  <img
+                    className="aa-broll-item__thumb"
+                    src={seg.asset.thumbnail_url || seg.asset.file_url}
+                    alt={seg.asset.name}
+                  />
+                ) : (
+                  <span className="aa-broll-item__thumb" style={{ padding: "6px 4px" }}>
+                    🎬
+                  </span>
+                )}
+                <div className="aa-broll-item__info">
+                  <div
+                    style={{
+                      overflow: "hidden",
+                      textOverflow: "ellipsis",
+                      whiteSpace: "nowrap",
+                    }}
+                  >
+                    {seg.asset.name}
+                  </div>
+                  <div style={{ fontSize: 11, color: "#8c8ca1", marginTop: 2 }}>
+                    {BROLL_MODE_LABEL[seg.mode]} · {formatTime(seg.start_time)}-
+                    {formatTime(seg.end_time)}
+                  </div>
+                </div>
+                <button
+                  type="button"
+                  className="aa-broll-item__remove"
+                  title="删除"
+                  onClick={() => onRemoveBRoll(seg.id)}
+                >
+                  ✕
+                </button>
+              </div>
+            ))}
+          </div>
+        )}
+      </div>
+
+      {/* ── 下半区：对口型预览（竖屏 9:16） ── */}
+      <div className="aa-lipsync-section">
+        <div className="aa-lipsync-section__title">对口型预览</div>
+
+        <div className="aa-lipsync-preview">
+          {isDone && lipsyncJob?.output_video_url ? (
+            <video src={lipsyncJob.output_video_url} controls />
+          ) : isGenerating ? (
+            <div style={{ width: "80%", textAlign: "center", color: "#fff" }}>
+              <div style={{ fontSize: 13, marginBottom: 8 }}>
+                {statusText} {Math.round(lipsyncJob?.progress ?? 0)}%
+              </div>
+              <div className="aa-progress">
+                <div
+                  className="aa-progress__bar"
+                  style={{ width: `${lipsyncJob?.progress ?? 0}%` }}
+                />
+              </div>
+            </div>
+          ) : (
+            <div className="aa-video-preview__placeholder">
+              {isFailed ? (
+                <>
+                  <div style={{ fontSize: 28, marginBottom: 8 }}>❌</div>
+                  <div>对口型生成失败</div>
+                  {lipsyncJob?.error_message && (
+                    <div style={{ fontSize: 11, marginTop: 4, color: "#fca5a5" }}>
+                      {lipsyncJob.error_message}
+                    </div>
+                  )}
+                </>
+              ) : (
+                "生成对口型视频后在此预览"
+              )}
+            </div>
+          )}
+        </div>
+
+        <div className="aa-lipsync-actions">
+          {isDone ? (
+            <button type="button" className="aa-btn aa-btn--full" onClick={onGenerateLipsync}>
+              🔄 重新生成对口型
+            </button>
+          ) : isGenerating ? (
+            <button type="button" className="aa-btn aa-btn--full" disabled>
+              对口型生成中…
+            </button>
+          ) : (
+            <button
+              type="button"
+              className="aa-btn aa-btn--primary aa-btn--full"
+              onClick={onGenerateLipsync}
+            >
+              🎬 生成对口型视频
+            </button>
+          )}
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default PanelScriptAndLipsync

apps/web/src/pages/ai-avatar/components/PanelTitleConfig.tsx

@@ -0,0 +1,134 @@
+/**
+ * AI数字人 — 面板4：标题配置
+ *
+ * 关键：直接复用智能剪辑（generate）模块的 TitleStylePanel 标题样式面板，
+ * 不重新开发标题预设/字体/位置等样式能力。本组件只负责：
+ * - 主标题文字输入
+ * - AiAvatarTitleConfig ↔ TitleSettings 的双向适配
+ * - 自动生成字幕开关
+ */
+import React, { useMemo, useState, useEffect } from "react"
+import { Input } from "antd"
+import TitleStylePanel from "@/pages/generate/components/title/TitleStylePanel"
+import TitleLibraryAutoComplete from "@/pages/generate/components/title/TitleLibraryAutoComplete"
+import type { TitleOption } from "@/pages/generate/components/title/TitleLibraryAutoComplete"
+import type { TitleSettings } from "@/pages/generate/types"
+import { POSITION_OPTIONS, FONT_OPTIONS, TITLE_PRESETS } from "@/pages/generate/constants"
+import type { AiAvatarTitleConfig } from "../types"
+import { getTitles } from "@/api/titles"
+
+const { TextArea } = Input
+
+interface PanelTitleConfigProps {
+  titleConfig: AiAvatarTitleConfig
+  onUpdate: (partial: Partial<AiAvatarTitleConfig>) => void
+}
+
+const PanelTitleConfig: React.FC<PanelTitleConfigProps> = ({ titleConfig, onUpdate }) => {
+  /** TitleStylePanel 内部高亮的预设 key（面板本地状态） */
+  const [activePreset, setActivePreset] = useState<string | null>(null)
+
+  /** 标题库选项（复用智能剪辑的标题库） */
+  const [titleOptions, setTitleOptions] = useState<TitleOption[]>([])
+  useEffect(() => {
+    getTitles()
+      .then((items) => setTitleOptions(items.map((t) => ({ label: t.content, value: t.content }))))
+      .catch(() => setTitleOptions([]))
+  }, [])
+
+  /** AiAvatarTitleConfig → TitleSettings（补齐 aiAutoSelect / 自由坐标字段） */
+  const titleSettings: TitleSettings = useMemo(
+    () => ({
+      aiAutoSelect: false,
+      title: titleConfig.title,
+      position: titleConfig.position,
+      font: titleConfig.font,
+      size: titleConfig.size,
+      bold: titleConfig.bold,
+      italic: titleConfig.italic,
+      stroke: titleConfig.stroke,
+      shadow: titleConfig.shadow,
+      color: titleConfig.color,
+      posX: null,
+      posY: null,
+    }),
+    [titleConfig],
+  )
+
+  /** 应用预设：与智能剪辑一致，只覆盖 color/bold/italic/stroke/shadow，不改变字号 */
+  const handleApplyPreset = (presetKey: string) => {
+    const preset = TITLE_PRESETS.find((p) => p.key === presetKey)
+    if (!preset) return
+    setActivePreset(presetKey)
+    onUpdate({
+      color: preset.style.color,
+      bold: preset.style.bold,
+      italic: preset.style.italic,
+      stroke: preset.style.stroke,
+      shadow: preset.style.shadow,
+    })
+  }
+
+  return (
+    <div className="aa-title-config">
+      {/* 主标题输入 — TextArea 多行 + 标题库选择 */}
+      <div className="aa-form-field">
+        <label className="aa-label">主标题</label>
+        <TextArea
+          className="aa-title-input"
+          placeholder="输入视频标题（支持 / 分行）"
+          value={titleConfig.title}
+          autoSize={{ minRows: 2, maxRows: 4 }}
+          maxLength={200}
+          onChange={(e) => onUpdate({ title: e.target.value })}
+          style={{ fontSize: 15 }}
+        />
+        <div style={{ marginTop: 8, display: "flex", alignItems: "center", gap: 8 }}>
+          <span style={{ fontSize: 12, color: "#8c8ca1", whiteSpace: "nowrap" }}>📚 标题库</span>
+          <TitleLibraryAutoComplete
+            key={titleConfig.title}
+            placeholder="选择标题填入上方"
+            value=""
+            onChange={(val) => {
+              if (val) onUpdate({ title: val })
+            }}
+            options={titleOptions}
+            maxLength={200}
+            style={{ flex: 1 }}
+          />
+        </div>
+      </div>
+
+      {/* 标题样式：直接复用智能剪辑 TitleStylePanel（位置/字体/字号/样式/预设） */}
+      <TitleStylePanel
+        settings={titleSettings}
+        onUpdatePosition={(position) => onUpdate({ position })}
+        onUpdateFont={(font) => onUpdate({ font })}
+        onUpdateSize={(size) => onUpdate({ size: Math.min(128, Math.max(16, size)) })}
+        onToggleBold={() => onUpdate({ bold: !titleConfig.bold })}
+        onToggleItalic={() => onUpdate({ italic: !titleConfig.italic })}
+        onToggleStroke={() => onUpdate({ stroke: !titleConfig.stroke })}
+        onToggleShadow={() => onUpdate({ shadow: !titleConfig.shadow })}
+        onApplyPreset={handleApplyPreset}
+        activePreset={activePreset}
+        titlePresets={TITLE_PRESETS}
+        POSITION_OPTIONS={POSITION_OPTIONS}
+        FONT_OPTIONS={FONT_OPTIONS}
+      />
+
+      {/* 自动生成字幕 */}
+      <div className="aa-subtitle-toggle">
+        <label className="aa-checkbox-row">
+          <input
+            type="checkbox"
+            checked={titleConfig.auto_subtitle}
+            onChange={(e) => onUpdate({ auto_subtitle: e.target.checked })}
+          />
+          自动生成字幕
+        </label>
+      </div>
+    </div>
+  )
+}
+
+export default PanelTitleConfig

apps/web/src/pages/ai-avatar/components/PanelVideoSelector.tsx

@@ -0,0 +1,124 @@
+/**
+ * AI数字人 — 出镜视频选择面板
+ * - 未选视频：虚线上传区，点击打开素材库弹窗
+ * - 已选视频：竖屏 9:16 预览播放器 + 视频信息卡片 + 移除按钮
+ */
+import type { AssetItem } from "@/api/assets"
+import type { AiAvatarTitleConfig } from "../types"
+import { getFontFamily } from "@/pages/generate/constants"
+
+export interface PanelVideoSelectorProps {
+  selectedVideo: AssetItem | null
+  /** 触发打开素材库弹窗 */
+  onSelectVideo: () => void
+  onRemoveVideo: () => void
+  titleConfig?: AiAvatarTitleConfig
+}
+
+/** 格式化时长（秒 → mm:ss） */
+function formatDuration(seconds?: number): string {
+  if (typeof seconds !== "number" || !Number.isFinite(seconds) || seconds <= 0) {
+    return "00:00"
+  }
+  return `${Math.floor(seconds / 60)}:${String(Math.floor(seconds % 60)).padStart(2, "0")}`
+}
+
+export function PanelVideoSelector({
+  selectedVideo,
+  onSelectVideo,
+  onRemoveVideo,
+  titleConfig,
+}: PanelVideoSelectorProps) {
+  /* 未选视频：虚线上传区，点击打开素材库弹窗 */
+  if (!selectedVideo) {
+    return (
+      <div
+        className="aa-upload-zone"
+        role="button"
+        tabIndex={0}
+        onClick={onSelectVideo}
+        onKeyDown={(e) => {
+          if (e.key === "Enter" || e.key === " ") {
+            e.preventDefault()
+            onSelectVideo()
+          }
+        }}
+      >
+        <div className="aa-upload-zone__icon">🎬</div>
+        <div className="aa-upload-zone__text">从素材库选择视频</div>
+      </div>
+    )
+  }
+
+  const width = selectedVideo.metadata?.width
+  const height = selectedVideo.metadata?.height
+  const duration = selectedVideo.duration ?? selectedVideo.metadata?.duration
+  const fileUrl = selectedVideo.file_url ?? ""
+
+  return (
+    <div>
+      {/* 竖屏 9:16 视频预览播放器 + 标题实时预览 */}
+      <div className="aa-video-preview" style={{ position: "relative" }}>
+        {fileUrl ? (
+          <video src={fileUrl} poster={selectedVideo.thumbnail_url} controls playsInline />
+        ) : (
+          <div className="aa-video-preview__placeholder">视频暂不可预览</div>
+        )}
+        {titleConfig?.title && (
+          <div
+            style={{
+              position: "absolute",
+              left: "50%",
+              transform: "translateX(-50%)",
+              ...(titleConfig.position === "top"
+                ? { top: "10%" }
+                : titleConfig.position === "bottom"
+                  ? { bottom: "10%" }
+                  : { top: "50%", transform: "translate(-50%, -50%)" }),
+              fontSize: Math.max(titleConfig.size, 32),
+              fontFamily: getFontFamily(titleConfig.font),
+              color: titleConfig.color,
+              fontWeight: titleConfig.bold ? 700 : 400,
+              fontStyle: titleConfig.italic ? "italic" : "normal",
+              textShadow: "0 2px 4px rgba(0,0,0,0.5)",
+              WebkitTextStroke: "2px #000",
+              pointerEvents: "none",
+              zIndex: 10,
+              maxWidth: "90%",
+              textAlign: "center",
+              whiteSpace: "pre-wrap",
+              lineHeight: 1.3,
+            }}
+          >
+            {titleConfig.title}
+          </div>
+        )}
+      </div>
+
+      {/* 视频信息卡片：文件名 / 时长 / 分辨率 */}
+      <div className="aa-video-info">
+        <div className="aa-video-info__row">
+          <span>文件名</span>
+          <span title={selectedVideo.name}>{selectedVideo.name}</span>
+        </div>
+        <div className="aa-video-info__row">
+          <span>时长</span>
+          <span>{formatDuration(duration)}</span>
+        </div>
+        <div className="aa-video-info__row">
+          <span>分辨率</span>
+          <span>{width && height ? `${width}×${height}` : "—"}</span>
+        </div>
+      </div>
+
+      <button
+        type="button"
+        className="aa-btn aa-btn--danger aa-btn--full"
+        style={{ marginTop: 10 }}
+        onClick={onRemoveVideo}
+      >
+        移除视频
+      </button>
+    </div>
+  )
+}

apps/web/src/pages/ai-avatar/components/PanelVoiceSelector.tsx

@@ -0,0 +1,334 @@
+/**
+ * AI数字人 — 配音库面板（面板3）
+ * 音色来源切换（系统预设 / 我的音色）、音色选择与试听、情绪/语速/语言参数
+ */
+import { useEffect, useRef, useState } from "react"
+import { message } from "antd"
+import { fetchVoices } from "@/api/voices/voices"
+import { previewTts } from "@/api/tts"
+import { normalizeEmotion } from "../utils/contract"
+import type { UnifiedVoiceItem } from "@/api/voices/types"
+import {
+  type VoiceSource,
+  type VoiceEmotion,
+  type VoiceLanguage,
+  VOICE_EMOTION_OPTIONS,
+  VOICE_LANGUAGE_OPTIONS,
+} from "../types"
+
+interface PanelVoiceSelectorProps {
+  voiceSource: VoiceSource
+  onVoiceSourceChange: (source: VoiceSource) => void
+  selectedVoice: UnifiedVoiceItem | null
+  onSelectVoice: (voice: UnifiedVoiceItem) => void
+  emotion: VoiceEmotion
+  onEmotionChange: (e: VoiceEmotion) => void
+  speed: number
+  onSpeedChange: (s: number) => void
+  language: VoiceLanguage
+  onLanguageChange: (l: VoiceLanguage) => void
+}
+
+export function PanelVoiceSelector({
+  voiceSource,
+  onVoiceSourceChange,
+  selectedVoice,
+  onSelectVoice,
+  emotion,
+  onEmotionChange,
+  speed,
+  onSpeedChange,
+  language,
+  onLanguageChange,
+}: PanelVoiceSelectorProps) {
+  const [voices, setVoices] = useState<UnifiedVoiceItem[]>([])
+  const [loading, setLoading] = useState(false)
+  const [error, setError] = useState<string | null>(null)
+  const [previewingId, setPreviewingId] = useState<string | null>(null)
+  const audioRef = useRef<HTMLAudioElement | null>(null)
+  /** 克隆音色试听合成缓存：voiceId -> url，对齐配音库 useAudioPlayer */
+  const previewCacheRef = useRef<Map<string, string>>(new Map())
+  const VOICE_PREVIEW_TEXT = "你好呀，欢迎使用小虾智剪，这是我的配音效果，希望你喜欢。"
+
+  /* 切换来源时重新获取音色列表 */
+  useEffect(() => {
+    let cancelled = false
+    const loadVoices = async () => {
+      setLoading(true)
+      setError(null)
+      try {
+        const res = await fetchVoices({ type: voiceSource })
+        if (!cancelled) setVoices(Array.isArray(res?.items) ? res.items : [])
+      } catch (err) {
+        if (!cancelled) setError(err instanceof Error ? err.message : "音色加载失败")
+      } finally {
+        if (!cancelled) setLoading(false)
+      }
+    }
+    loadVoices()
+    return () => {
+      cancelled = true
+    }
+  }, [voiceSource])
+
+  /* 卸载时停止试听 */
+  useEffect(() => {
+    return () => {
+      if (audioRef.current) {
+        audioRef.current.pause()
+        audioRef.current = null
+      }
+    }
+  }, [])
+
+  const stopPreview = () => {
+    if (audioRef.current) {
+      audioRef.current.pause()
+      audioRef.current = null
+    }
+    setPreviewingId(null)
+  }
+
+  const NO_PREVIEW_TIP = "该音色暂无试听音频，请先用此音色生成一段配音后再试听"
+
+  /** 用指定 URL 真实播放（抽取公共） */
+  const playAudioUrl = (voiceId: string, url: string) => {
+    // 临时兼容：后端 /tts/preview 返回 HTTP URL，staging 是 HTTPS，Mixed Content 会阻止加载
+    // OSS 同时支持 HTTP/HTTPS，直接替换协议即可
+    const safeUrl = url.startsWith("http://") ? url.replace("http://", "https://") : url
+    if (audioRef.current) {
+      audioRef.current.pause()
+      audioRef.current = null
+    }
+    const audio = new Audio(safeUrl)
+    audioRef.current = audio
+    setPreviewingId(voiceId)
+    audio.onended = () => {
+      if (audioRef.current === audio) {
+        audioRef.current = null
+        setPreviewingId(null)
+      }
+    }
+    audio.onerror = () => {
+      if (audioRef.current === audio) {
+        audioRef.current = null
+        setPreviewingId(null)
+        message.error("试听音频加载失败")
+      }
+    }
+    void audio.play().catch(() => {
+      setPreviewingId(null)
+      message.error("试听播放失败")
+    })
+  }
+
+  const handlePreview = async (voice: UnifiedVoiceItem) => {
+    /* 再次点击当前试听音色 → 停止 */
+    if (previewingId === voice.id) {
+      stopPreview()
+      return
+    }
+
+    /* 克隆音色：preview_url/audio_url 通常为空，需走 POST /tts/preview
+     * 现合成示例文案再播放，对齐配音库 useAudioPlayer 行为 */
+    if (voice.type === "clone") {
+      const cached = previewCacheRef.current.get(voice.voice_clone_profile_id || voice.id)
+      if (cached) {
+        playAudioUrl(voice.id, cached)
+        return
+      }
+      const targetId = voice.voice_clone_profile_id || voice.id
+      // DEBUG: 打印请求参数，帮助定位 /tts/preview 失败原因
+      console.log("[AI数字人-克隆试听] previewTts 请求:", {
+        voice_id: targetId,
+        voice_name: voice.name,
+        voice_type: voice.type,
+        voice_clone_profile_id: voice.voice_clone_profile_id,
+        voice_id_field: voice.voice_id,
+      })
+      setPreviewingId(voice.id)
+      try {
+        const res = await previewTts({
+          text: VOICE_PREVIEW_TEXT,
+          voice_id: targetId,
+          speed: speed, // 透传用户选择的语速（#1822）
+          emotion: normalizeEmotion(emotion), // 情绪中文→英文枚举
+        })
+        console.log("[AI数字人-克隆试听] previewTts 响应:", {
+          audio_url: res.audio_url?.substring(0, 80),
+          duration: res.duration,
+        })
+        if (!res.audio_url) {
+          setPreviewingId(null)
+          message.error("合成试听失败：未返回音频")
+          return
+        }
+        previewCacheRef.current.set(targetId, res.audio_url)
+        playAudioUrl(voice.id, res.audio_url)
+      } catch (err) {
+        setPreviewingId(null)
+        // DEBUG: 打印详细错误信息
+        console.error("[AI数字人-克隆试听] previewTts 失败:", {
+          status: (err as { response?: { status?: number } })?.response?.status,
+          data: (err as { response?: { data?: unknown } })?.response?.data,
+          message: err instanceof Error ? err.message : String(err),
+        })
+        // apiClient 拦截器已统一 toast
+      }
+      return
+    }
+
+    /* 系统预设音色：沿用 preview_url/audio_url 直链播放 */
+    const url = voice.preview_url || voice.audio_url
+    if (!url) {
+      message.warning(NO_PREVIEW_TIP)
+      return
+    }
+    playAudioUrl(voice.id, url)
+  }
+
+  const handleSpeedChange = (value: string) => {
+    const parsed = parseFloat(value)
+    if (Number.isNaN(parsed)) return
+    const clamped = Math.min(2.0, Math.max(0.5, parsed))
+    onSpeedChange(clamped)
+  }
+
+  return (
+    <div className="aa-voice-selector">
+      {/* 音色来源切换 */}
+      <div className="aa-voice-source-toggle">
+        <button
+          type="button"
+          className={`aa-voice-source-btn${voiceSource === "preset" ? " active" : ""}`}
+          onClick={() => onVoiceSourceChange("preset")}
+        >
+          系统预设
+        </button>
+        <button
+          type="button"
+          className={`aa-voice-source-btn${voiceSource === "clone" ? " active" : ""}`}
+          onClick={() => onVoiceSourceChange("clone")}
+        >
+          我的音色
+        </button>
+      </div>
+
+      {/* 音色列表 */}
+      {loading ? (
+        <div className="aa-empty">
+          <div className="aa-empty__icon">⏳</div>
+          <div>音色加载中…</div>
+        </div>
+      ) : error ? (
+        <div className="aa-empty">
+          <div className="aa-empty__icon">⚠️</div>
+          <div>{error}</div>
+        </div>
+      ) : voices.length === 0 ? (
+        <div className="aa-empty">
+          <div className="aa-empty__icon">🎙️</div>
+          <div>{voiceSource === "clone" ? "还没有克隆音色" : "暂无预置音色"}</div>
+        </div>
+      ) : (
+        <div className="aa-voice-list">
+          {voices.map((voice) => {
+            const selected = selectedVoice?.id === voice.id
+            const previewUrl = voice.preview_url || voice.audio_url
+            return (
+              <div
+                key={voice.id}
+                className={`aa-voice-card${selected ? " selected" : ""}`}
+                onClick={() => onSelectVoice(voice)}
+              >
+                <span className="aa-voice-card__radio" />
+                <div className="aa-voice-card__info">
+                  <div className="aa-voice-card__name">{voice.name}</div>
+                  {voice.description && (
+                    <div className="aa-voice-card__desc">{voice.description}</div>
+                  )}
+                </div>
+                <button
+                  type="button"
+                  className="aa-voice-card__preview"
+                  title={previewingId === voice.id ? "停止试听" : "试听"}
+                  disabled={voice.type === "preset" && !previewUrl}
+                  onClick={(e) => {
+                    e.stopPropagation()
+                    handlePreview(voice)
+                  }}
+                >
+                  {previewingId === voice.id ? "⏸" : "▶"}
+                </button>
+              </div>
+            )
+          })}
+        </div>
+      )}
+
+      {/* 我的音色：克隆入口 */}
+      {voiceSource === "clone" && (
+        <div className="aa-clone-entry">
+          <a href="/app/voice-clone">+ 克隆新音色</a>
+        </div>
+      )}
+
+      {/* 配音参数 */}
+      <div className="aa-voice-params">
+        <div className="aa-voice-params__row">
+          <div className="aa-voice-params__field">
+            <label className="aa-label" htmlFor="aa-voice-emotion">
+              情绪
+            </label>
+            <select
+              id="aa-voice-emotion"
+              className="aa-select"
+              value={emotion}
+              onChange={(e) => onEmotionChange(e.target.value as VoiceEmotion)}
+            >
+              {VOICE_EMOTION_OPTIONS.map((opt) => (
+                <option key={opt.value} value={opt.value}>
+                  {opt.label}
+                </option>
+              ))}
+            </select>
+          </div>
+          <div className="aa-voice-params__field">
+            <label className="aa-label" htmlFor="aa-voice-language">
+              语言
+            </label>
+            <select
+              id="aa-voice-language"
+              className="aa-select"
+              value={language}
+              onChange={(e) => onLanguageChange(e.target.value as VoiceLanguage)}
+            >
+              {VOICE_LANGUAGE_OPTIONS.map((opt) => (
+                <option key={opt.value} value={opt.value}>
+                  {opt.label}
+                </option>
+              ))}
+            </select>
+          </div>
+        </div>
+        <div className="aa-voice-params__field">
+          <label className="aa-label" htmlFor="aa-voice-speed">
+            语速（{speed.toFixed(1)}x）
+          </label>
+          <input
+            id="aa-voice-speed"
+            type="number"
+            className="aa-input"
+            min={0.5}
+            max={2.0}
+            step={0.1}
+            value={speed}
+            onChange={(e) => handleSpeedChange(e.target.value)}
+          />
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default PanelVoiceSelector

apps/web/src/pages/ai-avatar/components/TitleLibraryModal.tsx

@@ -0,0 +1,97 @@
+/**
+ * AI数字人 — 标题库选择弹窗
+ * 复用智能剪辑的标题库 API，选择标题后填入输入框
+ */
+import React, { useEffect, useState } from "react"
+import { getTitles } from "@/api/titles"
+import type { TitleItem } from "@/api/titles/types"
+
+interface TitleLibraryModalProps {
+  open: boolean
+  onClose: () => void
+  onSelect: (title: string) => void
+}
+
+const TitleLibraryModal: React.FC<TitleLibraryModalProps> = ({ open, onClose, onSelect }) => {
+  const [titles, setTitles] = useState<TitleItem[]>([])
+  const [loading, setLoading] = useState(false)
+  const [search, setSearch] = useState("")
+
+  useEffect(() => {
+    if (!open) return
+    setLoading(true)
+    getTitles()
+      .then((items) => setTitles(items))
+      .catch(() => setTitles([]))
+      .finally(() => setLoading(false))
+  }, [open])
+
+  const filtered = titles.filter(
+    (t) => !search || t.content.toLowerCase().includes(search.toLowerCase()),
+  )
+
+  if (!open) return null
+
+  return (
+    <div className="aa-modal-overlay" onClick={onClose}>
+      <div className="aa-modal" onClick={(e) => e.stopPropagation()} style={{ maxWidth: 600 }}>
+        <div className="aa-modal__header">
+          <span className="aa-modal__title">从标题库选择</span>
+          <button className="aa-modal__close" onClick={onClose}></button>
+        </div>
+        <div className="aa-modal__body">
+          <div style={{ marginBottom: 12 }}>
+            <input
+              className="aa-input"
+              placeholder="搜索标题..."
+              value={search}
+              onChange={(e) => setSearch(e.target.value)}
+            />
+          </div>
+          {loading ? (
+            <div style={{ textAlign: "center", padding: 40, color: "#8c8ca1" }}>加载中...</div>
+          ) : filtered.length === 0 ? (
+            <div style={{ textAlign: "center", padding: 40, color: "#8c8ca1" }}>
+              暂无标题，请先在标题库创建
+            </div>
+          ) : (
+            <div style={{ maxHeight: 400, overflowY: "auto" }}>
+              {filtered.map((t) => (
+                <div
+                  key={t.id}
+                  style={{
+                    padding: "12px 16px",
+                    marginBottom: 8,
+                    background: "#f8f8fc",
+                    borderRadius: 8,
+                    cursor: "pointer",
+                    transition: "background 0.2s",
+                  }}
+                  onMouseEnter={(e) => (e.currentTarget.style.background = "#eef0ff")}
+                  onMouseLeave={(e) => (e.currentTarget.style.background = "#f8f8fc")}
+                  onClick={() => {
+                    onSelect(t.content)
+                    onClose()
+                  }}
+                >
+                  <div style={{ fontSize: 14, color: "#1a1a2e", marginBottom: 4 }}>{t.content}</div>
+                  <div style={{ fontSize: 12, color: "#8c8ca1" }}>
+                    {t.word_count ?? t.content.length}字 ·{" "}
+                    {t.created_at ? new Date(t.created_at).toLocaleDateString() : ""}
+                  </div>
+                </div>
+              ))}
+            </div>
+          )}
+        </div>
+        <div className="aa-modal__footer">
+          <button className="aa-btn" onClick={onClose}>
+            取消
+          </button>
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default TitleLibraryModal

apps/web/src/pages/ai-avatar/hooks/useAiAvatar.ts

@@ -0,0 +1,141 @@
+/**
+ * AI数字人 — 页面全局状态管理 hook（v3）
+ */
+import { useState, useCallback } from "react"
+import type { AssetItem } from "@/api/assets"
+import type { UnifiedVoiceItem } from "@/api/voices/types"
+import {
+  type VoiceSource,
+  type VoiceEmotion,
+  type VoiceLanguage,
+  type Script,
+  type LipsyncJob,
+  type BRollSegment,
+  type AiAvatarTitleConfig,
+  type AiAvatarCoverConfig,
+  DEFAULT_TITLE_CONFIG,
+  DEFAULT_COVER_CONFIG,
+} from "../types"
+
+export function useAiAvatar() {
+  /* ── 面板1：出镜视频 ── */
+  const [selectedVideo, setSelectedVideo] = useState<AssetItem | null>(null)
+  const [showAssetPicker, setShowAssetPicker] = useState(false)
+
+  /* ── 面板2：配音库 ── */
+  const [voiceSource, setVoiceSource] = useState<VoiceSource>("preset")
+  const [selectedVoice, setSelectedVoice] = useState<UnifiedVoiceItem | null>(null)
+  const [emotion, setEmotion] = useState<VoiceEmotion>("natural")
+  const [speed, setSpeed] = useState(1.0)
+  const [language, setLanguage] = useState<VoiceLanguage>("mandarin")
+
+  /* ── 面板3：文案 & 对口型 ── */
+  const [script, setScript] = useState<Script | null>(null)
+  const [scriptText, setScriptText] = useState("")
+  const [lipsyncJob, setLipsyncJob] = useState<LipsyncJob | null>(null)
+  const [showScriptModal, setShowScriptModal] = useState(false)
+  const [showBRollModal, setShowBRollModal] = useState(false)
+
+  /* ── 面板3.5：B-roll ── */
+  const [bRollSegments, setBRollSegments] = useState<BRollSegment[]>([])
+
+  /* ── 面板4：标题配置 ── */
+  const [titleConfig, setTitleConfig] = useState<AiAvatarTitleConfig>(DEFAULT_TITLE_CONFIG)
+
+  /* ── 面板5：封面 & 生成 ── */
+  const [coverConfig, setCoverConfig] = useState<AiAvatarCoverConfig>(DEFAULT_COVER_CONFIG)
+  const [resolution, setResolution] = useState("1080p")
+  const [isGenerating, setIsGenerating] = useState(false)
+
+  /* ── Actions ── */
+  const selectVideo = useCallback((asset: AssetItem) => {
+    setSelectedVideo(asset)
+    setShowAssetPicker(false)
+  }, [])
+
+  const removeVideo = useCallback(() => {
+    setSelectedVideo(null)
+  }, [])
+
+  const selectScript = useCallback((s: Script) => {
+    setScript(s)
+    setScriptText(s.content)
+    setShowScriptModal(false)
+  }, [])
+
+  const addBRollSegment = useCallback((segment: BRollSegment) => {
+    setBRollSegments((prev) => [...prev, segment])
+  }, [])
+
+  const removeBRollSegment = useCallback((id: string) => {
+    setBRollSegments((prev) => prev.filter((s) => s.id !== id))
+  }, [])
+
+  const updateTitleConfig = useCallback((partial: Partial<AiAvatarTitleConfig>) => {
+    setTitleConfig((prev) => ({ ...prev, ...partial }))
+  }, [])
+
+  const reset = useCallback(() => {
+    setSelectedVideo(null)
+    setSelectedVoice(null)
+    setScript(null)
+    setScriptText("")
+    setLipsyncJob(null)
+    setBRollSegments([])
+    setTitleConfig(DEFAULT_TITLE_CONFIG)
+    setCoverConfig(DEFAULT_COVER_CONFIG)
+    setResolution("1080p")
+    setIsGenerating(false)
+  }, [])
+
+  return {
+    // 面板1
+    selectedVideo,
+    showAssetPicker,
+    setShowAssetPicker,
+    selectVideo,
+    removeVideo,
+    // 面板2
+    voiceSource,
+    setVoiceSource,
+    selectedVoice,
+    setSelectedVoice,
+    emotion,
+    setEmotion,
+    speed,
+    setSpeed,
+    language,
+    setLanguage,
+    // 面板3
+    script,
+    setScript,
+    scriptText,
+    setScriptText,
+    lipsyncJob,
+    setLipsyncJob,
+    showScriptModal,
+    setShowScriptModal,
+    showBRollModal,
+    setShowBRollModal,
+    selectScript,
+    // B-roll
+    bRollSegments,
+    addBRollSegment,
+    removeBRollSegment,
+    // 面板4
+    titleConfig,
+    updateTitleConfig,
+    setTitleConfig,
+    // 面板5
+    coverConfig,
+    setCoverConfig,
+    resolution,
+    setResolution,
+    isGenerating,
+    setIsGenerating,
+    // 全局
+    reset,
+  }
+}
+
+export type UseAiAvatarReturn = ReturnType<typeof useAiAvatar>

apps/web/src/pages/ai-avatar/types.ts

@@ -0,0 +1,131 @@
+/**
+ * AI数字人 — TypeScript 类型定义（v3）
+ */
+import type { AssetItem } from "@/api/assets"
+
+/* ── 音色来源切换 ── */
+export type VoiceSource = "preset" | "clone"
+
+/* ── 情绪 ── */
+export type VoiceEmotion = "natural" | "excited" | "calm" | "friendly"
+
+export const VOICE_EMOTION_OPTIONS: { value: VoiceEmotion; label: string }[] = [
+  { value: "natural", label: "自然" },
+  { value: "excited", label: "兴奋" },
+  { value: "calm", label: "沉稳" },
+  { value: "friendly", label: "亲切" },
+]
+
+/* ── 语言 ── */
+export type VoiceLanguage = "mandarin" | "english" | "cantonese"
+
+export const VOICE_LANGUAGE_OPTIONS: { value: VoiceLanguage; label: string }[] = [
+  { value: "mandarin", label: "普通话" },
+  { value: "english", label: "English" },
+  { value: "cantonese", label: "粤语" },
+]
+
+/* ── 对口型任务状态 ── */
+export type LipsyncStatus = "idle" | "pending" | "processing" | "completed" | "failed"
+
+/* ── 文案 ── */
+export interface Script {
+  id: string
+  title: string
+  content: string
+  char_count: number
+  created_at: string
+  updated_at?: string
+}
+
+/* ── 对口型任务 ── */
+export interface LipsyncJob {
+  id: string
+  status: LipsyncStatus
+  progress: number
+  output_video_url: string | null
+  /** 对口型成片总时长（秒），后端返回；用于 B-roll 时间自动估算（#1809 ⑥） */
+  output_duration?: number
+  error_message: string | null
+  created_at: string
+}
+
+/* ── B-roll 画面插入 ── */
+export type BRollInsertMode = "fullscreen" | "pip"
+export type PipPosition = "top-left" | "top-right" | "bottom-left" | "bottom-right"
+
+export interface BRollSegment {
+  id: string
+  asset: AssetItem
+  script_segment_index: number
+  start_time: number
+  end_time: number
+  mode: BRollInsertMode
+  pip_position: PipPosition
+  pip_scale: number
+}
+
+/* ── 标题配置 ── */
+export interface AiAvatarTitleConfig {
+  title: string
+  position: string
+  font: string
+  size: number
+  bold: boolean
+  italic: boolean
+  stroke: boolean
+  shadow: boolean
+  color: string
+  auto_subtitle: boolean
+  /** 自定义位置坐标（position=custom 时生效，像素） */
+  pos_x?: number
+  pos_y?: number
+}
+
+/* ── 封面配置 ── */
+export interface AiAvatarCoverConfig {
+  enabled: boolean
+  mode: "auto_frame" | "upload"
+  frame_time: number
+  upload_url: string | null
+  thumbnail_url: string | null
+  /** 智能封面（MediaKit 选帧）返回的 OSS 非临时 URL（#1822） */
+  smart_cover_url: string | null
+}
+
+/* ── 渲染任务 ── */
+export type RenderStatus = "pending" | "processing" | "completed" | "failed" | "cancelled"
+
+export interface RenderJob {
+  id: string
+  status: RenderStatus
+  progress: number
+  output_video_url: string | null
+  error_message: string | null
+  created_at: string
+}
+
+/* ── 默认值 ── */
+export const DEFAULT_TITLE_CONFIG: AiAvatarTitleConfig = {
+  title: "",
+  position: "bottom",
+  font: "思源黑体",
+  size: 28,
+  bold: true,
+  italic: false,
+  stroke: false,
+  shadow: false,
+  color: "#ffffff",
+  auto_subtitle: true,
+  pos_x: undefined,
+  pos_y: undefined,
+}
+
+export const DEFAULT_COVER_CONFIG: AiAvatarCoverConfig = {
+  enabled: true,
+  mode: "auto_frame",
+  frame_time: 0,
+  upload_url: null,
+  thumbnail_url: null,
+  smart_cover_url: null,
+}

apps/web/src/pages/ai-avatar/utils/contract.ts

@@ -0,0 +1,76 @@
+/**
+ * AI数字人 — 前后端接口契约转换工具（#1822）
+ *
+ * 以 packages/domain/video_filter_builder.py 的 build_title_drawtext_filter() 为唯一口径
+ * （契约文档第 5 节的 titles[]/fontSize/frame/start/end 为误写，后端不认，禁止使用）。
+ */
+import type { AiAvatarTitleConfig, AiAvatarCoverConfig, VoiceEmotion } from "../types"
+
+/* ── 情绪：中文 → 英文（防御性映射；state 默认已是英文） ── */
+const EMOTION_ZH_TO_EN: Record<string, VoiceEmotion> = {
+  自然: "natural",
+  兴奋: "excited",
+  沉稳: "calm",
+  亲切: "friendly",
+}
+const VALID_EMOTIONS: VoiceEmotion[] = ["natural", "excited", "calm", "friendly"]
+
+/** 归一化为后端英文枚举 natural/excited/calm/friendly；非法/空值回退 natural。 */
+export function normalizeEmotion(raw: string | undefined | null): VoiceEmotion {
+  if (!raw) return "natural"
+  const v = raw.trim()
+  if ((VALID_EMOTIONS as string[]).includes(v)) return v as VoiceEmotion
+  return EMOTION_ZH_TO_EN[v] ?? "natural"
+}
+
+/* ── 标题：前端 state → 后端 build_title_drawtext_filter 字段（单个 title_config dict） ── */
+/**
+ * 后端真实字段：text(或content)、font(或font_preset)、font_size(或size)、
+ * font_color(或color，可传 #RRGGBB)、position(top/center/bottom/custom)、
+ * enabled、bold、stroke{enabled,width,color}、shadow{enabled,color,offset_x,offset_y}、
+ * pos_x/pos_y(custom 时)。
+ * 口播标题默认 position=bottom（不传后端会默认 top 跑到画面顶部）。
+ */
+export function buildTitleConfigPayload(cfg: AiAvatarTitleConfig): Record<string, unknown> {
+  const text = (cfg.title || "").trim()
+  if (!text) return {}
+  const position = cfg.position || "bottom"
+  const payload: Record<string, unknown> = {
+    text,
+    enabled: true,
+    font: cfg.font || "思源黑体",
+    font_size: Math.round(cfg.size) || 36,
+    font_color: cfg.color || "#ffffff",
+    position,
+    bold: !!cfg.bold,
+    stroke: cfg.stroke ? { enabled: true, width: 2, color: "#000000" } : { enabled: false },
+    shadow: cfg.shadow
+      ? { enabled: true, color: "#000000", offset_x: 2, offset_y: 2 }
+      : { enabled: false },
+  }
+  // 自定义坐标（custom 位置）
+  if (position === "custom" && typeof cfg.pos_x === "number" && typeof cfg.pos_y === "number") {
+    payload.pos_x = cfg.pos_x
+    payload.pos_y = cfg.pos_y
+  }
+  return payload
+}
+
+/* ── 封面：前端 state → 后端 render cover_config ── */
+export function buildCoverConfigPayload(
+  cfg: AiAvatarCoverConfig,
+  smartCoverUrl: string | null,
+): Record<string, unknown> {
+  const payload: Record<string, unknown> = {
+    enabled: !!cfg.enabled,
+    mode: cfg.mode,
+    // build_cover_extract_command 读取 timestamp（截帧秒数）
+    timestamp: cfg.frame_time || 0,
+  }
+  if (smartCoverUrl) payload.cover_url = smartCoverUrl
+  // 自定义上传：blob: 本地预览地址无法给后端，仅 OSS URL 可用
+  if (cfg.mode === "upload" && cfg.upload_url && !cfg.upload_url.startsWith("blob:")) {
+    payload.upload_url = cfg.upload_url
+  }
+  return payload
+}

apps/web/src/pages/ai-avatar/utils/sentences.ts

@@ -0,0 +1,62 @@
+/**
+ * AI数字人 — 文案分句 & B-roll 时间自动估算（#1809 ⑤⑥）
+ */
+
+export interface ScriptSentence {
+  /** 句子序号（从 0 开始，对应提交给后端的 script_segment_index） */
+  index: number
+  /** 句子文本（去掉首尾空白） */
+  text: string
+  /** 句子字数（按中文/字符计，去除空白） */
+  charCount: number
+  /** 累计起始字数（用于时间估算） */
+  startChar: number
+  /** 估算的对口型视频内起始时间（秒） */
+  startTime: number
+  /** 估算的对口型视频内结束时间（秒） */
+  endTime: number
+}
+
+/**
+ * 按句号/问号/感叹号/分号/换行分句（兼容中英文标点）。
+ * 空文案返回空数组。时间按「该句字数 ÷ 全文总字数 × 口播总时长」线性估算。
+ */
+export function splitScriptIntoSentences(
+  scriptText: string,
+  outputDuration: number,
+): ScriptSentence[] {
+  const text = (scriptText || "").trim()
+  if (!text) return []
+
+  const rawParts = text
+    .split(/[。！？!?；;\n\r]+/)
+    .map((part) => part.trim())
+    .filter((part) => part.length > 0)
+
+  const totalChars = rawParts.reduce((sum, part) => sum + part.replace(/\s/g, "").length, 0)
+  const duration = outputDuration > 0 ? outputDuration : 0
+
+  const sentences: ScriptSentence[] = []
+  let accChar = 0
+  rawParts.forEach((part, i) => {
+    const charCount = part.replace(/\s/g, "").length
+    const startTime = duration > 0 && totalChars > 0 ? (accChar / totalChars) * duration : 0
+    const endTime =
+      duration > 0 && totalChars > 0 ? ((accChar + charCount) / totalChars) * duration : 0
+    sentences.push({
+      index: i,
+      text: part,
+      charCount,
+      startChar: accChar,
+      startTime: round1(startTime),
+      endTime: round1(endTime),
+    })
+    accChar += charCount
+  })
+
+  return sentences
+}
+
+function round1(n: number): number {
+  return Math.round(n * 10) / 10
+}

apps/web/src/pages/assets/AssetLibrary.tsx

@@ -42,6 +42,7 @@ const AssetLibrary: React.FC = () => {
     assetsError,
     assetsErrorObj,
     refetchAssets,
+    stalledAssetIds,
     searchText,
     setSearchText,
     filterType,
@@ -77,6 +78,7 @@ const AssetLibrary: React.FC = () => {
     removeUpload,
     clearFinished,
     uploading,
+    transferActive,
     activeCount,
     pendingCount,
   } = useAssetUpload({ effectiveLibId })
@@ -168,6 +170,7 @@ const AssetLibrary: React.FC = () => {
           {/* 上传区域 */}
           <AssetUploadZone
             uploading={uploading}
+            transferActive={transferActive}
             activeCount={activeCount}
             pendingCount={pendingCount}
             onUpload={enqueueUploads}
@@ -214,6 +217,7 @@ const AssetLibrary: React.FC = () => {
             selectedIds={selectedIds}
             diagnosingId={diagnosingId}
             uploadProgressMap={uploadProgressMap}
+            stalledAssetIds={stalledAssetIds}
             onRetry={refetchAssets}
             onToggleSelect={toggleSelect}
             onDiagnose={handleDiagnose}