feat: 正式生成时片段随机重排（降重，默认开启无开关）#1663

在 _distribute_assets 方法中，smart_match 评分排序完成后、
distribute_assets 之前，对 asset_ids 做 random.shuffle。

- smart_match 决定选哪些素材（评分排序保留）
- shuffle 只改变最终分配到 clips 的顺序
- scene_points 缓存不受影响（shuffle 之前已读取）
- asset_ids 先 list() 复制再 shuffle，不修改调用方原列表
- 新增 3 个单元测试验证 shuffle 行为

Closes #1663

.ci-retigger-marker

@@ -0,0 +1 @@
+CI re-trigger after runner add-host/DNS fix. This file is harmless and not referenced.

.gitea/workflows/ci-base-image.yml

@@ -0,0 +1,105 @@
+name: CI Base Image Build
+
+on:
+  push:
+    branches:
+      - develop
+      - main
+    paths:
+      - 'requirements-base.txt'
+      - 'requirements-dev.txt'
+      - 'infra/docker/ci.Dockerfile'
+  workflow_dispatch:
+    inputs:
+      reason:
+        description: "触发原因"
+        required: false
+        default: "手动触发 - ci-base 镜像重建"
+
+concurrency:
+  group: ci-base-image-build
+  cancel-in-progress: false
+
+jobs:
+  build-ci-base:
+    name: Build CI Base Image
+    runs-on: runtime-builder
+    timeout-minutes: 60
+    steps:
+    - name: Checkout code
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+      run: |
+        curl -sH "Authorization: token $GITHUB_TOKEN" \
+          "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" \
+          | bash
+
+    - name: Docker login to Gitea Registry
+      shell: sh
+      env:
+        GITEA_REGISTRY_USER: xiaoxia
+        GITEA_REGISTRY_TOKEN: ${{ secrets.REGISTRY_TOKEN }}
+      run: |
+        set -eu
+        for i in 1 2 3; do
+          echo "=== Docker login 尝试 $i/3 ==="
+          if docker login git.xiaoxiajianji.com -u "${GITEA_REGISTRY_USER}" -p "${GITEA_REGISTRY_TOKEN}"; then
+            echo "✅ Docker login successful"
+            break
+          fi
+          echo "❌ Docker login 失败（尝试 $i/3），5s 后重试..."
+          sleep 5
+        done
+
+    - name: Build and push CI base image
+      shell: sh
+      run: |
+        set -eu
+        IMAGE="git.xiaoxiajianji.com/xiaoxia/xiaoxia-saas/ci-base"
+        VERSION_TAG="deps-$(date +%Y%m%d-%H%M)-${GITHUB_SHA::8}"
+
+        echo "=== Building CI base image (tags: latest, ${VERSION_TAG}) ==="
+        docker build --progress=plain \
+          -f infra/docker/ci.Dockerfile \
+          -t "${IMAGE}:latest" \
+          -t "${IMAGE}:${VERSION_TAG}" \
+          .
+        echo "✅ Image built successfully"
+
+        echo "=== Pushing ${VERSION_TAG} ==="
+        docker push "${IMAGE}:${VERSION_TAG}"
+        echo "=== Pushing latest ==="
+        docker push "${IMAGE}:latest"
+        echo "✅ Pushed to Gitea Registry"
+
+    - name: Verify image
+      shell: sh
+      run: |
+        set -eu
+        IMAGE="git.xiaoxiajianji.com/xiaoxia/xiaoxia-saas/ci-base:latest"
+        echo "=== Verifying pinned deps in fresh image ==="
+        docker run --rm "${IMAGE}" /opt/xiaoxia-ci-venv/bin/python -c \
+          "import httpcore, h2, numpy, httpx; print('VERSIONS:', httpcore.__version__, h2.__version__, numpy.__version__, httpx.__version__)"
+
+    - name: Notify result
+      if: always()
+      continue-on-error: true
+      shell: sh
+      env:
+        CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
+      run: |
+        set +e
+        if [ "${{ job.status }}" = "success" ]; then
+          NOTIFY_MODE=success JOB_NAME="CI Base Image Build" python3 scripts/ci_notify.py
+        else
+          NOTIFY_MODE=failure JOB_NAME="CI Base Image Build" python3 scripts/ci_notify.py
+        fi
+
+    - name: Cleanup
+      if: always()
+      shell: sh
+      run: |
+        IMAGE="git.xiaoxiajianji.com/xiaoxia/xiaoxia-saas/ci-base"
+        docker rmi "${IMAGE}:latest" 2>/dev/null || true
+        echo "Cleanup done"

.gitea/workflows/ci-canary.yml

@@ -0,0 +1,51 @@
+name: CI Canary Check
+on:
+  schedule:
+    - cron: '*/30 * * * *'
+  workflow_dispatch:
+
+jobs:
+  canary:
+    runs-on: ci-l2
+    timeout-minutes: 10
+    steps:
+      - name: Canary (runner -> docker -> network -> gitea)
+        run: |
+          set -e
+          echo "== runner/container basic =="
+          date; hostname; whoami
+          echo "== gitea api reachability =="
+          code=$(curl -s -o /tmp/v.json -w '%{http_code}' -m 15 "$GITHUB_API_URL/version")
+          echo "gitea api http_code=$code"
+          [ "$code" = "200" ] || { echo "::error::Gitea API unreachable, http_code=$code"; exit 1; }
+          cat /tmp/v.json; echo
+          echo "== external egress =="
+          ext=$(curl -s -o /dev/null -w '%{http_code}' -m 15 https://www.baidu.com || echo 000)
+          echo "external http_code=$ext"
+          echo "== gitea domain resolves NOT to loopback =="
+          set -o pipefail
+          ip=$(getent hosts git.xiaoxiajianji.com | awk '{print $1}' | head -1)
+          echo "git.xiaoxiajianji.com -> $ip"
+          if [ -z "$ip" ]; then
+            echo "::error::DNS resolution failed, git.xiaoxiajianji.com unresolvable"; exit 1
+          fi
+          if [ "$ip" = "127.0.0.1" ] || [ "$ip" = "::1" ]; then
+            echo "::error::Gitea domain resolves to loopback inside job container (hosts/DNS leak)"; exit 1
+          fi
+          echo "CANARY OK"
+      - name: Notify failure
+        if: failure()
+        env:
+          CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
+        run: |
+          set +e
+          if [ -n "$CI_NOTIFY_WEBHOOK" ]; then
+            MSG="🚨 CI 金丝雀失败：runner->docker->网络->Gitea 链路异常，时间 $(date '+%Y-%m-%d %H:%M:%S')，请立即检查构建服务器"
+            python3 - "$CI_NOTIFY_WEBHOOK" "$MSG" <<'PY'
+          import json,sys,urllib.request
+          hook,msg=sys.argv[1],sys.argv[2]
+          data=json.dumps({"msg_type":"text","content":{"text":msg}}).encode()
+          urllib.request.urlopen(urllib.request.Request(hook,data=data,headers={"Content-Type":"application/json"}),timeout=10)
+          PY
+          fi
+          exit 0

.gitea/workflows/ci-trigger-monitor.yml

@@ -2,13 +2,13 @@ name: CI Trigger Monitor
 
 on:
   schedule:
-    - cron: '*/5 * * * *'  # 每5分钟检查一次
+    - cron: '*/10 * * * *'  # 每10分钟检查一次（与pr-auto-scan同步降频）
   workflow_dispatch:
     inputs:
       stale_threshold:
         description: 'CI未触发告警阈值（分钟）'
         required: false
-        default: '5'
+        default: '10'
 
 permissions:
   contents: read

.gitea/workflows/playwright-base-image.yml

@@ -0,0 +1,59 @@
+name: Playwright Base Image Build
+
+on:
+  workflow_dispatch:
+    inputs:
+      reason:
+        description: "触发原因"
+        required: false
+        default: "构建 playwright 基础镜像"
+
+jobs:
+  build-playwright:
+    name: Build Playwright Base Image
+    runs-on: runtime-builder
+    timeout-minutes: 30
+    steps:
+    - name: Docker login to Gitea Registry
+      shell: sh
+      env:
+        GITEA_REGISTRY_USER: xiaoxia
+        GITEA_REGISTRY_TOKEN: ${{ secrets.REGISTRY_TOKEN }}
+      run: |
+        set -eu
+        for i in 1 2 3; do
+          echo "=== Docker login attempt $i/3 ==="
+          if printf '%s' "${GITEA_REGISTRY_TOKEN}" | docker login git.xiaoxiajianji.com -u "${GITEA_REGISTRY_USER}" --password-stdin; then
+            echo "Docker login successful"
+            break
+          fi
+          echo "Docker login failed (attempt $i/3), retrying in 5s..."
+          sleep 5
+          [ $i -eq 3 ] && exit 1
+        done
+
+    - name: Pull, retag and push Playwright image
+      shell: sh
+      run: |
+        set -eu
+        OFFICIAL_IMAGE="mcr.microsoft.com/playwright:v1.45.0-jammy"
+        GITEA_IMAGE="git.xiaoxiajianji.com/xiaoxia/base/playwright:v1.45.0-jammy"
+
+        echo "=== Pulling official Playwright image ==="
+        docker pull "${OFFICIAL_IMAGE}"
+
+        echo "=== Tagging ==="
+        docker tag "${OFFICIAL_IMAGE}" "${GITEA_IMAGE}"
+
+        echo "=== Pushing to Gitea Registry ==="
+        docker push "${GITEA_IMAGE}"
+
+        echo "Done: ${GITEA_IMAGE}"
+
+    - name: Cleanup
+      if: always()
+      shell: sh
+      run: |
+        docker rmi "mcr.microsoft.com/playwright:v1.45.0-jammy" 2>/dev/null || true
+        docker rmi "git.xiaoxiajianji.com/xiaoxia/base/playwright:v1.45.0-jammy" 2>/dev/null || true
+        echo "Cleanup done"

.gitea/workflows/pr-auto-scan.yml

@@ -3,7 +3,7 @@ name: PR Auto Scan
 # 作为短作业模式的兜底，防止事件驱动遗漏
 on:
   schedule:
-    - cron: "*/5 * * * *"  # 每5分钟扫描一次
+    # - cron: "*/15 * * * *"  # DISABLED: temporarily to stop failure spam (2026-09-02)  # 每10分钟扫描一次（脚本自带240s墙钟上限，降频减负）
   workflow_dispatch:
 
 permissions:

.gitea/workflows/pr-automation.yml

@@ -18,7 +18,7 @@ jobs:
     name: Auto Approve on CI Green
     runs-on: ci-check
     if: github.event_name == 'pull_request' && !github.event.pull_request.draft
-    timeout-minutes: 3  # 长等待模式：等CI全绿后自动合并，不遗漏任何PR
+    timeout-minutes: 10  # 等待CI全绿+审批，需要充足时间
     steps:
       - name: Checkout code
         shell: sh
@@ -61,7 +61,8 @@ jobs:
     name: Auto Merge on CI Green + Approved
     runs-on: ci-check
     if: github.event_name == 'pull_request' && !github.event.pull_request.draft && github.event.pull_request.base.ref == 'develop'
-    timeout-minutes: 3  # 短作业模式：检查一次，不满足就退出，由pr-auto-scan每5分钟定时兜底
+    needs: [auto-approve]  # 修复竞态：必须等审批完成后再尝试合并
+    timeout-minutes: 15  # 等待审批+CI就绪+合并，需要充足时间
     steps:
       - name: Checkout code
         shell: sh

.gitignore

@@ -24,6 +24,11 @@ ruff_cache/
 .env.production
 .env.staging
 !.env.example
+# 配置模板不受忽略规则限制
+!deploy/configs/.env.staging
+!deploy/configs/.env.production
+# 渲染后的 env 文件包含真实密钥，绝不能提交
+.env.rendered
 
 # OS / editor
 .DS_Store
@@ -54,3 +59,4 @@ frontend-v21-ui-prototype-final.html
 !.vscode/settings.json
 .vscode/extensions.json
 .coverage
+.env.current

alembic/versions/058_uq_asset_lib_project_kind.py

@@ -0,0 +1,49 @@
+"""Add unique index on asset_libraries(project_id, kind)
+
+Revision ID: 058_uq_asset_lib_project_kind
+Revises: 057_title_config
+Create Date: 2026-08-30
+
+同一项目下同 kind 的素材库业务上唯一（前端 getOrCreate 语义、TTS 保存自动建库）。
+加唯一索引兜底并发创建竞态，避免重复素材库。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "058_uq_asset_lib_project_kind"
+down_revision = "057_title_config"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 建唯一索引前清洗历史重复：同 (project_id, kind) 只保留 created_at 最新的一条。
+    # project_id 为 NULL 的系统级行不参与去重（NULL 在唯一索引中互不冲突）。
+    op.execute("""
+        DELETE FROM asset_libraries
+        WHERE id IN (
+            SELECT id FROM (
+                SELECT id,
+                       ROW_NUMBER() OVER (
+                           PARTITION BY project_id, kind
+                           ORDER BY created_at DESC, id DESC
+                       ) AS rn
+                FROM asset_libraries
+                WHERE project_id IS NOT NULL
+            ) t
+            WHERE t.rn > 1
+        )
+        """)
+    # 与 model 的 UniqueConstraint 定义保持一致（pg_constraint + pg_index 同时注册），
+    # 避免 Alembic autogenerate 检测到 schema drift
+    op.create_unique_constraint(
+        "uq_asset_libraries_project_kind",
+        "asset_libraries",
+        ["project_id", "kind"],
+    )
+
+
+def downgrade() -> None:
+    op.drop_constraint("uq_asset_libraries_project_kind", "asset_libraries", type_="unique")

alembic/versions/059_add_duplicate_rate_to_generated_videos.py

@@ -0,0 +1,23 @@
+"""add duplicate_rate to generated_videos
+
+Revision ID: 059_duplicate_rate
+Revises: 058_uq_asset_lib_project_kind
+Create Date: 2026-08-31
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "059_duplicate_rate"
+down_revision = "058_uq_asset_lib_project_kind"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("generated_videos", sa.Column("duplicate_rate", sa.Float(), nullable=True))
+
+
+def downgrade() -> None:
+    op.drop_column("generated_videos", "duplicate_rate")

alembic/versions/060_migrate_template_segments_to_clip_configs.py

@@ -0,0 +1,57 @@
+"""migrate template_segments data to template_clip_configs
+
+Revision ID: 060_migrate_segments
+Revises: 059_duplicate_rate
+Create Date: 2026-08-31
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "060_migrate_segments"
+down_revision = "059_duplicate_rate"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    dialect = op.get_bind().dialect.name
+
+    if dialect == "postgresql":
+        config_expr = (
+            "CASE WHEN s.material_type IS NOT NULL AND s.material_type != '' "
+            "THEN json_build_object('material_type', s.material_type)::jsonb "
+            "ELSE '{}'::jsonb END"
+        )
+        empty_json = "'{}'::jsonb"
+    else:
+        config_expr = (
+            "CASE WHEN s.material_type IS NOT NULL AND s.material_type != '' "
+            "THEN JSON_OBJECT('material_type', s.material_type) "
+            "ELSE '{}' END"
+        )
+        empty_json = "'{}'"
+
+    sql_str = (
+        "INSERT INTO template_clip_configs "
+        '(id, template_id, clip_type, "order", min_duration, max_duration, '
+        "text_template, material_requirements, transition_effect, config, "
+        "created_at, updated_at) "
+        "SELECT "
+        "s.id, s.template_id, 'main', s.segment_order, "
+        "s.duration_min, s.duration_max, "
+        "'', " + empty_json + ", "
+        "'cut', " + config_expr + ", "
+        "s.created_at, s.updated_at "
+        "FROM template_segments s "
+        "WHERE NOT EXISTS ("
+        "  SELECT 1 FROM template_clip_configs c "
+        "  WHERE c.template_id = s.template_id"
+        ")"
+    )
+    op.execute(sa.text(sql_str))
+
+
+def downgrade() -> None:
+    pass

alembic/versions/061_add_sort_order_to_template_categories.py

@@ -0,0 +1,26 @@
+"""add sort_order to template_categories
+
+Revision ID: 061_sort_order
+Revises: 060_migrate_segments
+Create Date: 2026-09-02
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "061_sort_order"
+down_revision = "060_migrate_segments"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "template_categories",
+        sa.Column("sort_order", sa.Integer, nullable=False, server_default="0"),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("template_categories", "sort_order")

alembic/versions/062_readd_edit_plan_id_to_generation_tasks.py

@@ -0,0 +1,28 @@
+"""re-add edit_plan_id to generation_tasks (align staging with production)
+
+Revision ID: 062_edit_plan_id
+Revises: 061_sort_order
+Create Date: 2026-09-02
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "062_edit_plan_id"
+down_revision = "061_sort_order"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "generation_tasks",
+        sa.Column("edit_plan_id", sa.String(36), nullable=True),
+    )
+    op.create_index("ix_generation_tasks_edit_plan_id_2", "generation_tasks", ["edit_plan_id"])
+
+
+def downgrade() -> None:
+    op.drop_index("ix_generation_tasks_edit_plan_id_2", table_name="generation_tasks")
+    op.drop_column("generation_tasks", "edit_plan_id")

apps/api/app/api/routes/assets.py

@@ -26,6 +26,7 @@ from app.schemas.asset import (
     UpdateAssetReviewRequest,
 )
 from app.schemas.tag import TagAssetsRequest
+from app.services.asset_segment_tracker import compute_asset_availability, get_asset_recent_use_counts
 from fastapi import APIRouter, Depends, HTTPException, Query, Response
 
 from packages.domain.smart_match import smart_select_assets
@@ -35,6 +36,23 @@ logger = logging.getLogger(__name__)
 router = APIRouter()
 
 
+def _asset_availability_fields(item) -> dict:
+    """视频素材返回余量四字段；非视频/无时长/异常时返回 None + usable=True（零影响）。"""
+    try:
+        info = compute_asset_availability(item)
+    except Exception:
+        logger.warning("计算素材余量失败，按可用处理: asset_id=%s", getattr(item, "id", "?"), exc_info=True)
+        info = None
+    if info is None:
+        return {
+            "used_duration": None,
+            "available_duration": None,
+            "used_ratio": None,
+            "usable": True,
+        }
+    return info
+
+
 def _to_asset_response(item, storage_service=None) -> AssetResponse:
     # 生成签名文件 URL（用于视频播放 / 文件下载）
     file_url = None
@@ -46,10 +64,16 @@ def _to_asset_response(item, storage_service=None) -> AssetResponse:
             logger.warning("生成签名URL失败: storage_key=%s", item.storage_key, exc_info=True)
             file_url = None
 
-    # 缩略图：优先用已有 thumbnail_url，否则对视频素材复用文件签名 URL
-    thumbnail_url = item.thumbnail_url
-    if not thumbnail_url and item.mime_type and item.mime_type.startswith("video") and file_url:
-        thumbnail_url = file_url
+    # 缩略图：存储的是 storage_key，需要生成签名 URL 供前端使用
+    # 不再降级使用视频文件 URL（浏览器 <img> 无法渲染 .mp4，会显示黑屏）
+    thumbnail_url = None
+    if item.thumbnail_url:
+        try:
+            svc = storage_service or get_storage_service()
+            thumbnail_url = svc.get_download_url(item.thumbnail_url)
+        except Exception:
+            logger.warning("生成缩略图签名URL失败: key=%s", item.thumbnail_url, exc_info=True)
+            thumbnail_url = None
 
     return AssetResponse(
         id=item.id,
@@ -73,6 +97,7 @@ def _to_asset_response(item, storage_service=None) -> AssetResponse:
         created_at=format_utc_datetime(item.created_at),
         uploaded_by_user_id=item.uploaded_by_user_id,
         tag_ids=getattr(item, "tag_ids", []),
+        **_asset_availability_fields(item),
     )
 
 
@@ -265,7 +290,7 @@ def list_assets(
         else:
             total = asset_repository.count_by_project_ids(project_ids, status=status_list)
             # 跨项目分页：逐项目累积直到凑够一页
-            paged_items: list = []
+            paged_items = []
             offset = skip
             remaining = limit
             for pid in project_ids:
@@ -551,21 +576,89 @@ def smart_match_assets(
             request.library_id, request.kind, status=["ready"], limit=10000
         )
     else:
-        filtered_assets = asset_repository.find_by_library(
-            request.library_id, status=["ready"], limit=10000
-        )
+        filtered_assets = asset_repository.find_by_library(request.library_id, status=["ready"], limit=10000)
     total_candidates = len(filtered_assets)
 
-    # 调用统一智能选素材算法（kind 已在 DB 层过滤，无需重复过滤）
-    results = smart_select_assets(
-        filtered_assets,
-        limit=request.limit,
-        kind=None,
-    )
+    # ── 过滤前置：余量 + 高频使用，过滤在评分/截取 limit 之前完成 ──────────
+    # 旧实现先 smart_select_assets(limit=N) 再对这 N 条做过滤，过滤后不回补，
+    # 当排名靠前的素材恰好都被排除时返回空 items（前端回退全选，smart-match 名存实亡）。
+    # 现在先过滤全量候选，每级过滤后为空/不足则回退上一级，最后才评分截取。
 
+    # 1) 余量过滤：usable=False（零重复可切区间耗尽且历史区间均达复用上限）的素材排除
+    usable_assets = []
+    exhausted_assets = []
+    for a in filtered_assets:
+        try:
+            avail = compute_asset_availability(a)
+        except Exception:
+            logger.warning(
+                "smart-match 余量计算失败，按可用处理: asset_id=%s",
+                getattr(a, "id", "?"),
+                exc_info=True,
+            )
+            avail = None
+        if avail is not None and not avail["usable"]:
+            exhausted_assets.append(a)
+        else:
+            usable_assets.append(a)
+
+    if exhausted_assets:
+        logger.info(
+            "smart-match 余量过滤: 候选 %d，可切区间耗尽 %d",
+            len(filtered_assets), len(exhausted_assets),
+        )
+
+    # 回退策略：余量过滤后为空（全部耗尽）时，保留全部候选，不返回空结果。
+    # 宁可让用户在已耗尽素材上复用，也比 smart-match 空结果回退全选更可控
+    # （全选同样会选到这些素材，且不经过评分排序）。
+    pool = usable_assets if usable_assets else filtered_assets
+
+    # 2) 高频使用排除：同一素材在最近 5 个视频中出现超过 3 次则排除
+    MAX_RECENT_USE_COUNT = 3
+    high_freq_assets = set()
+    if pool:
+        asset_ids = [getattr(a, "id", "") for a in pool if getattr(a, "id", "")]
+        if asset_ids:
+            try:
+                use_counts = get_asset_recent_use_counts(
+                    db=asset_repository.session,
+                    asset_ids=asset_ids,
+                    recent_video_count=5,
+                )
+                for a in pool:
+                    aid = getattr(a, "id", "")
+                    count = use_counts.get(aid, 0)
+                    if count > MAX_RECENT_USE_COUNT:
+                        high_freq_assets.add(aid)
+                        logger.info(
+                            "smart-match 排除高频使用素材: asset_id=%s use_count=%d limit=%d",
+                            aid, count, MAX_RECENT_USE_COUNT,
+                        )
+                # 回退策略：排除后剩余素材不足（为空或不够 limit）时，
+                # 不再全部排除，保留全部可用素材
+                if high_freq_assets:
+                    remaining_count = len(pool) - len(high_freq_assets)
+                    enough = request.limit is None or remaining_count >= request.limit
+                    if remaining_count > 0 and enough:
+                        pool = [a for a in pool if getattr(a, "id", "") not in high_freq_assets]
+                    else:
+                        logger.info(
+                            "smart-match 高频排除后素材不足(%d<%s)，保留全部 %d 条",
+                            remaining_count,
+                            request.limit if request.limit is not None else "不限",
+                            len(pool),
+                        )
+            except Exception:
+                logger.warning("smart-match 高频使用查询失败，跳过排除", exc_info=True)
+
+    # 3) 调用统一智能选素材算法（kind 已在 DB 层过滤，无需重复过滤）
+    results = smart_select_assets(pool, limit=request.limit, kind=None)
+
+    # 扁平结构：SmartMatchItem 继承 AssetResponse，素材字段直接在条目顶层，
+    # 前端无需解析 item.asset 包装层，item.id / item.usable / 余量字段直接可读
     items = [
         SmartMatchItem(
-            asset=_to_asset_response(r.asset),
+            **_to_asset_response(r.asset).model_dump(),
             score=r.score,
             breakdown=r.breakdown,
         )

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

@@ -456,7 +456,7 @@ async def wechat_callback(
     user = user_repository.find_by_id(response.user_id)
     binding_complete = False
     if user:
-        binding_complete = (
+        binding_complete = bool(
             user.phone_verified and user.email_verified and user.email and "@wechat.local" not in user.email
         )
 

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

@@ -1,15 +1,18 @@
 """封面生成路由 — Generation 模块.
 
 端点：
-- POST /generate-cover     AI 生成封面（从预览视频中抽帧）
+- POST /generate-cover     AI 生成封面（从最终成片视频中抽帧，兼容预览片段回退）
 
 挂载路径: /api/v1/generation/generate-cover
 """
 
 from __future__ import annotations
 
+import ipaddress
 import logging
+import re
 from typing import Any, List, Optional
+from urllib.parse import urlparse
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.dependencies import get_db_session, get_generated_video_repository
@@ -24,6 +27,7 @@ from packages.adapters.sqlalchemy_impl.generation_task_repository import (
 )
 from packages.application import ListGeneratedVideosByTaskUseCase
 from packages.domain.config_schemas import normalize_plan_config
+from packages.shared.storage import get_shared_storage_service
 
 from .templates_editor.dependencies import get_draft_plan_id, get_editor_services
 
@@ -51,6 +55,14 @@ class GenerateCoverRequest(BaseModel):
         default=None,
         description="上传的封面图片 URL，仅 cover_type=upload 时有效",
     )
+    generated_video_id: Optional[str] = Field(
+        default=None,
+        description="确认生成产出的最终视频 ID。传入后封面从该视频文件抽帧，而非预览片段。",
+    )
+    video_url: Optional[str] = Field(
+        default=None,
+        description="最终视频 URL（兜底）。当 generated_video_id 不可用时，直接从此 URL 对应的视频抽帧。",
+    )
 
 
 class GenerateCoverResponse(BaseModel):
@@ -121,8 +133,6 @@ def _persist_cover_frame(
                     exc_info=True,
                 )
 
-        from packages.shared.storage import get_shared_storage_service
-
         storage = get_shared_storage_service()
         cover_key = f"covers/{plan_id}/cover_{uuid.uuid4().hex[:8]}.jpg"
         storage.upload_file(
@@ -140,6 +150,106 @@ def _persist_cover_frame(
             Path(tmp_path).unlink(missing_ok=True)
 
 
+def _get_task_video_url(db: Session, task_id: str) -> Optional[str]:
+    """从 GenerationTask 关联的 GeneratedVideo 中获取视频 storage_key / URL."""
+    try:
+        video_repo = get_generated_video_repository(db)
+        use_case = ListGeneratedVideosByTaskUseCase(video_repo)
+        videos = use_case.execute(task_id)
+        if videos:
+            return getattr(videos[0], "file_url", "") or ""
+    except Exception:
+        logger.warning("[封面生成] 获取任务视频失败: task_id=%s", task_id, exc_info=True)
+    return None
+
+
+def _resolve_storage_key_to_url(storage_key: str) -> Optional[str]:
+    """将 storage_key 或完整 URL 转换为可访问的裸 URL。"""
+    if not storage_key:
+        return None
+    try:
+        if storage_key.startswith("http"):
+            url = storage_key
+        else:
+            storage_svc = get_shared_storage_service()
+            url = storage_svc.get_url(storage_key)
+        if url:
+            url = re.sub(r"(?<!:)//", "/", url)
+        return url
+    except Exception as e:
+        logger.warning("[封面生成] storage_key 转 URL 失败: key=%s err=%s", storage_key, e)
+        return None
+
+
+def _endpoint_host(value: str) -> str:
+    """从 endpoint / URL 字符串中安全提取主机名（兼容有无 scheme 两种配置）。"""
+    v = (value or "").strip().lower()
+    if not v:
+        return ""
+    if "://" in v:
+        return (urlparse(v).hostname or "").lower()
+    # 无 scheme：去掉可能的端口（host:port），urlparse 补 // 以正确解析
+    return (urlparse("//" + v).hostname or "").lower()
+
+
+def _is_private_or_reserved_host(host: str) -> bool:
+    """判断主机名是否为内网/回环/链路本地/保留地址（IPv4 与 IPv6 统一处理）。
+
+    使用标准库 ipaddress 判定；非 IP 主机名（如 localhost）单独处理。
+    """
+    h = host.strip().lower()
+    if h in {"localhost", "0.0.0.0", "::", "::1"}:
+        return True
+    try:
+        addr = ipaddress.ip_address(h)
+        # is_private 覆盖 10/8、172.16/12、192.168/16、127/8、169.254/16、
+        # ::1、fc00::/7、fe80::/10 等全部私有/保留段
+        return bool(addr.is_private or addr.is_loopback or addr.is_link_local or addr.is_reserved)
+    except ValueError:
+        return False
+
+
+def _is_trusted_media_url(url: str) -> bool:
+    """校验 URL 是否指向受信任的存储域名（OSS bucket / 本地存储），防止 SSRF。
+
+    用户可通过 video_url 传入视频地址，但服务端（MediaKit）会主动请求该 URL，
+    因此必须限制为自家存储域名，拒绝内网地址、元数据地址等任意主机。
+    """
+    if not url:
+        return False
+    try:
+        parsed = urlparse(url.strip())
+        if parsed.scheme not in ("http", "https"):
+            return False
+        host = (parsed.hostname or "").lower()
+        if not host:
+            return False
+        # 拒绝一切内网/回环/链路本地/保留地址（IPv4 + IPv6，标准库判定）
+        if _is_private_or_reserved_host(host):
+            return False
+        # 允许：自家 OSS bucket 域名（<bucket>.<endpoint>）或 endpoint 自身及其子域
+        try:
+            storage_svc = get_shared_storage_service()
+            trusted_hosts = set()
+            public_base = getattr(storage_svc, "public_url", "") or ""
+            h1 = _endpoint_host(public_base)
+            if h1:
+                trusted_hosts.add(h1)
+            h2 = _endpoint_host(getattr(storage_svc, "endpoint", "") or "")
+            if h2:
+                trusted_hosts.add(h2)
+            for trusted in trusted_hosts:
+                if host == trusted or host.endswith("." + trusted):
+                    return True
+        except Exception:
+            logger.warning("[封面生成] 存储域名白名单初始化失败，URL 校验从严拒绝", exc_info=True)
+            return False
+        return False
+    except Exception:
+        logger.warning("[封面生成] video_url 白名单校验异常，从严拒绝: url=%s", url[:80], exc_info=True)
+        return False
+
+
 @router.post("/generate-cover", response_model=GenerateCoverResponse)
 def generate_cover(
     body: GenerateCoverRequest,
@@ -149,12 +259,16 @@ def generate_cover(
     db: Session = Depends(get_db_session),
     current_user: AuthenticatedUser = Depends(get_current_user),
 ) -> GenerateCoverResponse:
-    """AI 生成封面 — 从预览视频中抽帧.
+    """AI 生成封面 — 优先从最终成片视频中抽帧，回退到预览片段.
 
     流程（串行）：
-    1. 预览视频已渲染完成（通过 3 步查找获取 URL）
-    2. 用裸 URL 让 MediaKit 下载视频并抽帧
-    3. 帧图下载后上传到 OSS covers/ 路径
+    1. 优先使用前端传入的 generation_task_id 定位最终成片任务，
+       或自动查找 plan 关联的已完成最终成片任务（is_preview=False）
+    2. 回退：从预览片段获取视频 URL（兼容旧流程）
+    3. 用裸 URL 让 MediaKit 下载视频并抽帧
+    4. 帧图下载后上传到 OSS covers/ 路径
+
+    MediaKit 的调用方式（strategy / max_frames / 轮询 / 重试 / 降级）不变。
     """
     _, plan_svc = services
     plan = plan_svc.get_plan_or_raise(plan_id)
@@ -182,27 +296,111 @@ def generate_cover(
         )
         return GenerateCoverResponse(plan_id=plan_id, cover=cover_data)
 
-    # ── 3 步查找预览视频 URL ──────────────────────────────────────────
-    # 第一步：从 plan.config 读取
+    # ── 查找用于抽帧的视频 URL ────────────────────────────────────────
+    # 优先级：
+    #   0. 请求体显式传入的 generation_task_id（最终成片任务）
+    #   1. plan.config.rendered_storage_key
+    #   2. plan.config.generation_task_id 对应的任务
+    #   3. source_edit_plan_id 关联的已完成「最终成片」任务（is_preview=False）
+    #   4. source_edit_plan_id 关联的已完成预览任务（is_preview=True，兼容回退）
+    #   5. user + template 最近的已完成预览任务（兜底）
     logger.info("[封面生成] 步骤1: 从 plan.config 查找 rendered_storage_key: plan_id=%s", plan_id)
     rendered_storage_key = (plan.config or {}).get("rendered_storage_key", "")
 
-    # 第二步：如果还没有，通过 generation_task_id 查找预览任务的产物
+    # 步骤 0：请求体传入最终视频标识（generated_video_id 或 video_url）
+    if not rendered_storage_key:
+        # 0a：通过 generated_video_id 查找最终成片视频
+        if body.generated_video_id:
+            logger.info(
+                "[封面生成] 步骤0a: 使用 generated_video_id: plan_id=%s video_id=%s",
+                plan_id,
+                body.generated_video_id,
+            )
+            try:
+                gv_repo = get_generated_video_repository(db)
+                gv = gv_repo.get(body.generated_video_id)
+                if gv:
+                    file_url = getattr(gv, "file_url", "") or ""
+                    if file_url:
+                        # 权限校验（双重，任何一层确认归属不符即拒绝）：
+                        # 1) GeneratedVideo.user_id 直接归属（老数据可能为空，为空时不据此放行）
+                        gv_owner = (getattr(gv, "user_id", "") or "").strip()
+                        if gv_owner and gv_owner != current_user.user.id:
+                            raise HTTPException(status_code=403, detail="无权访问该视频")
+                        # 2) 关联 generation_task 归属校验；关联任务缺失时不可静默放行：
+                        #    若 video 自身无 owner 信息且关联任务也查不到，拒绝访问
+                        gv_task_id = getattr(gv, "generation_task_id", "") or ""
+                        task0 = None
+                        if gv_task_id:
+                            try:
+                                task0 = SQLAlchemyGenerationTaskRepository(db).get(gv_task_id)
+                            except Exception:
+                                logger.warning(
+                                    "[封面生成] 步骤0a关联任务查询异常: plan_id=%s task_id=%s",
+                                    plan_id,
+                                    gv_task_id,
+                                    exc_info=True,
+                                )
+                        if task0 is not None:
+                            task_owner = (getattr(task0, "created_by_user_id", "") or "").strip()
+                            if task_owner and task_owner != current_user.user.id:
+                                raise HTTPException(status_code=403, detail="无权访问该视频")
+                        elif not gv_owner:
+                            # video 无 owner 且关联任务不存在/无法确认归属 → 拒绝，防止越权
+                            logger.warning(
+                                "[封面生成] 步骤0a视频归属无法确认，拒绝访问: plan_id=%s video_id=%s",
+                                plan_id,
+                                body.generated_video_id,
+                            )
+                            raise HTTPException(status_code=403, detail="无权访问该视频")
+                        rendered_storage_key = file_url
+                        logger.info(
+                            "[封面生成] ✅ 步骤0a找到最终成片: plan_id=%s video_id=%s url=%s",
+                            plan_id,
+                            body.generated_video_id,
+                            file_url[:80],
+                        )
+            except HTTPException:
+                raise
+            except Exception:
+                logger.warning(
+                    "[封面生成] 步骤0a查找视频失败: plan_id=%s video_id=%s",
+                    plan_id,
+                    body.generated_video_id,
+                    exc_info=True,
+                )
+
+        # 0b：直接使用 video_url（兜底）— 必须通过存储域名白名单校验，防止 SSRF
+        if not rendered_storage_key and body.video_url:
+            if _is_trusted_media_url(body.video_url):
+                logger.info(
+                    "[封面生成] 步骤0b: 使用请求体传入的 video_url(白名单通过): plan_id=%s url=%s",
+                    plan_id,
+                    body.video_url[:80],
+                )
+                rendered_storage_key = body.video_url
+            else:
+                logger.warning(
+                    "[封面生成] 步骤0b: video_url 不在受信任存储域名白名单内，已忽略: plan_id=%s url=%s",
+                    plan_id,
+                    body.video_url[:80],
+                )
+
+    # 步骤 2：通过 plan.config.generation_task_id 查找
     if not rendered_storage_key:
         generation_task_id = (plan.config or {}).get("generation_task_id", "")
-        logger.info(
-            "[封面生成] 步骤2: 通过 generation_task_id 查找: plan_id=%s task_id=%s", plan_id, generation_task_id
-        )
         if generation_task_id:
+            logger.info(
+                "[封面生成] 步骤2: 通过 plan.config.generation_task_id 查找: plan_id=%s task_id=%s",
+                plan_id,
+                generation_task_id,
+            )
             try:
-                gen_task_repo = SQLAlchemyGenerationTaskRepository(db)
-                task = gen_task_repo.get(generation_task_id)
+                _repo = SQLAlchemyGenerationTaskRepository(db)
+                task = _repo.get(generation_task_id)
                 if task:
-                    video_repo = get_generated_video_repository(db)
-                    use_case = ListGeneratedVideosByTaskUseCase(video_repo)
-                    videos = use_case.execute(task.id)
-                    if videos:
-                        rendered_storage_key = getattr(videos[0], "file_url", "") or ""
+                    rendered_storage_key = _get_task_video_url(db, task.id) or ""
+                    if rendered_storage_key:
                         logger.info(
                             "[封面生成] ✅ 步骤2找到视频: plan_id=%s task_id=%s url=%s",
                             plan_id,
@@ -211,26 +409,23 @@ def generate_cover(
                         )
             except Exception:
                 logger.warning(
-                    "封面生成: 通过 generation_task_id 查找视频失败: plan_id=%s",
+                    "[封面生成] 步骤2查找失败: plan_id=%s",
                     plan_id,
                     exc_info=True,
                 )
 
-    # 第 2.5 步：通过 plan_id 作为 source_edit_plan_id 查找关联的已完成预览任务
+    # 步骤 3：通过 source_edit_plan_id 查找已完成「最终成片」任务（is_preview=False）
     if not rendered_storage_key:
         try:
-            gen_task_repo = SQLAlchemyGenerationTaskRepository(db)
-            logger.info("[封面生成] 步骤2.5: 通过 source_edit_plan_id 查找: plan_id=%s", plan_id)
-            preview_tasks = gen_task_repo.list_by_source_edit_plan(plan_id)
-            for pt in preview_tasks:
-                if getattr(pt, "status", "") == "completed" and getattr(pt, "is_preview", False):
-                    video_repo = get_generated_video_repository(db)
-                    use_case = ListGeneratedVideosByTaskUseCase(video_repo)
-                    videos = use_case.execute(pt.id)
-                    if videos:
-                        rendered_storage_key = getattr(videos[0], "file_url", "") or ""
+            _repo = SQLAlchemyGenerationTaskRepository(db)
+            logger.info("[封面生成] 步骤3: 查找最终成片任务(is_preview=False): plan_id=%s", plan_id)
+            all_tasks = _repo.list_by_source_edit_plan(plan_id)
+            for pt in all_tasks:
+                if getattr(pt, "status", "") == "completed" and not getattr(pt, "is_preview", False):
+                    rendered_storage_key = _get_task_video_url(db, pt.id) or ""
+                    if rendered_storage_key:
                         logger.info(
-                            "[封面生成] ✅ 步骤2.5找到视频: plan_id=%s task_id=%s url=%s",
+                            "[封面生成] ✅ 步骤3找到最终成片: plan_id=%s task_id=%s url=%s",
                             plan_id,
                             pt.id,
                             rendered_storage_key[:80],
@@ -238,66 +433,74 @@ def generate_cover(
                         break
         except Exception:
             logger.warning(
-                "封面生成: 通过 source_edit_plan_id 查找预览任务失败: plan_id=%s",
+                "[封面生成] 步骤3查找最终成片失败: plan_id=%s",
                 plan_id,
                 exc_info=True,
             )
 
-    # 第三步：按 user + template 查找最近的已完成预览任务（兜底）
+    # 步骤 4：兼容回退 — 通过 source_edit_plan_id 查找已完成预览任务
     if not rendered_storage_key:
         try:
-            gen_task_repo = SQLAlchemyGenerationTaskRepository(db)
-            logger.info("[封面生成] 步骤3: 通过 user+template 查找: plan_id=%s template_id=%s", plan_id, template_id)
-            preview_tasks = gen_task_repo.list_latest_completed_preview(
+            _repo = SQLAlchemyGenerationTaskRepository(db)
+            logger.info("[封面生成] 步骤4: 回退查找预览任务(is_preview=True): plan_id=%s", plan_id)
+            preview_tasks = _repo.list_by_source_edit_plan(plan_id)
+            for pt in preview_tasks:
+                if getattr(pt, "status", "") == "completed" and getattr(pt, "is_preview", False):
+                    rendered_storage_key = _get_task_video_url(db, pt.id) or ""
+                    if rendered_storage_key:
+                        logger.info(
+                            "[封面生成] ✅ 步骤4找到预览视频: plan_id=%s task_id=%s url=%s",
+                            plan_id,
+                            pt.id,
+                            rendered_storage_key[:80],
+                        )
+                        break
+        except Exception:
+            logger.warning(
+                "[封面生成] 步骤4查找预览任务失败: plan_id=%s",
+                plan_id,
+                exc_info=True,
+            )
+
+    # 步骤 5：按 user + template 查找最近的已完成预览任务（兜底）
+    if not rendered_storage_key:
+        try:
+            _repo = SQLAlchemyGenerationTaskRepository(db)
+            logger.info(
+                "[封面生成] 步骤5: 通过 user+template 查找预览任务: plan_id=%s template_id=%s",
+                plan_id,
+                template_id,
+            )
+            preview_tasks = _repo.list_latest_completed_preview(
                 user_id=str(current_user.user.id),
                 template_id=template_id,
             )
             if preview_tasks:
-                completed_preview = preview_tasks[0]
-                video_repo = get_generated_video_repository(db)
-                use_case = ListGeneratedVideosByTaskUseCase(video_repo)
-                videos = use_case.execute(completed_preview.id)
-                if videos:
-                    rendered_storage_key = getattr(videos[0], "file_url", "") or ""
+                rendered_storage_key = _get_task_video_url(db, preview_tasks[0].id) or ""
+                if rendered_storage_key:
                     logger.info(
-                        "封面视频: 通过 user+template 找到预览任务: plan_id=%s template_id=%s task_id=%s",
+                        "[封面生成] ✅ 步骤5找到预览视频: plan_id=%s task_id=%s",
                         plan_id,
-                        template_id,
-                        completed_preview.id,
+                        preview_tasks[0].id,
                     )
         except Exception:
             logger.warning(
-                "封面警告: user+template 查找预览任务失败: plan_id=%s template_id=%s",
+                "[封面生成] 步骤5 user+template 查找失败: plan_id=%s",
                 plan_id,
-                template_id,
                 exc_info=True,
             )
 
-    # 使用裸 URL（rendered/* 已配置公开读）；找不到渲染视频时不立即报错，
-    # 因为步骤 E 可以直接从源素材抽帧（历史数据或 Worker 抽帧失败时的兜底）
+    # 将 storage_key 转换为可访问 URL；找不到视频时不立即报错，
+    # 因为步骤 E2 可以直接从源素材抽帧（历史数据或 Worker 抽帧失败时的兜底）
     primary_video_url = None
     if rendered_storage_key:
         plan_svc.update_plan_config(plan_id, {"rendered_storage_key": rendered_storage_key})
-        try:
-            if rendered_storage_key.startswith("http"):
-                primary_video_url = rendered_storage_key
-            else:
-                from packages.shared.storage import get_shared_storage_service
-
-                storage_svc = get_shared_storage_service()
-                primary_video_url = storage_svc.get_url(rendered_storage_key)
-            if primary_video_url:
-                import re as _re
-
-                primary_video_url = _re.sub(r"(?<!:)//", "/", primary_video_url)
-            logger.info(
-                "获取预览视频URL用于封面生成: plan_id=%s url=%s",
-                plan_id,
-                primary_video_url[:80] if primary_video_url else "",
-            )
-        except Exception as e:
-            logger.warning("获取预览视频URL失败: plan_id=%s err=%s", plan_id, e)
-            primary_video_url = None
+        primary_video_url = _resolve_storage_key_to_url(rendered_storage_key)
+        logger.info(
+            "[封面生成] 封面抽帧视频URL: plan_id=%s url=%s",
+            plan_id,
+            primary_video_url[:80] if primary_video_url else "",
+        )
 
     # 统一封面管道：优先从 GenerationTask.cover_url 读取渲染后视频抽帧的封面
     # 多步查找 cover_url，和查找视频 URL 一样的 fallback 逻辑
@@ -310,7 +513,7 @@ def generate_cover(
         if generation_task_id:
             try:
                 task = gen_task_repo.get(generation_task_id)
-                if task and getattr(task, "cover_url", ""):
+                if task and getattr(task, "cover_url", ""):  # type: ignore[arg-type]
                     cover_url_from_task = task.cover_url
                     logger.info(
                         "[封面生成] 统一管道封面(步骤A-direct): plan_id=%s task_id=%s url=%s",
@@ -326,20 +529,67 @@ def generate_cover(
                     exc_info=True,
                 )
 
-        # 步骤 B：通过 source_edit_plan_id 查找关联预览任务的 cover_url
+        # 步骤 A2：通过 generated_video_id 查找其关联任务的 cover_url
+        if not cover_url_from_task and body.generated_video_id:
+            try:
+                gv_repo = get_generated_video_repository(db)
+                gv = gv_repo.get(body.generated_video_id)
+                if gv:
+                    gv_task_id = getattr(gv, "generation_task_id", "") or ""
+                    if gv_task_id:
+                        task_a2 = gen_task_repo.get(gv_task_id)
+                        if task_a2 and getattr(task_a2, "cover_url", ""):  # type: ignore[arg-type]
+                            cover_url_from_task = task_a2.cover_url
+                            logger.info(
+                                "[封面生成] 封面(步骤A2-video-task): plan_id=%s video_id=%s url=%s",
+                                plan_id,
+                                body.generated_video_id,
+                                cover_url_from_task[:80],
+                            )
+            except Exception:
+                logger.warning(
+                    "[封面生成] 步骤A2读取 cover_url 失败: plan_id=%s video_id=%s",
+                    plan_id,
+                    body.generated_video_id,
+                    exc_info=True,
+                )
+
+        # 步骤 B：通过 source_edit_plan_id 查找关联任务的 cover_url
+        # 优先最终成片任务(is_preview=False)，其次预览任务
         if not cover_url_from_task:
             try:
-                preview_tasks = gen_task_repo.list_by_source_edit_plan(plan_id)
-                for pt in preview_tasks:
-                    if getattr(pt, "status", "") == "completed" and getattr(pt, "cover_url", ""):
+                all_tasks = gen_task_repo.list_by_source_edit_plan(plan_id)
+                # 先找最终成片
+                for pt in all_tasks:
+                    if (
+                        getattr(pt, "status", "") == "completed"
+                        and not getattr(pt, "is_preview", False)
+                        and getattr(pt, "cover_url", "")
+                    ):
                         cover_url_from_task = pt.cover_url
                         logger.info(
-                            "[封面生成] 统一管道封面(步骤B-source_plan): plan_id=%s task_id=%s url=%s",
+                            "[封面生成] 封面(步骤B-final): plan_id=%s task_id=%s url=%s",
                             plan_id,
                             pt.id,
                             cover_url_from_task[:80],
                         )
                         break
+                # 再找预览
+                if not cover_url_from_task:
+                    for pt in all_tasks:
+                        if (
+                            getattr(pt, "status", "") == "completed"
+                            and getattr(pt, "is_preview", False)
+                            and getattr(pt, "cover_url", "")
+                        ):
+                            cover_url_from_task = pt.cover_url
+                            logger.info(
+                                "[封面生成] 封面(步骤B-preview): plan_id=%s task_id=%s url=%s",
+                                plan_id,
+                                pt.id,
+                                cover_url_from_task[:80],
+                            )
+                            break
             except Exception:
                 logger.warning(
                     "[封面生成] 步骤B查找 cover_url 失败: plan_id=%s",
@@ -497,7 +747,7 @@ def generate_cover(
 
         if cover_url_from_task:
             # 标题已在预览视频渲染时烧录（ASS字幕），封面帧自然包含标题
-            cover_data = {
+            cover_data: dict[str, object] = {  # type: ignore[no-redef]
                 "type": "ai_frame",
                 "image_url": cover_url_from_task,
                 "frame_time": 0.0,

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

@@ -98,7 +98,7 @@ def _resolve_strategy_id_from_template(template_id: str, db: Session, user_id: s
     try:
         new_repo = SQLAlchemyEditTemplateRepository(db)
         new_template = new_repo.get(template_id)
-        if new_template and getattr(new_template, "editing_mode", ""):
+        if new_template and getattr(new_template, "editing_mode", ""):  # type: ignore[arg-type]
             mode = new_template.editing_mode.strip()
             if mode:
                 logger.info(
@@ -358,6 +358,41 @@ def create_preview_generation_task(
                 exc_info=True,
             )
 
+    # 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:
+        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,
+                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")
+            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
+
     # 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(

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

@@ -1,5 +1,4 @@
 import logging
-import random
 import uuid
 from typing import Any
 
@@ -116,8 +115,8 @@ def _select_assets_from_library(
 
     Args:
         assets: 素材库中所有素材（Asset 实体列表）
-        mode: 选取模式 — all=全部, random=随机, smart=智能匹配（多维度评分+多样性）
-        count: 选取数量，0 表示全部（仅 random/smart 模式有效）
+        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）
 
     Returns:
         选中的素材 ID 列表
@@ -127,12 +126,6 @@ def _select_assets_from_library(
     if not ready_video_assets:
         return []
 
-    if mode == "random":
-        selected = (
-            ready_video_assets if count <= 0 else random.sample(ready_video_assets, min(count, len(ready_video_assets)))
-        )
-        return [a.id for a in selected]
-
     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%)
@@ -169,8 +162,22 @@ def _writeback_edit_plan_config(
         current_config = plan_model.config if isinstance(plan_model.config, dict) else {}
         merged = dict(current_config)
         merged["generation_task_id"] = task_id
+        
+        # 检查标题是否发生变化，如果变化则清除 cover 字段强制重新生成封面
         if title_config:
+            old_title_config = merged.get("title_config", {}) or {}
+            old_title_text = (old_title_config.get("text") or "").strip()
+            new_title_text = (title_config.get("text") or "").strip()
+            if old_title_text != new_title_text:
+                # 标题变化，清除旧封面
+                if "cover" in merged:
+                    del merged["cover"]
+                    logger.info(
+                        "[生成任务] 标题变化，清除旧封面: plan_id=%s old_title=%s new_title=%s",
+                        plan_id, old_title_text, new_title_text,
+                    )
             merged["title_config"] = title_config
+        
         plan_model.config = merged
         db.commit()
         logger.info(
@@ -278,8 +285,8 @@ def create_generation_task(
                 mode=request.asset_select_mode,
                 count=request.asset_select_count,
             )
-    elif project_id and not resolved_asset_ids and request.asset_select_mode in ("random", "smart"):
-        # 项目级模式：未指定 asset_ids 且选择了 random/smart 模式时，也自动选取
+    elif project_id and not resolved_asset_ids and request.asset_select_mode in ("smart",):
+        # 项目级模式：未指定 asset_ids 且选择了 smart 模式时，也自动选取
         assets = asset_repository.find_by_project(project_id)
         if assets:
             resolved_asset_ids = _select_assets_from_library(
@@ -378,7 +385,7 @@ def create_generation_task(
 
     use_case = CreateGenerationTaskUseCase(generation_task_repository)
     count = request.count
-    created_tasks = []
+    created_tasks: list = []
     failed_tasks = []
     user_id = authenticated_user.user.id
     # 同批次任务共享 batch_id，用于视频查重时批次内比对
@@ -411,8 +418,56 @@ 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）。
+    variant_plan_ids: list[str] = []
+    if count > 1 and request.source_edit_plan_id:
+        from app.services.edit_plan_service import EditPlanService
+
+        _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}",
+                    )
+                    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,
+                )
+                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)
+
     try:
-        for _ in range(count):
+        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]
+
             task = use_case.execute(
                 CreateGenerationTaskCommand(
                     project_id=project_id,
@@ -424,7 +479,7 @@ def create_generation_task(
                     title_ids=request.title_ids,
                     voice_ids=request.voice_ids,
                     created_by_user_id=user_id,
-                    source_edit_plan_id=request.source_edit_plan_id,
+                    source_edit_plan_id=effective_plan_id,
                     asset_select_mode=request.asset_select_mode,
                     batch_id=batch_id,
                     video_title=request.video_title,

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

@@ -3,7 +3,7 @@ from typing import Any
 from app.core.celery_app import celery_app
 from app.dependencies import get_ingest_job_repository
 from app.schemas.ingest_job import IngestJobResponse, SubmitIngestJobRequest
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter, Depends, HTTPException
 
 from packages.application import SubmitIngestJobCommand, SubmitIngestJobUseCase
 
@@ -17,7 +17,7 @@ def get_ingest_job(
 ) -> IngestJobResponse:
     job = ingest_job_repository.get(job_id)
     if job is None:
-        raise ValueError(f"IngestJob {job_id} not found")
+        raise HTTPException(status_code=404, detail=f"IngestJob {job_id} not found")
     return IngestJobResponse(
         id=job.id,
         project_id=job.project_id,

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

@@ -15,16 +15,47 @@
 
 from __future__ import annotations
 
+import json
 import logging
+import random
+import re
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.storage import get_storage_service
-from app.dependencies import get_asset_repository
+from app.dependencies import get_asset_repository, get_db_session
+
+# 默认转场时长（与 worker 端保持一致）
+_DEFAULT_TRANSITION_DURATION = 0.5
+
+from app.services.asset_segment_tracker import (
+    REUSE_RATIO_LIMIT,
+    SEGMENT_EDGE_GAP,
+    get_used_segments,
+    make_reuse_callback,
+    record_used_segments,
+    remove_used_segment,
+)
 from app.services.edit_plan_service import EditPlanService
 from app.services.edit_template_service import EditTemplateService
-from fastapi import APIRouter, Depends, HTTPException, Query, status
+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,
+    extract_scene_points_from_metadata,
+    pick_scene_aware_start,
+    pick_start_in_scene_segment,
+)
+from packages.domain.smart_match import SCORE_RANDOM_NOISE_MAX, score_asset
+from packages.shared.mediakit_client import get_mediakit_client
 
 from .dependencies import get_draft_plan_id, get_editor_services
 from .schemas import (
@@ -45,6 +76,9 @@ from .schemas import (
 logger = logging.getLogger(__name__)
 router = APIRouter(tags=["Template Editor"])
 
+# 编辑器默认片段时长（秒）
+_DEFAULT_EDITOR_CLIP_DURATION = 5.0
+
 
 def _clip_to_response(clip, asset_url: str | None = None) -> EditorClipResponse:
     """统一构造片段响应 — 与 edit_plan_clips 表字段完全对齐"""
@@ -155,10 +189,7 @@ def list_draft_clips(
     url_map = _build_asset_url_map(asset_ids, asset_repo)
 
     return EditorClipListResponse(
-        items=[
-            _clip_to_response(c, asset_url=url_map.get(getattr(c, "asset_id", "") or ""))
-            for c in clips
-        ],
+        items=[_clip_to_response(c, asset_url=url_map.get(getattr(c, "asset_id", "") or "")) for c in clips],
         total=total,
     )
 
@@ -272,9 +303,7 @@ def split_draft_clip(
     try:
         result = plan_svc.split_clip(clip_id, body.split_time)
     except ValueError as exc:
-        raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST, detail=str(exc)
-        ) from exc
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=str(exc)) from exc
     left = result["left_clip"]
     right = result["right_clip"]
     asset_ids = [getattr(left, "asset_id", "") or "", getattr(right, "asset_id", "") or ""]
@@ -304,9 +333,7 @@ def merge_draft_clips(
     try:
         merged = plan_svc.merge_clips(body.clip_ids)
     except ValueError as exc:
-        raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST, detail=str(exc)
-        ) from exc
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=str(exc)) from exc
     asset_id = getattr(merged, "asset_id", "") or ""
     url_map = _build_asset_url_map([asset_id], asset_repo) if asset_id else {}
     return {
@@ -354,40 +381,814 @@ def batch_delete_editor_clips(
     return ClipBatchDeleteResponse(deleted_count=deleted, plan_id=plan_id)
 
 
+def _safe_segment_duration(value, default: float) -> float:
+    """安全地将数据库中的时长值转换为正浮点数.
+
+    处理 None、无效类型、负数、NaN 等异常情况。
+    """
+    if value is None:
+        return default
+    try:
+        result = float(value)
+    except (ValueError, TypeError):
+        return default
+    if result != result or result <= 0:  # NaN check or non-positive
+        return default
+    return result
+
+
+def _get_template_segments(
+    template_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）。
+
+    Returns:
+        [(segment_order, duration_min, duration_max), ...] 按 order 排序
+    """
+    # 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)
+
+    # 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 []
+
+
+def _recommended_time_conflicts(
+    start: float,
+    duration: float,
+    used: list[tuple[float, float]],
+    edge_gap: float = SEGMENT_EDGE_GAP,
+) -> bool:
+    """检查推荐起始时间是否与已使用时间段冲突.
+
+    冲突检测统一加 ``edge_gap`` 秒边缘间隙：已用区间按 [s-gap, e+gap] 扩边后判定，
+    避免推荐片段与已用片段首尾紧贴导致画面观感重复。
+    """
+    end = start + duration
+    for used_start, used_end in used:
+        if start < used_end + edge_gap and end > used_start - edge_gap:
+            return True
+    return False
+
+
+# 向后兼容别名：镜头段构建/段内取点逻辑已下沉到 packages.domain.plan_generator_utils，
+# 旧测试与历史代码仍按 clips._build_scene_segments / _pick_start_in_scene_segment 导入
+_build_scene_segments = build_scene_segments
+_pick_start_in_scene_segment = pick_start_in_scene_segment
+
+
+def _get_mediakit_recommendations(
+    asset_ids: list[str],
+    asset_repo,
+) -> dict[str, float]:
+    """调用 MediaKit 视频理解，获取智能选片推荐起始时间.
+
+    尝试让 MediaKit 分析视频内容，返回每个素材的推荐起始时间。
+    任何异常都优雅降级，返回空字典（调用方降级到随机选择）。
+    """
+    try:
+        client = get_mediakit_client()
+        if not client.is_available:
+            logger.info("MediaKit 未配置，使用随机起始时间")
+            return {}
+
+        storage = get_storage_service()
+
+        video_urls: list[str] = []
+        valid_asset_ids: list[str] = []
+        for asset_id in asset_ids[:10]:
+            asset = asset_repo.get(asset_id)
+            if not asset or not getattr(asset, "storage_key", None):
+                continue
+            mime = getattr(asset, "mime_type", "")
+            if not mime.startswith("video/"):
+                continue
+            try:
+                url = storage.get_download_url(asset.storage_key)
+                if url:
+                    video_urls.append(url)
+                    valid_asset_ids.append(asset_id)
+            except Exception as e:
+                logger.warning("获取素材URL失败: asset_id=%s error=%s", asset_id, e)
+
+        if not video_urls:
+            return {}
+
+        prompt = (
+            "请分析每段视频，找出最精彩的5秒片段应该从哪个时间点开始。"
+            "考虑因素：画面清晰度、主体是否明确、是否有明显的动作或场景变化。"
+            "请严格以JSON数组格式返回，不要包含其他文字："
+            '[{"asset_id": "素材ID", "recommended_start_time": 12.5, "reason": "原因"}]'
+        )
+
+        contents = client.analyze_videos(
+            video_urls=video_urls,
+            prompt=prompt,
+            level="Economy",
+            poll_interval=2.0,
+            max_poll_attempts=15,
+        )
+
+        if not contents:
+            logger.info("MediaKit 分析无结果，降级为随机选择")
+            return {}
+
+        # 按索引映射结果：contents[i] 对应 valid_asset_ids[i]
+        recommendations: dict[str, float] = {}
+        for idx, content_text in enumerate(contents):
+            if idx >= len(valid_asset_ids):
+                break
+            asset_id = valid_asset_ids[idx]
+            if not content_text:
+                continue
+
+            # 尝试从文本中提取 JSON
+            parsed = False
+            # 尝试直接解析
+            try:
+                data = json.loads(content_text.strip())
+                if isinstance(data, list) and data:
+                    for item in data:
+                        if isinstance(item, dict) and "recommended_start_time" in item:
+                            recommendations[asset_id] = float(item["recommended_start_time"])
+                            parsed = True
+                            break
+            except (json.JSONDecodeError, ValueError, TypeError):
+                pass
+
+            # 尝试从 markdown 代码块中提取 JSON
+            if not parsed:
+                json_match = re.search(r"\[\s*(\{.*?\})\s*\]", content_text, re.DOTALL)
+                if json_match:
+                    try:
+                        item = json.loads(json_match.group(1))
+                        if isinstance(item, dict) and "recommended_start_time" in item:
+                            recommendations[asset_id] = float(item["recommended_start_time"])
+                            parsed = True
+                    except (json.JSONDecodeError, ValueError, TypeError):
+                        pass
+
+            # 尝试正则提取
+            if not parsed:
+                time_match = re.search(r'recommended_start_time["\s:]+([\d.]+)', content_text)
+                if time_match:
+                    try:
+                        recommendations[asset_id] = float(time_match.group(1))
+                    except (ValueError, TypeError):
+                        pass
+
+        if recommendations:
+            logger.info("MediaKit 智能选片推荐: %s", recommendations)
+        else:
+            logger.info("MediaKit 结果解析失败，降级为随机选择")
+
+        return recommendations
+
+    except Exception as e:
+        logger.warning("MediaKit 智能选片异常，降级为随机选择: %s", e)
+        return {}
+
+
+def _calc_plan_internal_duplicate_rate(clips_data: list[dict]) -> float:
+    """估算单条成片内部重复率（%）.
+
+    检查本条成片中同一素材是否有重叠的时间区间。
+    重叠时长 / 成片总时长 * 100 = 内部重复率。
+    这是一个轻量估算，不依赖视频指纹；完整查重由 worker 异步完成。
+    """
+    if not clips_data:
+        return 0.0
+
+    # 按素材分组
+    by_asset: dict[str, list[tuple[float, float]]] = {}
+    total_duration = 0.0
+    for c in clips_data:
+        aid = c.get("asset_id", "")
+        if not aid:
+            continue
+        start = c.get("start_time", 0.0)
+        end = start + c.get("duration", 0.0)
+        by_asset.setdefault(aid, []).append((start, end))
+        total_duration += c.get("duration", 0.0)
+
+    if total_duration <= 0:
+        return 0.0
+
+    # 检查同素材内的区间重叠
+    overlap_duration = 0.0
+    for segments in by_asset.values():
+        if len(segments) < 2:
+            continue
+        segments_sorted = sorted(segments, key=lambda s: s[0])
+        last_end = segments_sorted[0][1]
+        for start, end in segments_sorted[1:]:
+            overlap = max(0.0, min(end, last_end) - start)
+            if overlap > 0:
+                overlap_duration += overlap
+            last_end = max(last_end, end)
+
+    return round(overlap_duration / total_duration * 100, 1)
+
+
 @router.post("/clips/from-assets", response_model=ClipsFromAssetsResponse)
 def create_clips_from_assets_editor(
     template_id: str,
     body: ClipsFromAssetsRequest,
+    background_tasks: BackgroundTasks,
     plan_id: str = Depends(get_draft_plan_id),
     services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
+    asset_repo: SQLAlchemyAssetRepository = Depends(get_asset_repository),
+    db: Session = Depends(get_db_session),
     current_user: AuthenticatedUser = Depends(get_current_user),
 ) -> ClipsFromAssetsResponse:
-    """从素材批量创建片段"""
-    _, plan_svc = services
-    clips = []
-    for i, asset_id in enumerate(body.asset_ids):
-        try:
-            clip = plan_svc.create_clip(
-                plan_id,
-                clip_type="main",
-                order=body.start_order + i if hasattr(body, "start_order") else i,
-                duration=5.0,
-                asset_id=asset_id,
+    """从素材批量创建片段（按模板segment配置创建，MediaKit异步更新）.
+
+    逻辑：
+    1. 从模板读取 segments，片段数量 = segment 数量（忽略前端传的 required_clips_count）
+    2. 每个片段时长在 segment 的 duration_min ~ duration_max 之间随机取值（保留一位小数）
+    3. 素材按片段顺序轮询分配，素材不够时同一素材切多个片段
+    4. 使用 replace_all_clips_transactional 原子性地清空旧片段并创建新的（随机起始时间）
+    5. 立即返回响应（目标 <1秒）
+    6. 后台异步任务：调用 MediaKit 智能选片并更新片段的 start_time
+    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
+
+    # 1. 查询模板 segments
+    segments = _get_template_segments(template_id, tpl_svc, db)
+    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",
+        )
+
+    # 防御：schema validator 已过滤 null/空串，这里再归一化一次，
+    # 避免异常入参（undefined → null）导致后续 /assets/{id} 404 / 422
+    asset_ids = [str(aid).strip() for aid in (body.asset_ids or []) if isinstance(aid, str) and aid.strip()]
+    if not asset_ids:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="素材列表为空，无法创建片段",
+        )
+
+    # 2. 获取素材实际时长（去重查询）
+    unique_asset_ids = list(dict.fromkeys(asset_ids))
+    asset_durations: dict[str, float] = {}
+    asset_smart_scores: dict[str, float] = {}
+    # 素材 metadata 中缓存的场景切换点（由后台 MediaKit SceneChange 检测写入）：
+    # 有缓存时片段起点从随机镜头段中选取（不同片段来自不同镜头），无缓存回退随机起点
+    asset_scene_points: dict[str, list[float]] = {}
+    for asset_id in unique_asset_ids:
+        asset = asset_repo.get(asset_id)
+        if asset and hasattr(asset, "duration"):
+            asset_durations[asset_id] = float(asset.duration or 0.0)
+            # 计算 smart_match 综合评分，用于候选排序
+            smart_score, _ = score_asset(asset)
+            asset_smart_scores[asset_id] = smart_score
+            # 读取场景切换点缓存（新素材未检测过时为 None，走随机起点兜底）
+            cached_points = extract_scene_points_from_metadata(getattr(asset, "metadata", None))
+            if cached_points:
+                asset_scene_points[asset_id] = cached_points
+    logger.info(
+        "from-assets 场景缓存命中: %d/%d 个素材有场景切换点",
+        len(asset_scene_points),
+        len(unique_asset_ids),
+    )
+
+    # 3. 在内存中计算所有片段数据（使用随机起始时间，不调用MediaKit）
+    # 读取素材 metadata 中持久化的历史已用区间（跨任务/跨调用去重），
+    # 格式与 _calc_random_start_time 的 used_segments 参数一致
+    used_segments: dict[str, list[tuple[float, float]]] = get_used_segments(db, unique_asset_ids)
+    # 受控复用回调：可用区间耗尽时复用最久未用且未达复用上限（3次）的历史区间，
+    # 复用片段时长累加到 reused_durations 供 15% 占比控制
+    reused_durations: dict[str, float] = {}
+    # 本条成片中每个素材被分配的片段总时长（复用占比分母）
+    asset_assigned_durations: dict[str, float] = {}
+    # 受控复用回调：区间耗尽时复用最久未用且 use_count<3 的历史区间；
+    # 回调内部预判复用后占比是否超 15%，超限拒绝复用（返回 None）
+    reuse_cb = make_reuse_callback(
+        db,
+        asset_durations,
+        reused_durations,
+        assigned_tracker=asset_assigned_durations,
+    )
+    clips_data: list[dict] = []
+
+    def _reuse_ratio_exceeded(aid: str, extra: float = 0.0) -> bool:
+        """该素材在本条成片中「已复用片段时长 / 已分配片段总时长」是否已超 15%。
+
+        在为下一片段选素材时调用：本片段尚未分配，复用状态只在分配后的回调里
+        更新，因此直接检查当前占比——一旦已超 15%，该素材不再参与后续分配。
+        assigned=0（首个片段）放行；reused=0（尚未发生复用）时不误拦正常分配。
+        """
+        assigned = asset_assigned_durations.get(aid, 0.0)
+        if assigned <= 0:
+            return False
+        return reused_durations.get(aid, 0.0) / assigned > REUSE_RATIO_LIMIT
+
+    # 素材耗尽标志：某轮循环中所有素材均被跳过时为 True
+    all_assets_exhausted = False
+
+    # 计算转场重叠补偿：每个 clip 需要额外增加的时长
+    # 目标：渲染后视频总时长 = 模板设定的各片段时长之和
+    # 公式：每 clip 增加 (n_segments - 1) * td / n_segments
+    n_segments = len(segments)
+    if n_segments > 1:
+        transition_compensation = (n_segments - 1) * _DEFAULT_TRANSITION_DURATION / n_segments
+    else:
+        transition_compensation = 0.0
+
+    # 打乱 segments 的处理顺序（分配素材的顺序随机化），但最终 clips_data 按原始 order 排序
+    shuffled_indices = list(range(len(segments)))
+    random.shuffle(shuffled_indices)
+
+    for idx in shuffled_indices:
+        _seg_order, dur_min, dur_max = segments[idx]
+        # 在 segment 的 duration_min ~ duration_max 之间随机取值（保留一位小数）
+        raw_duration = random.uniform(dur_min, dur_max)
+        # 加上转场补偿，确保最终输出时长 = 模板设定总时长
+        raw_duration += transition_compensation
+
+        # 贪心分配素材：按"已使用次数"升序排列候选素材（使用最少的优先），
+        # 同次数随机打散，避免"A-B-C-D"的固定组合反复出现。
+        # 跳过时长缺失、复用占比已超 10% 阈值的素材；
+        # 选中后计算起点，若该素材可用区间耗尽且复用被闸门拒绝（calc 返回 None），
+        # 继续尝试下一个素材
+        asset_id = ""
+        clip_duration = 0.0
+        start_time: float | None = None
+        # 动态按使用次数排序：优先选使用最少的素材，同次数随机打散
+        asset_use_counts = {aid: len(used_segments.get(aid, [])) for aid in asset_ids}
+        # 排序键：smart_match 评分（注入随机噪声）→ 使用次数 → 纯随机。
+        # 噪声让得分接近的素材排名每次浮动，避免同一批素材反复选出相同组合，
+        # 从素材组合层面降低成片查重率；分差 > SCORE_RANDOM_NOISE_MAX 时排名稳定，
+        # 质量差距显著的素材仍保持优先级。
+        sorted_candidates = sorted(
+            asset_ids,
+            key=lambda aid: (
+                -(asset_smart_scores.get(aid, 0.0) + random.uniform(0.0, SCORE_RANDOM_NOISE_MAX)),
+                asset_use_counts.get(aid, 0),
+                random.random(),
+            ),
+        )
+        for candidate in sorted_candidates:
+            candidate_total = asset_durations.get(candidate, 0.0)
+            if candidate_total <= 0:
+                continue
+            candidate_duration = min(round(raw_duration, 1), candidate_total)
+            if candidate_duration <= 0:
+                continue
+            if _reuse_ratio_exceeded(candidate, candidate_duration):
+                logger.info(
+                    "from-assets 素材复用占比超 %.0f%% 阈值，跳过分配: asset_id=%s",
+                    REUSE_RATIO_LIMIT * 100,
+                    candidate,
+                )
+                continue
+            # 起始时间选取（不调用 MediaKit，保证接口快速返回）：
+            # 1) 素材有场景切换点缓存时，优先从随机镜头段中选起点（不同片段来自不同镜头，
+            #    画面内容本质不同），与 used_segments 做冲突避让（含 1.5s 边缘间隙）
+            # 2) 无缓存 / 镜头段全冲突 → _calc_random_start_time 随机起点兜底；
+            #    100 次避不开历史区间时走受控复用回调（复用片段累加 reused_durations，
+            #    回调内部预判复用后占比超 10% 则拒绝并返回 None）
+            candidate_start = None
+            if candidate in asset_scene_points:
+                candidate_start = pick_scene_aware_start(
+                    candidate,
+                    candidate_duration,
+                    asset_durations,
+                    asset_scene_points,
+                    used_segments,
+                    edge_gap=SEGMENT_EDGE_GAP,
+                )
+            if candidate_start is None:
+                candidate_start = _calc_random_start_time(
+                    candidate,
+                    candidate_duration,
+                    asset_durations,
+                    used_segments,
+                    on_exhausted=reuse_cb,
+                )
+            if candidate_start is None:
+                # 该素材可用区间耗尽且复用被闸门/use_count 上限拒绝 → 尝试下一素材
+                logger.info(
+                    "from-assets 素材无可用可切区间（复用被拒），轮询下一素材: asset_id=%s",
+                    candidate,
+                )
+                continue
+            asset_id = candidate
+            clip_duration = candidate_duration
+            start_time = candidate_start
+            break
+
+        if not asset_id or start_time is None:
+            # 所有素材时长缺失、复用占比超阈值，或区间耗尽且复用被拒 → 素材可切区间不足
+            all_assets_exhausted = True
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="素材可切区间不足，请补充新素材",
             )
-            clips.append(clip)
-        except ValueError:
-            pass
+
+        # 记录已使用时间段（内存，供本次后续片段避开）
+        used_segments.setdefault(asset_id, []).append((start_time, start_time + clip_duration))
+        asset_assigned_durations[asset_id] = asset_assigned_durations.get(asset_id, 0.0) + clip_duration
+        # 同步写入素材 metadata（不 commit，与下方 replace_all_clips_transactional
+        # 处于同一事务，任一步失败整体回滚，不留脏数据）；
+        # 复用区间与历史记录高度重叠时 record 内部自动累加 use_count
+        record_used_segments(db, asset_id, start_time, start_time + clip_duration, plan_id)
+
+        clips_data.append(
+            {
+                "order": _seg_order,
+                "asset_id": asset_id,
+                "start_time": start_time,
+                "duration": clip_duration,
+                "clip_type": body.clip_type or "main",
+            }
+        )
+
+    # 按原始 segment order 排序，确保 clips_data 的 order 字段有序（0,1,2,3...）
+    clips_data.sort(key=lambda c: c["order"])
+
+    # 4. 事务性替换：清空旧片段 → 创建新片段 → 标记ready（单事务，失败自动回滚）
+    created_count = plan_svc.replace_all_clips_transactional(plan_id, clips_data)
 
     logger.info(
-        "模板编辑器从素材创建片段: template_id=%s plan_id=%s count=%d by user=%s",
+        "from-assets按模板创建片段(异步): template_id=%s plan_id=%s segments=%d created=%d by user=%s",
         template_id,
         plan_id,
-        len(clips),
+        len(segments),
+        created_count,
         current_user.user.id,
     )
 
-    return ClipsFromAssetsResponse(
-        created_count=len(clips),
-        plan_id=plan_id,
-        clip_ids=[c.id for c in clips],
+    # 5. 触发后台任务：异步调用 MediaKit 并更新片段起始时间
+    background_tasks.add_task(
+        _update_mediakit_recommendations_async,
+        plan_id,
+        unique_asset_ids,
     )
+
+    # 6. 估算成片内部重复率（本条成片中同一素材的重叠片段时长占比）
+    dup_rate = _calc_plan_internal_duplicate_rate(clips_data)
+    duplicate_warning = None
+    if dup_rate > 50:
+        duplicate_warning = f"查重率 {dup_rate:.1f}% 超过50%，建议更换素材或模板"
+        logger.warning(
+            "from-assets 成片查重率超标: plan_id=%s dup_rate=%.1f%%",
+            plan_id,
+            dup_rate,
+        )
+
+    # 7. 素材耗尽提示
+    exhaustion_warning = None
+    if all_assets_exhausted and created_count < len(segments):
+        exhaustion_warning = (
+            "素材可切区间不足，部分片段使用了复用素材。" "建议：1) 补充更多素材到素材库 2) 使用不同的素材组合生成"
+        )
+
+    # 8. 立即返回响应
+    return ClipsFromAssetsResponse(
+        created_count=created_count,
+        plan_id=plan_id,
+        clip_ids=[],
+        duplicate_warning=duplicate_warning,
+        exhaustion_warning=exhaustion_warning,
+    )
+
+
+def _update_mediakit_recommendations_async(  # pragma: no cover
+    plan_id: str,
+    asset_ids: list[str],
+) -> None:
+    """后台任务：使用 SceneChange 智能选帧并更新片段的起始时间.
+
+    优先使用 SceneChange 策略检测视频镜头切换点，将每个素材按镜头段拆分，
+    各片段优先从不同镜头段中选取起始时间，实现「不同片段展示不同场景」的效果。
+
+    降级策略：
+    1. SceneChange 优先 → detect_scene_changes 内部已含 TimeInterval 降级
+    2. 若 detect_scene_changes 仍返回 None → 回退到旧的 analyze_videos 方式
+    3. 所有方式都失败 → 保持现有随机 start_time，不影响视频生成
+
+    此函数在后台异步执行，不影响接口响应时间。
+    失败时静默处理，不影响已创建的片段。
+    """
+    from collections import defaultdict
+
+    from packages.adapters.sqlalchemy_impl.asset_repository import SQLAlchemyAssetRepository
+    from packages.adapters.sqlalchemy_impl.session import SessionLocal
+
+    db = None
+    try:
+        # 复用应用全局 Session（避免每次创建新连接池导致资源泄漏）
+        if SessionLocal is None:
+            logger.warning("后台任务: SessionLocal 未初始化，跳过 MediaKit 更新")
+            return
+        db = SessionLocal()
+
+        # 初始化服务
+        asset_repo = SQLAlchemyAssetRepository(db)
+        plan_svc = EditPlanService(db)
+
+        # 查询该 plan 的所有片段（分批获取，避免硬编码 limit 截断）
+        batch_size = 500
+        all_clips = []
+        offset = 0
+        while True:
+            batch = plan_svc.list_clips(plan_id, skip=offset, limit=batch_size)
+            if not batch:
+                break
+            all_clips.extend(batch)
+            if len(batch) < batch_size:
+                break
+            offset += batch_size
+        clips = all_clips
+
+        if not clips:
+            logger.info("后台任务: plan_id=%s 无片段，跳过更新", plan_id)
+            return
+
+        # 批量预加载所有涉及的素材（消除 N+1 查询）
+        unique_asset_ids = list({getattr(c, "asset_id", "") or "" for c in clips} - {""})
+        assets_map: dict[str, object] = {a.id: a for a in asset_repo.find_by_ids(unique_asset_ids)}
+
+        # 按 asset_id 预分组片段对象（按 order 排序，保证按模板顺序分配镜头段）
+        clips_by_asset: dict[str, list] = defaultdict(list)
+        for clip in clips:
+            aid = getattr(clip, "asset_id", "") or ""
+            if aid:
+                clips_by_asset[aid].append(clip)
+        for aid in clips_by_asset:
+            clips_by_asset[aid].sort(key=lambda c: c.order)
+
+        # 读取素材全部历史已用区间（跨任务/跨 plan 持久化记录）
+        historical_segments = get_used_segments(db, unique_asset_ids)
+
+        # 已更新的片段ID（用于排除已移动的旧时间段）
+        updated_clip_ids: set[str] = set()
+        # 已更新的时间段
+        updated_segments: dict[str, list[tuple[float, float]]] = {}
+        updated_count = 0
+
+        # 尝试获取存储服务（用于生成视频 URL）
+        try:
+            storage = get_storage_service()
+        except Exception:
+            logger.warning("后台任务: 获取存储服务失败，跳过 SceneChange 更新")
+            return
+
+        # 获取 MediaKit 客户端
+        client = get_mediakit_client()
+
+        # 对每个素材，检测场景切换点并分配镜头段
+        for asset_id in unique_asset_ids:
+            asset_clips = clips_by_asset.get(asset_id, [])
+            if not asset_clips:
+                continue
+
+            asset = assets_map.get(asset_id)
+            if not asset:
+                continue
+            asset_total = float(getattr(asset, "duration", 0.0) or 0.0)
+            if asset_total <= 0:
+                continue
+
+            # 获取素材视频 URL
+            video_url: str | None = None
+            storage_key = getattr(asset, "storage_key", None) or ""
+            mime = getattr(asset, "mime_type", "") or ""
+            if storage_key and mime.startswith("video/"):
+                try:
+                    video_url = storage.get_download_url(storage_key)
+                except Exception as e:
+                    logger.warning("后台任务: 获取素材URL失败: asset_id=%s error=%s", asset_id, e)
+
+            # 构建该素材的占用区间列表（排除已更新片段）
+            def _get_other_segments(asset_id_inner, clip_id_inner):
+                segs: list[tuple[float, float]] = []
+                for c in clips_by_asset.get(asset_id_inner, []):
+                    cid = c.id
+                    if cid != clip_id_inner and cid not in updated_clip_ids:
+                        segs.append((c.start_time, c.start_time + c.duration))
+                segs.extend(updated_segments.get(asset_id_inner, []))
+
+                # 并入历史已用区间
+                def _norm(segs_in):
+                    return {(round(float(a), 3), round(float(b), 3)) for a, b in segs_in}
+
+                return list(_norm(segs) | _norm(historical_segments.get(asset_id_inner, [])))
+
+            # 优先使用 SceneChange 策略
+            scene_segments: list[tuple[float, float]] = []
+            # 先查素材 metadata 中的场景点缓存：命中则直接复用，跳过 MediaKit 检测
+            # （缓存由本任务首次检测后写入，跨任务/跨 plan 复用）
+            cached_points = extract_scene_points_from_metadata(getattr(asset, "metadata", None))
+            if cached_points:
+                scene_segments = build_scene_segments(cached_points, asset_total)
+                logger.info(
+                    "后台任务: 命中场景点缓存: asset_id=%s scenes=%d",
+                    asset_id,
+                    len(scene_segments),
+                )
+
+            if not scene_segments and client.is_available and video_url:
+                scene_changes = client.detect_scene_changes(video_url)
+                if scene_changes is not None:
+                    scene_segments = build_scene_segments(scene_changes, asset_total)
+                    logger.info(
+                        "后台任务: 素材场景检测完成: asset_id=%s scenes=%d",
+                        asset_id,
+                        len(scene_segments),
+                    )
+                    # 检测结果写入素材 metadata 缓存：首次生成用随机起点，
+                    # 检测完成后后续生成的渲染前同步路径即可读缓存选镜头段
+                    try:
+                        existing_meta = dict(getattr(asset, "metadata", None) or {})
+                        existing_meta["scene_change_points"] = scene_changes
+                        asset.metadata = existing_meta  # type: ignore[attr-defined]
+                        asset_repo.update(asset)  # type: ignore[arg-type]
+                        logger.info(
+                            "后台任务: 场景点已写入素材缓存: asset_id=%s points=%d",
+                            asset_id,
+                            len(scene_changes),
+                        )
+                    except Exception as cache_err:
+                        # 缓存写入失败不影响本次片段更新
+                        logger.warning(
+                            "后台任务: 场景点缓存写入失败: asset_id=%s error=%s",
+                            asset_id,
+                            cache_err,
+                        )
+
+            # SceneChange 未获得有效结果 → 尝试 analyze_videos 作为 fallback
+            if not scene_segments and video_url:
+                fallback_recs = _get_mediakit_recommendations([asset_id], asset_repo)
+                if fallback_recs and asset_id in fallback_recs:
+                    # analyze_videos 只返回单个推荐点，转为单镜头段
+                    rec_start = fallback_recs[asset_id]
+                    scene_segments = [(rec_start, asset_total)]
+                    logger.info(
+                        "后台任务: 使用 analyze_videos fallback: asset_id=%s start=%.2f",
+                        asset_id,
+                        rec_start,
+                    )
+
+            if not scene_segments:
+                # 所有方式都失败 → 保持现有随机 start_time
+                logger.info(
+                    "后台任务: SceneChange 与 analyze_videos 均无结果，保持随机起点: asset_id=%s",
+                    asset_id,
+                )
+                continue
+
+            # 为每个片段分配不同的镜头段
+            scene_segments_pool = list(scene_segments)  # 可消费的镜头段池
+            for clip in asset_clips:
+                clip_duration = clip.duration
+                recommended_start: float | None = None
+
+                # 从镜头段池中依次尝试，选一个不冲突的
+                for seg_idx, (seg_start, seg_end) in enumerate(scene_segments_pool):
+                    candidate_start = pick_start_in_scene_segment(seg_start, seg_end, clip_duration)
+                    if candidate_start is None:
+                        continue  # 镜头段太短，跳过
+
+                    # 检查越界
+                    if candidate_start + clip_duration > asset_total:
+                        continue
+
+                    # 检查与已用区间冲突
+                    other_segs = _get_other_segments(asset_id, clip.id)
+                    if _recommended_time_conflicts(candidate_start, clip_duration, other_segs):
+                        continue
+
+                    recommended_start = candidate_start
+                    # 消费该镜头段（从池中移除，下一个片段用不同镜头段）
+                    scene_segments_pool.pop(seg_idx)
+                    break
+
+                if recommended_start is None:
+                    # 镜头段用完或都冲突 → 尝试 _calc_random_start_time 兜底
+                    used_segs_for_calc: dict[str, list[tuple[float, float]]] = {
+                        asset_id: _get_other_segments(asset_id, clip.id)
+                    }
+                    fallback_start = _calc_random_start_time(
+                        asset_id,
+                        clip_duration,
+                        {asset_id: asset_total},
+                        used_segs_for_calc,
+                    )
+                    if fallback_start is None:
+                        continue  # 完全无法分配，保持原起点
+                    recommended_start = fallback_start
+
+                # 更新片段起始时间
+                try:
+                    old_start = clip.start_time
+                    old_end = old_start + clip_duration
+
+                    plan_svc.update_clip(clip.id, start_time=recommended_start)
+                    try:
+                        if remove_used_segment(db, asset_id, old_start, old_end, plan_id=plan_id):
+                            record_used_segments(
+                                db,
+                                asset_id,
+                                recommended_start,
+                                recommended_start + clip_duration,
+                                plan_id,
+                            )
+                    except Exception as me:
+                        logger.warning(
+                            "后台任务: 同步素材区间记录失败，回滚本次片段更新: clip_id=%s error=%s",
+                            clip.id,
+                            me,
+                        )
+                        db.rollback()
+                        continue
+                    db.commit()
+                    updated_count += 1
+                    updated_clip_ids.add(clip.id)
+                    updated_segments.setdefault(asset_id, []).append(
+                        (recommended_start, recommended_start + clip_duration)
+                    )
+                    logger.info(
+                        "后台任务: 更新片段起始时间(场景选帧): clip_id=%s asset_id=%s start_time=%.2f",
+                        clip.id,
+                        asset_id,
+                        recommended_start,
+                    )
+                except Exception as ue:
+                    logger.warning("后台任务: 单个片段更新失败: clip_id=%s error=%s", clip.id, ue)
+                    try:
+                        db.rollback()
+                    except Exception:
+                        pass
+                    continue
+
+        logger.info("后台任务完成: plan_id=%s 成功更新 %d 个片段", plan_id, updated_count)
+
+    except Exception as e:
+        # 后台任务失败不影响已创建的片段，静默处理
+        logger.warning("后台任务异常: plan_id=%s error=%s", plan_id, e, exc_info=True)
+        if db:
+            try:
+                db.rollback()
+            except Exception:
+                pass
+    finally:
+        if db:
+            try:
+                db.close()
+            except Exception:
+                pass

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

@@ -41,17 +41,17 @@ def list_editor_transition_presets(
     _: AuthenticatedUser = Depends(get_current_user),
 ) -> TransitionPresetListResponse:
     """获取转场预设列表"""
-    from packages.domain.transition_presets import TRANSITION_PRESETS
+    from packages.domain.transition_presets import TRANSITION_PRESET_LIBRARY
 
     items = [
         {
-            "id": p["id"],
-            "name": p["name"],
-            "category": p.get("category", "通用"),
-            "duration": p.get("default_duration", 0.5),
-            "description": p.get("description", ""),
+            "id": p.id,
+            "name": p.name,
+            "category": p.category,
+            "duration": p.default_duration,
+            "description": p.description,
         }
-        for p in TRANSITION_PRESETS
+        for p in TRANSITION_PRESET_LIBRARY
     ]
     return TransitionPresetListResponse(items=items, total=len(items))
 
@@ -123,17 +123,17 @@ def list_editor_filter_presets(
     _: AuthenticatedUser = Depends(get_current_user),
 ) -> FilterPresetListResponse:
     """获取滤镜预设列表"""
-    from packages.domain.filter_presets import FILTER_PRESETS
+    from packages.domain.filter_presets import FILTER_PRESET_LIBRARY
 
     items = [
         {
-            "id": p["id"],
-            "name": p["name"],
-            "category": p.get("category", "通用"),
-            "thumbnail": p.get("thumbnail", ""),
-            "description": p.get("description", ""),
+            "id": p.id,
+            "name": p.name,
+            "category": p.category,
+            "thumbnail": p.lut_url,
+            "description": p.description,
         }
-        for p in FILTER_PRESETS
+        for p in FILTER_PRESET_LIBRARY
     ]
     return FilterPresetListResponse(items=items, total=len(items))
 

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

@@ -167,6 +167,18 @@ class ClipsFromAssetsRequest(BaseModel):
 
     asset_ids: List[str] = Field(..., min_length=1, max_length=200, description="素材 ID 列表，按顺序追加到时间线末尾")
     clip_type: str = Field(default="main", description="片段类型，默认 main")
+    required_clips_count: Optional[int] = Field(
+        default=None, ge=1, le=200, description="要求创建的片段数量；不传则等于素材数量"
+    )
+
+    @validator("asset_ids", pre=True)
+    def _drop_invalid_asset_ids(cls, v):  # noqa: N805
+        """容错过滤：前端异常情况下可能把 undefined 序列化成 null 或空串混入
+        asset_ids（会直接 422 或导致后续 /assets/{id} 404），这里统一剔除。
+        过滤后为空时由 Field(min_length=1) / 路由层 400 兜底。"""
+        if not isinstance(v, list):
+            return v
+        return [x for x in v if isinstance(x, str) and x.strip()]
 
 
 class ClipsFromAssetsResponse(BaseModel):
@@ -177,6 +189,8 @@ class ClipsFromAssetsResponse(BaseModel):
     plan_id: str = ""
     message: str = ""
     clip_ids: List[str] = Field(default_factory=list, description="创建的片段ID列表")
+    duplicate_warning: Optional[str] = Field(default=None, description="查重率超标警告")
+    exhaustion_warning: Optional[str] = Field(default=None, description="素材耗尽警告")
 
 
 # ── 封面配置 ────────────────────────────────────────────────────────────────

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

@@ -2,18 +2,24 @@
 
 from __future__ import annotations
 
+import json
 import logging
-from typing import Optional
+import subprocess
+import tempfile
+from pathlib import Path
+from typing import Any, Optional
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.celery_app import celery_app
+from app.core.storage import get_storage_service
 from app.dependencies import (
+    get_asset_library_repository,
+    get_asset_repository,
     get_audio_url_signer,
     get_cosyvoice_service,
     get_db_session,
-    get_user_repository,
+    get_project_repository,
     get_voice_clone_profile_repository,
-    get_voice_library_repository,
 )
 from app.schemas.tts import (
     ListTTSJobResponse,
@@ -27,12 +33,12 @@ from app.schemas.tts import (
     TTSSynthesizeResponse,
 )
 from fastapi import APIRouter, Depends, HTTPException, Query, Response, WebSocket, WebSocketDisconnect, status
+from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl.tts_job_repository import (
     SQLAlchemyTTSJobRepository,
 )
-from packages.adapters.sqlalchemy_impl.voice_library_repository import SQLAlchemyVoiceLibraryRepository
 from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
 from packages.application.tts_job.streaming_service import TTSStreamingService
 from packages.application.tts_job.use_cases import (
@@ -44,13 +50,12 @@ from packages.application.tts_job.use_cases import (
     TTSJobNotFoundError,
 )
 from packages.application.tts_job.workflow import TTSWorkflowService
-from packages.application.voice_library.commands import CreateVoiceLibraryCommand
-from packages.application.voice_library.use_cases import (
-    CreateVoiceLibraryUseCase,
-    QuotaExceededError,
-)
+from packages.domain import Asset, AssetLibrary, AssetLibraryKind, AssetStatus, ClassificationStatus
 from packages.domain.voice_presets import list_voices
-from packages.ports.user_repository import UserRepository
+from packages.ports.asset_library_repository import AssetLibraryRepository
+from packages.ports.asset_repository import AssetRepository
+from packages.ports.project_repository import ProjectRepository
+from packages.shared.storage import SharedStorageService
 
 logger = logging.getLogger(__name__)
 
@@ -134,27 +139,47 @@ def synthesize(
     """
     user_id = authenticated_user.user.id
 
-    # 校验 voice_clone_profile_id 归属（防止越权使用他人克隆音色）
-    if request.voice_clone_profile_id:
-        profile = voice_clone_repo.get(request.voice_clone_profile_id)
-        if profile is None:
+    # 解析 voice_id：前端可能传克隆音色 profile UUID（而非 CosyVoice voice_id），
+    # 与 /tts/preview 保持一致：命中 profile → 校验归属 → 取 CosyVoice voice_id
+    actual_voice_id = request.voice_id
+    voice_clone_profile_id = request.voice_clone_profile_id
+    resolved_profile = None
+    if actual_voice_id:
+        resolved_profile = voice_clone_repo.get(actual_voice_id)
+        if resolved_profile is not None:
+            voice_clone_profile_id = actual_voice_id
+
+    # 显式传了 voice_clone_profile_id（且与 voice_id 不同）时再查一次归属
+    if voice_clone_profile_id and (resolved_profile is None or resolved_profile.id != voice_clone_profile_id):
+        resolved_profile = voice_clone_repo.get(voice_clone_profile_id)
+        if resolved_profile is None:
             raise HTTPException(
                 status_code=status.HTTP_404_NOT_FOUND,
                 detail="Voice clone profile not found",
             )
-        if profile.user_id != user_id:
+
+    if resolved_profile is not None:
+        if resolved_profile.user_id != user_id:
             raise HTTPException(
                 status_code=status.HTTP_403_FORBIDDEN,
-                detail="Access denied to voice clone profile",
+                detail="无权访问该音色",
             )
+        if not resolved_profile.voice_id:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="音色克隆尚未完成，请稍后再试",
+            )
+        # 命中克隆音色：无论 voice_id 直接传 profile UUID 还是显式传 voice_clone_profile_id，
+        # job.voice_id 统一存解析后的 CosyVoice voice_id
+        actual_voice_id = resolved_profile.voice_id
 
     use_case = CreateTTSJobUseCase(repository)
     job = use_case.execute(
         user_id=user_id,
         input_text=request.text,
-        voice_id=request.voice_id,
+        voice_id=actual_voice_id,
         voice_model=request.voice_model,
-        voice_clone_profile_id=request.voice_clone_profile_id,
+        voice_clone_profile_id=voice_clone_profile_id,
         metadata=request.metadata_,
     )
 
@@ -284,6 +309,62 @@ def delete_tts_job(
     return
 
 
+def _find_or_create_voice_library(
+    *,
+    user_id: str,
+    project_repository: ProjectRepository,
+    asset_library_repository: Any,  # port Protocol 声明为 async，SQLAlchemy 实现为同步，与 upload/asset_libraries 路由惯例一致用 Any
+) -> AssetLibrary:
+    """在用户可访问的项目中找到（或自动创建）voice 素材库。
+
+    与前端配音素材页逻辑一致：素材库挂在项目下，配音素材读取
+    getAssetsByKind("voice") → 用户所有可访问项目中的 voice 库。
+    优先使用已有 voice 库；没有则在第一个可访问项目中自动创建。
+    """
+    projects = project_repository.find_accessible_projects(user_id)
+    if not projects:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="没有可用的项目，请先创建项目后再保存配音素材",
+        )
+
+    for project in projects:
+        for lib in asset_library_repository.find_by_project(project.id):
+            kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
+            if kind == AssetLibraryKind.VOICE.value:
+                return lib
+
+    # 所有项目都没有 voice 库 → 在第一个可访问项目中自动创建默认配音素材库。
+    # asset_libraries 有 (project_id, kind) 唯一索引兜底并发：若两个请求同时创建，
+    # 落败方捕获 IntegrityError 回滚后重新查询，返回抢先创建成功的库。
+    project = projects[0]
+    library = AssetLibrary.create(
+        project_id=project.id,
+        name="配音素材库",
+        kind=AssetLibraryKind.VOICE,
+    )
+    try:
+        return asset_library_repository.create(library)
+    except IntegrityError:
+        # 并发下另一个请求已抢先创建：回滚当前事务（立即 commit 模式下 session 已
+        # 自动回滚，rollback 为幂等 no-op；UoW/flush 模式下必须显式回滚才能继续查询），
+        # 再重查返回抢先创建成功的库。
+        session = getattr(asset_library_repository, "session", None)
+        if session is not None:
+            try:
+                session.rollback()
+            except Exception:
+                logger.warning("IntegrityError 后回滚 session 失败（可能已关闭）", exc_info=True)
+        for lib in asset_library_repository.find_by_project(project.id):
+            kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
+            if kind == AssetLibraryKind.VOICE.value:
+                return lib
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="配音素材库创建失败，请重试",
+        ) from None  # IntegrityError 已处理，不保留异常链
+
+
 @router.post(
     "/jobs/{job_id}/save-to-library",
     response_model=SaveToLibraryResponse,
@@ -294,13 +375,17 @@ def save_tts_job_to_library(
     request: SaveToLibraryRequest = SaveToLibraryRequest(),
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     tts_repository: SQLAlchemyTTSJobRepository = Depends(_get_repository),
-    voice_library_repository: SQLAlchemyVoiceLibraryRepository = Depends(get_voice_library_repository),
-    user_repository: UserRepository = Depends(get_user_repository),
+    asset_repository: AssetRepository = Depends(get_asset_repository),
+    asset_library_repository: AssetLibraryRepository = Depends(get_asset_library_repository),
+    project_repository: ProjectRepository = Depends(get_project_repository),
+    storage_service: SharedStorageService = Depends(get_storage_service),
     sign_url=Depends(get_audio_url_signer),
 ) -> SaveToLibraryResponse:
-    """将已完成的 TTS 合成结果保存到配音库。
+    """将已完成的 TTS 合成结果保存到配音素材库（assets 表新素材体系）。
 
-    自动携带音色名、时长、语速等元信息。
+    流程：把 TTS 输出音频转存到用户素材 OSS 路径 → 创建 file_type=audio、
+    status=ready 的 asset（挂用户 voice 素材库）→ 返回前端可用结构。
+    配额策略与素材上传一致（上传/ingest 链路无额外配额拦截）。
     """
     user_id = authenticated_user.user.id
 
@@ -318,64 +403,136 @@ def save_tts_job_to_library(
             detail="TTS job is not completed yet",
         )
 
-    # 构建配音素材名称
+    if not job.output_audio_url and not job.output_audio_key:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="TTS job 缺少输出音频，无法保存",
+        )
+
+    # 素材名称
     name = request.name or f"TTS-{job.id[:8]}"
 
-    # 构建元信息
-    metadata_ = {
+    # 找到（或自动创建）用户 voice 素材库
+    library = _find_or_create_voice_library(
+        user_id=user_id,
+        project_repository=project_repository,
+        asset_library_repository=asset_library_repository,
+    )
+
+    # 转存音频到素材 OSS 路径（tts-outputs/ 下的产物归 TTS 任务所有，
+    # 素材独立持有副本，删除 TTS 任务不影响配音库素材）
+    audio_format = (job.format or "mp3").strip() or "mp3"
+    content_type_map = {
+        "mp3": "audio/mpeg",
+        "wav": "audio/wav",
+        "pcm": "audio/pcm",
+        "opus": "audio/opus",
+    }
+    content_type = content_type_map.get(audio_format, "audio/mpeg")
+    storage_key = f"uploads/voice/tts/{job.id}.{audio_format}"
+
+    tmp_path: Path | None = None
+    audio_duration: float | None = None
+    file_size = 0
+    try:
+        with tempfile.NamedTemporaryFile(suffix=f".{audio_format}", delete=False) as tmp:
+            tmp_path = Path(tmp.name)
+        # 优先用 OSS storage_key（走 oss2 SDK，私有 bucket 也可下载）；
+        # 兜底用 output_audio_url（旧任务可能没有 key）。
+        # download_asset 自动识别输入：http(s):// 开头走 HTTP 下载，否则按 OSS key 走 SDK。
+        download_source = job.output_audio_key or job.output_audio_url
+        downloaded = storage_service.download_asset(download_source, tmp_path)
+        if not downloaded or not tmp_path.exists() or tmp_path.stat().st_size == 0:
+            raise HTTPException(
+                status_code=status.HTTP_502_BAD_GATEWAY,
+                detail="TTS 音频下载失败，无法保存到配音库",
+            )
+        file_size = tmp_path.stat().st_size
+        storage_service.upload_file(tmp_path, storage_key, content_type=content_type)
+
+        # 从音频文件提取时长（ffprobe），作为 job.duration 的兜底
+        try:
+            proc = subprocess.run(
+                [
+                    "ffprobe", "-v", "quiet", "-print_format", "json",
+                    "-show_format", str(tmp_path),
+                ],
+                capture_output=True, text=True, timeout=10,
+            )
+            if proc.returncode == 0:
+                fmt = json.loads(proc.stdout).get("format", {})
+                dur = float(fmt.get("duration", 0))
+                if dur > 0:
+                    audio_duration = dur
+        except Exception:
+            logger.warning("ffprobe 提取时长失败: job_id=%s", job.id, exc_info=True)
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error("TTS 音频转存素材失败: job_id=%s, error=%s", job.id, e, exc_info=True)
+        raise HTTPException(
+            status_code=status.HTTP_502_BAD_GATEWAY,
+            detail="TTS 音频转存失败，无法保存到配音库",
+        ) from e
+    finally:
+        if tmp_path and tmp_path.exists():
+            try:
+                tmp_path.unlink()
+            except OSError:
+                pass
+
+    # 构建素材元信息
+    metadata_: dict[str, object] = {
         "source": "tts_job",
         "tts_job_id": job.id,
         "format": job.format,
         "sample_rate": job.sample_rate,
+        "voice_id": job.voice_id,
+        "voice_name": job.voice_model or "",
     }
     if job.metadata:
-        # 保留原始 job 的有用元信息
         for key in ("speed", "language"):
             if key in job.metadata:
                 metadata_[key] = job.metadata[key]
 
-    # 获取用户套餐（用于配额检查）
-    user = user_repository.find_by_id(user_id)
-    plan_name = getattr(user, "subscription_plan", "free") if user else "free"
-
-    # 构建命令并执行
-    command = CreateVoiceLibraryCommand(
-        user_id=user_id,
+    asset = Asset.create(
+        project_id=library.project_id,
+        library_id=library.id,
         name=name,
-        text=job.input_text,
-        voice_provider="cosyvoice",
-        voice_id=job.voice_id,
-        voice_name=job.voice_model or "",
-        audio_url=job.output_audio_url,
-        duration=job.duration,
-        file_size=job.file_size,
-        status="completed",
-        project_id=job.project_id or "",
-        tags=[],
-        metadata_=metadata_,
+        storage_key=storage_key,
+        mime_type=content_type,
+        metadata=metadata_,
+        file_size=file_size,
+        duration=job.duration or audio_duration or None,
+        status=AssetStatus.READY,
+        classification_status=ClassificationStatus.PENDING,  # 音频不参与内容分类，保持 pending 与 ingest 链路一致
+        uploaded_by_user_id=user_id,
     )
-
-    use_case = CreateVoiceLibraryUseCase(voice_library_repository)
     try:
-        item = use_case.execute(command, plan_name=plan_name or "free")
-    except QuotaExceededError as exc:
+        asset = asset_repository.create(asset)
+    except Exception as e:
+        # DB 写入失败：清理已上传到 OSS 的素材文件，避免产生无法索引的孤儿文件
+        logger.error("素材记录创建失败，清理 OSS 文件: %s, error=%s", storage_key, e, exc_info=True)
+        try:
+            storage_service.delete_file(storage_key)
+        except Exception:
+            logger.warning("清理孤儿 OSS 文件失败: %s", storage_key, exc_info=True)
         raise HTTPException(
-            status_code=status.HTTP_429_TOO_MANY_REQUESTS,
-            detail=f"配音库配额已满（{exc.used}/{exc.limit}），请升级套餐",
-        ) from exc
+            status_code=status.HTTP_502_BAD_GATEWAY,
+            detail="素材保存失败，请重试",
+        ) from e
 
     return SaveToLibraryResponse(
-        id=item.id,
-        name=item.name,
-        audio_url=sign_url(item.audio_url) if item.audio_url else "",
-        duration=item.duration,
-        voice_id=item.voice_id,
-        voice_name=item.voice_name,
-        status=item.status,
+        id=asset.id,
+        name=asset.name,
+        audio_url=sign_url(storage_key),
+        duration=asset.duration or 0.0,
+        voice_id=job.voice_id,
+        voice_name=job.voice_model or "",
+        status="completed",
     )
 
 
-
 @router.post("/preview", response_model=TTSPreviewResponse)
 def preview_tts(
     request: TTSPreviewRequest,

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

@@ -23,6 +23,7 @@ from app.schemas.upload import (
 from fastapi import APIRouter, Depends, File, Form, HTTPException, UploadFile, status
 
 from packages.application import SubmitIngestJobCommand, SubmitIngestJobUseCase
+from packages.domain import Asset, AssetStatus
 
 logger = logging.getLogger(__name__)
 
@@ -80,6 +81,40 @@ def _validate_mime_type(content_type: str | None) -> str:
     return base_type
 
 
+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",
+    }
+    for ext, mime in _MIME_MAP.items():
+        if lower_filename.endswith(ext):
+            return mime
+    return "video/mp4"  # default
+
+
+def _create_pending_asset(
+    asset_repository, project_id, library_id, storage_key, filename, mime_type, user_id, file_hash=""
+):
+    """立即创建一条 PROCESSING 状态的 Asset 记录，使前端能马上看到新素材。"""
+    asset = Asset.create(
+        project_id=project_id,
+        library_id=library_id,
+        name=filename,
+        storage_key=storage_key,
+        mime_type=mime_type,
+        status=AssetStatus.PROCESSING,
+        uploaded_by_user_id=user_id,
+        file_hash=file_hash,
+    )
+    return asset_repository.create(asset)
+
+
 def _submit_ingest_job(
     project_id: str,
     library_id: str,
@@ -209,6 +244,20 @@ async def complete_direct_upload(
                 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,
+        project_id=request.project_id,
+        library_id=request.library_id,
+        storage_key=normalized_key,
+        filename=filename,
+        mime_type=mime_type,
+        user_id=authenticated_user.user.id,
+        file_hash=request.file_hash,
+    )
+
     job = _submit_ingest_job(
         project_id=request.project_id,
         library_id=request.library_id,
@@ -216,7 +265,12 @@ async def complete_direct_upload(
         ingest_job_repository=ingest_job_repository,
         file_hash=request.file_hash,
     )
-    return DirectUploadCompleteResponse(storage_key=normalized_key, ingest_job_id=job.id, url=storage_service.get_url(normalized_key))
+    return DirectUploadCompleteResponse(
+        storage_key=normalized_key,
+        ingest_job_id=job.id,
+        asset_id=pending_asset.id,
+        url=storage_service.get_url(normalized_key),
+    )
 
 
 @router.post(
@@ -284,6 +338,18 @@ async def upload_asset(
             detail=f"Failed to upload file: {type(error).__name__}",
         ) from error
 
+    # 立即创建 Asset 记录（PROCESSING 状态），使前端刷新后即可看到新素材
+    pending_asset = _create_pending_asset(
+        asset_repository=asset_repository,
+        project_id=project_id,
+        library_id=library_id,
+        storage_key=storage_key,
+        filename=safe_filename,
+        mime_type=validated_content_type,
+        user_id=authenticated_user.user.id,
+        file_hash=file_hash,
+    )
+
     job = _submit_ingest_job(
         project_id=project_id,
         library_id=library_id,
@@ -295,5 +361,6 @@ async def upload_asset(
     return UploadAssetResponse(
         storage_key=storage_key,
         ingest_job_id=job.id,
+        asset_id=pending_asset.id,
         url=file_url,
     )

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

@@ -52,6 +52,7 @@ def _to_video_response(item, storage: OSSStorageService | None = None) -> VideoI
         generation_params=item.generation_params,
         download_url=download_url,
         generated_at=format_utc_datetime(item.generated_at) if hasattr(item, "generated_at") else "",
+        duplicate_rate=getattr(item, "duplicate_rate", None),
     )
 
 

apps/api/app/api/routes/voice_clones.py

@@ -7,7 +7,13 @@ from typing import Optional
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.celery_app import celery_app
-from app.dependencies import get_cosyvoice_service, get_voice_clone_profile_repository
+from app.core.storage import get_storage_service
+from app.dependencies import (
+    get_asset_repository,
+    get_cosyvoice_service,
+    get_project_repository,
+    get_voice_clone_profile_repository,
+)
 from app.schemas.voice_clone import (
     CreateVoiceCloneRequest,
     ListVoiceCloneResponse,
@@ -32,6 +38,9 @@ from packages.application.voice_clone.use_cases import (
 from packages.application.voice_clone.workflow import (
     VoiceCloneWorkflowService,
 )
+from packages.ports.asset_repository import AssetRepository
+from packages.ports.project_repository import ProjectRepository
+from packages.shared.storage import SharedStorageService
 
 logger = logging.getLogger(__name__)
 
@@ -83,23 +92,68 @@ def create_voice_clone(
     request: CreateVoiceCloneRequest,
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     workflow: VoiceCloneWorkflowService = Depends(_get_workflow_service),
+    asset_repository: AssetRepository = Depends(get_asset_repository),
+    project_repository: ProjectRepository = Depends(get_project_repository),
+    storage_service: SharedStorageService = Depends(get_storage_service),
 ) -> VoiceCloneProfileResponse:
     """创建音色克隆任务。
 
     创建 VoiceCloneProfile → 提交 CosyVoice 克隆任务 → 触发 Celery 异步轮询。
-    如果有 source_audio_url，状态会变为 processing；否则保持 pending。
+    参考音频两种来源（二选一）：
+    - source_audio_url：前端直传后的音频 URL（兼容旧流程）
+    - asset_id：配音素材库中的音频素材，服务端用其 OSS storage_key 生成
+      预签名下载 URL（不依赖前端签名，避免签名过期导致克隆失败）
+    如果有参考音频，状态会变为 processing；否则保持 pending。
     """
     user_id = authenticated_user.user.id
+
+    source_audio_url = request.source_audio_url
+    clone_metadata = dict(request.metadata_ or {})
+
+    if request.asset_id:
+        if source_audio_url:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="asset_id 与 source_audio_url 只能传一个",
+            )
+        asset = asset_repository.find_by_id(request.asset_id)
+        if asset is None:
+            raise HTTPException(
+                status_code=status.HTTP_404_NOT_FOUND,
+                detail="素材不存在",
+            )
+        # 归属校验：素材挂在项目素材库下，用户必须能访问该项目
+        project = project_repository.find_by_id(asset.project_id)
+        if project is None or not project.can_access(user_id):
+            raise HTTPException(
+                status_code=status.HTTP_403_FORBIDDEN,
+                detail="无权使用该素材",
+            )
+        # 类型校验：仅支持音频素材
+        if asset.file_type != "audio":
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="仅支持音频素材进行音色克隆",
+            )
+        if not asset.storage_key:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="该素材缺少音频文件，无法用于克隆",
+            )
+        # 用 OSS storage_key 生成服务端预签名 URL（7 天有效，覆盖克隆重试周期）
+        source_audio_url = storage_service.get_download_url(asset.storage_key, expires_seconds=7 * 24 * 3600)
+        clone_metadata["source_asset_id"] = asset.id
+
     profile = workflow.start_clone(
         user_id=user_id,
         name=request.name,
         description=request.description,
-        source_audio_url=request.source_audio_url,
+        source_audio_url=source_audio_url,
         voice_model=request.voice_model,
         language=request.language,
         gender=request.gender,
         max_retries=request.max_retries,
-        metadata=request.metadata_,
+        metadata=clone_metadata,
     )
 
     # 如果 profile 处于 processing 且有 task_id，触发 Celery 异步轮询

apps/api/app/api/routes/voices.py

@@ -6,12 +6,26 @@
 from __future__ import annotations
 
 import logging
+import shutil
+import subprocess
+import tempfile
 import time
+from pathlib import Path
 from typing import Literal, Optional
+from uuid import uuid4
 
 from app.api.routes._helpers import get_user_plan
 from app.auth import AuthenticatedUser, get_current_user
-from app.dependencies import get_audio_url_signer, get_cosyvoice_service, get_db_session, get_user_repository
+from app.core.storage import get_storage_service
+from app.dependencies import (
+    get_asset_library_repository,
+    get_asset_repository,
+    get_audio_url_signer,
+    get_cosyvoice_service,
+    get_db_session,
+    get_project_repository,
+    get_user_repository,
+)
 from app.schemas.voice import (
     PresetVoiceItemResponse,
     PresetVoiceListResponse,
@@ -24,7 +38,7 @@ from app.schemas.voice_library import (
     UpdateVoiceLibraryRequest,
     VoiceLibraryItemResponse,
 )
-from fastapi import APIRouter, Depends, HTTPException, Query, Response, status
+from fastapi import APIRouter, Depends, File, Form, HTTPException, Query, Response, UploadFile, status
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl.voice_clone_profile_repository import SQLAlchemyVoiceCloneProfileRepository
@@ -40,8 +54,12 @@ from packages.application.voice_library.use_cases import (
     QuotaExceededError,
     UpdateVoiceLibraryUseCase,
 )
+from packages.domain import Asset, AssetStatus
+from packages.domain.classification import AssetLibraryKind, ClassificationStatus
+from packages.domain.entities import AssetLibrary
 from packages.domain.preset_voices import PRESET_VOICES, get_preset_voice_by_id
 from packages.ports.user_repository import UserRepository
+from packages.shared.storage import SharedStorageService
 
 router = APIRouter()
 logger = logging.getLogger(__name__)
@@ -507,3 +525,243 @@ def delete_voice(
     if not deleted:
         raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Voice not found")
     return
+
+
+# ── 提取视频配音 ─────────────────────────────────────────────────────
+
+# 支持的视频格式
+EXTRACT_VIDEO_MIMES = frozenset({"video/mp4", "video/quicktime", "video/webm", "video/x-msvideo"})
+MAX_EXTRACT_SIZE = 500 * 1024 * 1024  # 500MB
+
+
+@router.post(
+    "/extract-voice",
+    status_code=status.HTTP_201_CREATED,
+)
+def extract_voice_from_video(
+    file: UploadFile = File(...),
+    project_id: str = Form(...),
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    project_repository=Depends(get_project_repository),
+    asset_library_repository=Depends(get_asset_library_repository),
+    asset_repository=Depends(get_asset_repository),
+    storage_service: SharedStorageService = Depends(get_storage_service),
+    sign_url=Depends(get_audio_url_signer),
+):
+    """从上传的视频中提取人声配音。
+
+    流程：
+    1. 接收视频文件（mp4/mov/webm）
+    2. ffmpeg 提取音频 + 降噪 + 编码为 mp3
+    3. 上传到 OSS，创建 Asset 记录到配音素材库
+    4. 返回素材信息（时长、文件大小、URL）
+    """
+    user_id = authenticated_user.user.id
+
+    # 校验文件类型
+    content_type = file.content_type or ""
+    if content_type and content_type not in EXTRACT_VIDEO_MIMES:
+        # 兜底：按扩展名判断
+        ext = (file.filename or "").rsplit(".", 1)[-1].lower()
+        ext_to_mime = {"mp4": "video/mp4", "mov": "video/quicktime", "webm": "video/webm", "avi": "video/x-msvideo"}
+        if ext not in ext_to_mime:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="仅支持 mp4/mov/webm/avi 格式的视频文件",
+            )
+        content_type = ext_to_mime[ext]
+
+    # 找到（或自动创建）用户 voice 素材库（复用 TTS 的逻辑）
+    library = _find_or_create_voice_library_for_extract(
+        user_id=user_id,
+        project_repository=project_repository,
+        asset_library_repository=asset_library_repository,
+    )
+
+    tmp_dir = None
+    try:
+        tmp_dir = Path(tempfile.mkdtemp(prefix="voice_extract_"))
+        video_path = tmp_dir / f"input_{uuid4().hex[:8]}_{file.filename or 'video.mp4'}"
+        audio_path = tmp_dir / f"output_{uuid4().hex[:8]}.mp3"
+
+        # 保存上传的视频到临时文件
+        with open(video_path, "wb") as f:
+            total = 0
+            while chunk := file.file.read(1024 * 1024):  # 1MB chunks
+                total += len(chunk)
+                if total > MAX_EXTRACT_SIZE:
+                    raise HTTPException(
+                        status_code=status.HTTP_413_REQUEST_ENTITY_TOO_LARGE,
+                        detail="视频文件过大，最大支持 500MB",
+                    )
+                f.write(chunk)
+
+        if video_path.stat().st_size == 0:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="视频文件为空")
+
+        # ffmpeg: 提取音频 + 降噪 + 编码 mp3
+        # 滤镜链：highpass(去低频噪声) → afftdn(FFT降噪) → lowpass(去高频噪声)
+        ffmpeg_cmd = [
+            "ffmpeg",
+            "-y",
+            "-i",
+            str(video_path),
+            "-vn",  # 不要视频
+            "-af",
+            "highpass=f=80,afftdn=nf=-25:tn=1,lowpass=f=8000",
+            "-acodec",
+            "libmp3lame",
+            "-ab",
+            "192k",
+            "-ar",
+            "44100",
+            "-ac",
+            "1",  # 单声道（人声足够）
+            str(audio_path),
+        ]
+
+        result = subprocess.run(
+            ffmpeg_cmd,
+            capture_output=True,
+            timeout=300,  # 5 分钟超时
+        )
+
+        if result.returncode != 0:
+            stderr_text = result.stderr.decode("utf-8", errors="replace")[-500:]
+            logger.error("ffmpeg 提取配音失败: %s", stderr_text)
+            raise HTTPException(
+                status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
+                detail="视频音频提取失败，可能该视频没有音轨或格式不支持",
+            )
+
+        if not audio_path.exists() or audio_path.stat().st_size == 0:
+            raise HTTPException(
+                status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
+                detail="音频提取结果为空",
+            )
+
+        # 获取音频时长
+        duration = _get_audio_duration(audio_path)
+        file_size = audio_path.stat().st_size
+
+        # 上传到 OSS
+        audio_ext = "mp3"
+        storage_key = f"uploads/voice/extracted/{uuid4().hex}.{audio_ext}"
+        storage_service.upload_file(audio_path, storage_key, content_type="audio/mpeg")
+
+        # 创建 Asset 记录
+        original_name = (file.filename or "video").rsplit(".", 1)[0]
+        asset_name = f"{original_name}-配音"
+
+        asset = Asset.create(
+            project_id=library.project_id,
+            library_id=library.id,
+            name=asset_name,
+            storage_key=storage_key,
+            mime_type="audio/mpeg",
+            metadata={
+                "source": "video_extract",
+                "original_video": file.filename or "unknown",
+            },
+            file_size=file_size,
+            duration=duration,
+            status=AssetStatus.READY,
+            classification_status=ClassificationStatus.PENDING,
+            uploaded_by_user_id=user_id,
+        )
+        asset = asset_repository.create(asset)
+
+        return {
+            "id": asset.id,
+            "name": asset.name,
+            "audio_url": sign_url(storage_key),
+            "duration": duration,
+            "file_size": file_size,
+            "status": "completed",
+            "source": "video_extract",
+        }
+
+    except HTTPException:
+        raise
+    except subprocess.TimeoutExpired:
+        raise HTTPException(
+            status_code=status.HTTP_504_GATEWAY_TIMEOUT,
+            detail="视频处理超时，请尝试较短的视频",
+        )
+    except Exception as e:
+        logger.exception("提取视频配音失败: %s", e)
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="提取配音失败，请稍后重试",
+        )
+    finally:
+        # 清理临时文件
+        if tmp_dir and Path(tmp_dir).exists():
+            shutil.rmtree(tmp_dir, ignore_errors=True)
+
+
+def _find_or_create_voice_library_for_extract(*, user_id, project_repository, asset_library_repository):
+    """为用户找到或创建 voice 素材库（与 TTS 保存逻辑一致）。"""
+    projects = project_repository.find_accessible_projects(user_id)
+    if not projects:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="没有可用的项目，请先创建项目",
+        )
+
+    for project in projects:
+        for lib in asset_library_repository.find_by_project(project.id):
+            kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
+            if kind == AssetLibraryKind.VOICE.value:
+                return lib
+
+    # 自动创建
+    from sqlalchemy.exc import IntegrityError
+
+    project = projects[0]
+    library = AssetLibrary.create(
+        project_id=project.id,
+        name="配音素材库",
+        kind=AssetLibraryKind.VOICE,
+    )
+    try:
+        return asset_library_repository.create(library)
+    except IntegrityError:
+        session = getattr(asset_library_repository, "session", None)
+        if session is not None:
+            try:
+                session.rollback()
+            except Exception:
+                pass
+        for lib in asset_library_repository.find_by_project(project.id):
+            kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
+            if kind == AssetLibraryKind.VOICE.value:
+                return lib
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="配音素材库创建失败",
+        )
+
+
+def _get_audio_duration(audio_path: Path) -> float:
+    """用 ffprobe 获取音频时长（秒）。"""
+    try:
+        result = subprocess.run(
+            [
+                "ffprobe",
+                "-v",
+                "quiet",
+                "-show_entries",
+                "format=duration",
+                "-of",
+                "csv=p=0",
+                str(audio_path),
+            ],
+            capture_output=True,
+            timeout=10,
+        )
+        if result.returncode == 0 and result.stdout.strip():
+            return float(result.stdout.strip())
+    except (ValueError, subprocess.TimeoutExpired):
+        pass
+    return 0.0

apps/api/app/schemas/asset.py

@@ -53,6 +53,14 @@ class AssetResponse(BaseModel):
     created_at: str
     uploaded_by_user_id: str
     tag_ids: list[str] = Field(default_factory=list)
+    # 片段级余量信息（仅视频素材返回，非视频/无时长记录为 None，前端按可用处理）
+    used_duration: float | None = Field(default=None, description="已使用片段时长（秒，历史区间合并去重后）")
+    available_duration: float | None = Field(default=None, description="剩余可用时长（秒）= 素材总时长 - 已用时长")
+    used_ratio: float | None = Field(default=None, description="已用时长占比（0~1）")
+    usable: bool = Field(
+        default=True,
+        description="是否仍可用于新片段：零重复可切区间耗尽且所有历史区间复用次数" "（use_count）均达上限时为 false",
+    )
 
 
 MAX_BATCH_SIZE = 200
@@ -121,10 +129,13 @@ class SmartMatchRequest(BaseModel):
     )
 
 
-class SmartMatchItem(BaseModel):
-    """智能选素材结果条目。"""
+class SmartMatchItem(AssetResponse):
+    """智能选素材结果条目（扁平结构）。
+
+    素材字段（id/usable/余量等）直接挂在条目顶层，前端拿到 item 即可读 item.id，
+    与 AssetResponse 字段完全一致；score/breakdown 为智能匹配附加的评分字段。
+    """
 
-    asset: AssetResponse
     score: float = Field(..., ge=0, le=100, description="综合得分 0-100")
     breakdown: dict[str, float] = Field(default_factory=dict, description="各维度得分明细")
 

apps/api/app/schemas/generation_task.py

@@ -45,11 +45,9 @@ class CreateGenerationTaskRequest(BaseModel):
     # ── 素材库自动匹配 ──
     asset_select_mode: str = Field(
         default="all",
-        description="素材选取模式：all=全部ready视频, random=随机选取, smart=智能匹配（按质量/时长评分）",
-    )
-    asset_select_count: int = Field(
-        default=0, ge=0, le=100, description="选取数量，0表示全部（仅 random/smart 模式有效）"
+        description="素材选取模式：all=全部ready视频, smart=智能匹配（按质量/时长评分）",
     )
+    asset_select_count: int = Field(default=0, ge=0, le=100, description="选取数量，0表示全部（仅 smart 模式有效）")
     # ── 自动重试 ──
     auto_retry_enabled: bool = Field(
         default=False,

apps/api/app/schemas/video_center.py

@@ -22,6 +22,7 @@ class VideoItemResponse(BaseModel):
     generation_params: dict = Field(default_factory=dict)
     download_url: str | None = None
     generated_at: str = ""
+    duplicate_rate: float | None = None
 
 
 class ListVideosResponse(BaseModel):

apps/api/app/schemas/voice_clone.py

@@ -13,7 +13,8 @@ class CreateVoiceCloneRequest(BaseModel):
 
     name: str = Field(..., min_length=1, max_length=100, description="音色名称")
     description: str = Field("", description="音色描述")
-    source_audio_url: str = Field("", description="参考音频 URL")
+    source_audio_url: str = Field("", description="参考音频 URL（与 asset_id 二选一）")
+    asset_id: str = Field("", description="参考音频素材 ID（配音素材库中的音频 asset，与 source_audio_url 二选一）")
     voice_model: str = Field("", description="语音模型名称")
     language: str = Field("zh-CN", description="语言")
     gender: str = Field("unknown", description="性别")

apps/api/app/services/asset_segment_tracker.py

@@ -0,0 +1,494 @@
+"""素材片段级使用记录追踪与受控复用.
+
+在素材 metadata（assets.classification_result JSON）中持久化已使用的片段时间区间，
+供 from-assets 创建片段时避开历史区间，实现跨任务/跨调用的片段去重；
+素材可用区间耗尽后进入受控复用：允许有限次数（MAX_RANGE_USE_COUNT）复用最久未用
+的历史区间，配合调用方的成片复用占比控制（MAX_REUSE_RATIO = 10%），把任意两条
+成片的画面重复率控制在阈值内。
+
+metadata 中的记录字段 ``used_time_ranges``::
+
+    "used_time_ranges": [
+        {
+            "start": 12.5, "end": 20.3,
+            "plan_id": "plan-xxx",
+            "created_at": "2026-08-29T12:00:00+00:00",
+            "use_count": 1,              # 该区间累计被使用次数（复用一次 +1）
+            "last_used_at": "2026-08-29T12:00:00+00:00"  # 最近一次使用时间
+        },
+        ...
+    ]
+
+注意：本模块所有函数都不自行 commit，由调用方控制事务边界
+（from-assets 与 replace_all_clips_transactional 同事务；异步任务各自 commit）。
+历史记录永不自动清空（自动轮回重置已下线，reset_used_segments 仅保留给运维/测试）。
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from datetime import datetime, timezone
+from typing import Callable
+
+from sqlalchemy.orm import Session
+
+from packages.adapters.sqlalchemy_impl.models import AssetModel
+
+logger = logging.getLogger(__name__)
+
+USED_RANGES_KEY = "used_time_ranges"
+
+# ── 受控复用配置常量 ─────────────────────────────────────────────────────────
+MAX_RANGE_USE_COUNT = 2
+"""单条历史区间最多被使用次数（含首次），达到后不再参与复用。"""
+
+REUSE_RATIO_LIMIT = 0.10
+"""单条成片中，单个素材的复用片段累计时长 / 该素材在成片中的总时长上限（10%）。
+超过则该素材不再分配新片段（调用方在轮询分配时跳过）。"""
+
+SEGMENT_EDGE_GAP = 1.5
+"""冲突判定边缘间隙（秒）：历史区间按 [start-gap, end+gap] 扩边后参与冲突检测，
+避免两条片段首尾紧贴导致画面观感重复；记录仍存实际值。"""
+
+# 判定"新片段与历史区间为同一次使用（复用）"的重叠率阈值：
+# 重叠时长 / 新区间时长超过该比例视为复用该历史区间（累加 use_count）而非新增记录。
+_REUSE_OVERLAP_RATIO = 0.6
+
+
+def _now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def _read_meta(model) -> dict:
+    """读取素材 metadata dict。
+
+    兼容两种对象：
+    - ORM ``AssetModel``：metadata 以 JSON 字符串存在 ``classification_result`` 列；
+    - 领域实体 ``Asset``（路由层 repository 返回）：metadata 直接是 dict 属性
+      （repository 与 classification_result 互转，见 asset_repository.py）。
+    """
+    # 领域实体：metadata 已是 dict
+    meta = getattr(model, "metadata", None)
+    if isinstance(meta, dict):
+        return meta
+    raw = getattr(model, "classification_result", None)
+    if not raw:
+        return {}
+    try:
+        data = json.loads(raw) if isinstance(raw, str) else raw
+        return data if isinstance(data, dict) else {}
+    except Exception:
+        return {}
+
+
+def _get_model(db: Session, asset_id: str, for_update: bool = False) -> AssetModel | None:
+    query = db.query(AssetModel).filter(AssetModel.id == asset_id)
+    if for_update:
+        # 行级锁（PostgreSQL SELECT ... FOR UPDATE）：序列化同一素材的
+        # classification_result 读-改-写，避免并发事务丢失使用记录。
+        # SQLite 不支持时 SQLAlchemy 会忽略该子句（no-op）。
+        query = query.with_for_update()
+    return query.first()
+
+
+def get_used_segments(db: Session, asset_ids: list[str]) -> dict[str, list[tuple[float, float]]]:
+    """聚合多个素材的历史已用片段区间。
+
+    Returns:
+        ``{asset_id: [(start, end), ...]}`` 格式，与 ``_calc_random_start_time`` 的
+        ``used_segments`` 参数格式一致，可直接传入。
+    """
+    if not asset_ids:
+        return {}
+    result: dict[str, list[tuple[float, float]]] = {}
+    models = db.query(AssetModel).filter(AssetModel.id.in_(list(set(asset_ids)))).all()
+    for model in models:
+        meta = _read_meta(model)
+        ranges = meta.get(USED_RANGES_KEY) or []
+        segments: list[tuple[float, float]] = []
+        for r in ranges:
+            try:
+                segments.append((float(r["start"]), float(r["end"])))
+            except (KeyError, TypeError, ValueError):
+                continue
+        if segments:
+            result[model.id] = segments
+    return result
+
+
+def record_used_segments(
+    db: Session,
+    asset_id: str,
+    start: float,
+    end: float,
+    plan_id: str,
+) -> None:
+    """记录一次片段使用（不 commit）.
+
+    若新区间与某条历史区间高度重叠（复用场景，如受控复用回调返回的区间、
+    MediaKit 挪到历史区间），则累加该记录的 ``use_count`` 并刷新 ``last_used_at``，
+    不新增记录；否则追加一条新记录（use_count=1）。
+    """
+    # 行级锁读取：与并发生成任务互斥，保证区间记录读-改-写一致
+    model = _get_model(db, asset_id, for_update=True)
+    if model is None:
+        logger.warning("[片段追踪] 素材不存在，跳过记录: asset_id=%s", asset_id)
+        return
+    meta = _read_meta(model)
+    ranges = list(meta.get(USED_RANGES_KEY) or [])
+
+    new_start = round(float(start), 3)
+    new_end = round(float(end), 3)
+    new_dur = max(new_end - new_start, 1e-6)
+    now = _now_iso()
+
+    for r in ranges:
+        try:
+            rs, re_ = float(r["start"]), float(r["end"])
+        except (KeyError, TypeError, ValueError):
+            continue
+        overlap = max(0.0, min(new_end, re_) - max(new_start, rs))
+        if overlap / new_dur >= _REUSE_OVERLAP_RATIO:
+            # 复用同一条历史区间：累加次数、刷新时间
+            r["use_count"] = int(r.get("use_count", 1)) + 1
+            r["last_used_at"] = now
+            r["plan_id"] = plan_id
+            meta[USED_RANGES_KEY] = ranges
+            model.classification_result = json.dumps(meta, ensure_ascii=False)
+            model.updated_at = datetime.now(timezone.utc)
+            return
+
+    ranges.append(
+        {
+            "start": new_start,
+            "end": new_end,
+            "plan_id": plan_id,
+            "created_at": now,
+            "use_count": 1,
+            "last_used_at": now,
+        }
+    )
+    meta[USED_RANGES_KEY] = ranges
+    model.classification_result = json.dumps(meta, ensure_ascii=False)
+    model.updated_at = datetime.now(timezone.utc)
+
+
+def remove_used_segment(
+    db: Session,
+    asset_id: str,
+    start: float,
+    end: float,
+    plan_id: str | None = None,
+    tolerance: float = 0.5,
+) -> bool:
+    """删除素材 metadata 中匹配的一条使用记录（不 commit）.
+
+    匹配规则：start/end 与记录值相差不超过 tolerance 秒；plan_id 非空时，
+    记录有 plan_id 则需相等，记录缺 plan_id（本功能上线前的旧数据）时按时间匹配。
+    Returns:
+        是否找到并删除了记录。
+    """
+    model = _get_model(db, asset_id)
+    if model is None:
+        return False
+    meta = _read_meta(model)
+    ranges = list(meta.get(USED_RANGES_KEY) or [])
+    remaining: list[dict] = []
+    removed = False
+    for r in ranges:
+        try:
+            match = (
+                abs(float(r["start"]) - float(start)) <= tolerance and abs(float(r["end"]) - float(end)) <= tolerance
+            )
+        except (KeyError, TypeError, ValueError):
+            remaining.append(r)
+            continue
+        # plan_id 校验：传入 plan_id 时，记录有 plan_id 则必须相等；
+        # 记录本身缺 plan_id（旧数据）时退化为按时间匹配，避免旧区间永远删不掉
+        if plan_id is not None and r.get("plan_id") is not None and r.get("plan_id") != plan_id:
+            match = False
+        if match and not removed:
+            removed = True
+            continue
+        remaining.append(r)
+    if removed:
+        meta[USED_RANGES_KEY] = remaining
+        model.classification_result = json.dumps(meta, ensure_ascii=False)
+        model.updated_at = datetime.now(timezone.utc)
+    return removed
+
+
+def reset_used_segments(db: Session, asset_id: str) -> None:
+    """清空单个素材的历史片段使用记录（不 commit）.
+
+    仅供运维/测试使用；正常生成流程中历史记录永不自动清空（受控复用取代自动轮回）。
+    """
+    model = _get_model(db, asset_id)
+    if model is None:
+        return
+    meta = _read_meta(model)
+    if meta.get(USED_RANGES_KEY):
+        meta[USED_RANGES_KEY] = []
+        model.classification_result = json.dumps(meta, ensure_ascii=False)
+        model.updated_at = datetime.now(timezone.utc)
+        logger.info("[片段追踪] 素材区间记录手动清空: asset_id=%s", asset_id)
+
+
+# ── 素材余量/可用性计算（Task H：素材库角标 + smart-match 过滤）──────────────
+
+# 判定「是否还有空闲可切区间」时使用的最小片段时长（秒）：空闲段长于此值才视为可切
+_MIN_FREE_CLIP_DURATION = 3.0
+
+
+def _merge_intervals(intervals: list[tuple[float, float]]) -> list[tuple[float, float]]:
+    """合并重叠/相接的时间区间，返回升序不重叠区间列表。"""
+    if not intervals:
+        return []
+    ordered = sorted((float(a), float(b)) for a, b in intervals if b > a)
+    merged: list[tuple[float, float]] = [ordered[0]]
+    for start, end in ordered[1:]:
+        last_start, last_end = merged[-1]
+        if start <= last_end:
+            merged[-1] = (last_start, max(last_end, end))
+        else:
+            merged.append((start, end))
+    return merged
+
+
+def _has_free_gap(used: list[tuple[float, float]], total: float, min_free: float = _MIN_FREE_CLIP_DURATION) -> bool:
+    """素材 [0, total] 中是否存在长度 ≥ min_free 的空闲段（考虑边缘间隙）。"""
+    if total <= 0:
+        return False
+    # 历史区间按边缘间隙扩边后判定空闲（与选片冲突检测同一口径）
+    expanded = [(max(0.0, s - SEGMENT_EDGE_GAP), min(total, e + SEGMENT_EDGE_GAP)) for s, e in used]
+    merged = _merge_intervals(expanded)
+    cursor = 0.0
+    for start, end in merged:
+        if start - cursor >= min_free:
+            return True
+        cursor = max(cursor, end)
+    return total - cursor >= min_free
+
+
+def compute_asset_availability(
+    model: "AssetModel | None",
+    min_free_clip_duration: float = _MIN_FREE_CLIP_DURATION,
+) -> dict | None:
+    """计算单个素材的余量与可用性（纯函数，不读写 DB）。
+
+    Returns:
+        视频素材返回 ``{"used_duration", "available_duration", "used_ratio", "usable"}``；
+        非视频 / 无 model / 无时长信息返回 None（调用方按可用处理，零影响）。
+
+    usable=False 条件（与受控复用机制一致）：
+        零重复可切区间已耗尽（不存在 ≥ min_free 的空闲段）且
+        所有历史区间 use_count 均达 MAX_RANGE_USE_COUNT 上限（无区间可复用）。
+    """
+    if model is None:
+        return None
+    file_type = getattr(model, "file_type", None) or getattr(model, "mime_type", "") or ""
+    if file_type != "video" and not str(file_type).startswith("video/"):
+        return None
+    total = float(getattr(model, "duration", 0.0) or 0.0)
+    if total <= 0:
+        return None
+
+    meta = _read_meta(model)
+    raw_ranges = meta.get(USED_RANGES_KEY) or []
+
+    intervals: list[tuple[float, float]] = []
+    use_counts: list[int] = []
+    for r in raw_ranges:
+        try:
+            start = float(r["start"])
+            end = float(r["end"])
+        except (KeyError, TypeError, ValueError):
+            continue
+        if end <= start:
+            continue
+        intervals.append((start, end))
+        try:
+            use_counts.append(int(r.get("use_count", 1)))
+        except (TypeError, ValueError):
+            use_counts.append(1)
+
+    merged = _merge_intervals(intervals)
+    used_duration = round(sum(e - s for s, e in merged), 3)
+    used_duration = min(used_duration, total)
+    available_duration = round(max(total - used_duration, 0.0), 3)
+    used_ratio = round(min(used_duration / total, 1.0), 4)
+
+    has_free = _has_free_gap(intervals, total, min_free_clip_duration)
+    if has_free:
+        usable = True
+    else:
+        # 空闲段耗尽：仅当存在历史区间且全部达复用上限时才判定不可用；
+        # 无历史区间（理论上不会走到，因为 has_free=True）按可用处理
+        if not use_counts:
+            usable = True
+        else:
+            usable = any(uc < MAX_RANGE_USE_COUNT for uc in use_counts)
+
+    return {
+        "used_duration": used_duration,
+        "available_duration": available_duration,
+        "used_ratio": used_ratio,
+        "usable": usable,
+    }
+
+
+def find_reusable_range(
+    db: Session,
+    asset_id: str,
+    clip_duration: float,
+    asset_total: float,
+    *,
+    max_use_count: int = MAX_RANGE_USE_COUNT,
+) -> tuple[float, float] | None:
+    """受控复用：在素材历史区间中选一条可复用区间返回 (start, end)。
+
+    选择规则：
+    1. 仅选 ``use_count < max_use_count`` 的历史区间；
+    2. 优先返回能完整容纳当前 clip_duration（起点后不越素材边界）的最久未用区间；
+    3. 没有能容纳的，则返回 last_used_at 最老（或缺失 last_used_at 的旧数据优先）
+       且 use_count 最低的区间起点（可能与其他历史区间重叠，属降级复用）；
+    4. 无任何可复用区间（记录为空或全部达上限）返回 None。
+
+    本函数只读不写；复用次数的累加由后续 record_used_segments 完成。
+    """
+    model = _get_model(db, asset_id)
+    if model is None:
+        return None
+    meta = _read_meta(model)
+    ranges = [r for r in (meta.get(USED_RANGES_KEY) or []) if int(r.get("use_count", 1)) < max_use_count]
+    if not ranges:
+        return None
+
+    def _last_used(r: dict) -> str:
+        return str(r.get("last_used_at") or r.get("created_at") or "")
+
+    max_start = max(0.0, asset_total - clip_duration)
+    # 2. 能完整容纳当前片段的候选：按 last_used_at 升序（最久未用优先）
+    fit = sorted(
+        [r for r in ranges if float(r["start"]) <= max_start + 1e-6],
+        key=_last_used,
+    )
+    if fit:
+        start = min(float(fit[0]["start"]), max_start)
+        return (start, start + clip_duration)
+
+    # 3. 降级：最久未用 + use_count 最低的区间起点
+    fallback = sorted(ranges, key=lambda r: (_last_used(r), int(r.get("use_count", 1))))[0]
+    start = min(float(fallback["start"]), max_start)
+    return (start, start + clip_duration)
+
+
+def make_reuse_callback(
+    db: Session,
+    asset_durations: dict[str, float],
+    reused_tracker: dict[str, float] | None = None,
+    assigned_tracker: dict[str, float] | None = None,
+    ratio_limit: float = REUSE_RATIO_LIMIT,
+) -> Callable[[str, float], tuple[float, float] | None]:
+    """构造给 ``_calc_random_start_time`` 用的受控复用回调.
+
+    Args:
+        db: SQLAlchemy session
+        asset_durations: 素材 ID -> 总时长（回调需要素材总时长做边界约束）
+        reused_tracker: 可选的 ``{asset_id: 累计复用时长}``，回调成功返回复用区间时
+            会把本次片段时长累加进去，供调用方统计成片复用占比（10% 阈值）。
+        assigned_tracker: 可选的 ``{asset_id: 已分配片段总时长}``，配合 ratio_limit
+            在复用前预判：若复用本片段后占比 (reused + clip_duration) /
+            (assigned + clip_duration) 超过 ratio_limit，则拒绝复用、返回 None
+            （保证成片复用占比不超阈值）。
+        ratio_limit: 单条成片复用时长占比上限，默认 10%。
+
+    Returns:
+        回调函数 ``(asset_id, clip_duration) -> (start, end) | None``。
+        回调内吞掉 DB 异常返回 None，不影响主生成流程。
+    """
+
+    def _reuse(asset_id: str, clip_duration: float) -> tuple[float, float] | None:
+        try:
+            total = float(asset_durations.get(asset_id, 0.0) or 0.0)
+            if total <= 0:
+                return None
+            # 占比闸门：预判复用本片段后是否超限（仅当调用方提供了 assigned tracker）
+            if assigned_tracker is not None:
+                assigned = float(assigned_tracker.get(asset_id, 0.0) or 0.0)
+                reused_amt = float((reused_tracker or {}).get(asset_id, 0.0) or 0.0)
+                if assigned > 0 and (reused_amt + clip_duration) / (assigned + clip_duration) > ratio_limit:
+                    logger.info(
+                        "[片段追踪] 复用占比预判超 %.0f%% 阈值，拒绝复用: asset_id=%s "
+                        "reused=%.1f assigned=%.1f clip=%.1f",
+                        ratio_limit * 100,
+                        asset_id,
+                        reused_amt,
+                        assigned,
+                        clip_duration,
+                    )
+                    return None
+            result = find_reusable_range(db, asset_id, clip_duration, total)
+        except Exception:
+            logger.warning("[片段追踪] 受控复用查询异常: asset_id=%s", asset_id, exc_info=True)
+            return None
+        if result is not None and reused_tracker is not None:
+            reused_tracker[asset_id] = reused_tracker.get(asset_id, 0.0) + clip_duration
+        return result
+
+    return _reuse
+
+
+def get_asset_recent_use_counts(
+    db: Session,
+    asset_ids: list[str],
+    recent_video_count: int = 5,
+) -> dict[str, int]:
+    """统计每个素材在最近 N 个不同 plan_id 中的使用次数。
+
+    遍历素材 metadata 中的 used_time_ranges，统计有多少个不同的 plan_id（去重），
+    返回 {asset_id: count}。只统计最近 recent_video_count 个不同 plan_id 的使用次数。
+
+    Args:
+        db: 数据库会话
+        asset_ids: 素材 ID 列表
+        recent_video_count: 统计最近多少个不同 plan_id
+
+    Returns:
+        {asset_id: 在最近 recent_video_count 个 plan 中的使用次数}
+    """
+    if not asset_ids:
+        return {}
+
+    result: dict[str, int] = {}
+    models = db.query(AssetModel).filter(AssetModel.id.in_(asset_ids)).all()
+    for model in models:
+        meta = _read_meta(model)
+        ranges = meta.get(USED_RANGES_KEY) or []
+        if not ranges:
+            result[model.id] = 0
+            continue
+
+        # 按 created_at 倒序收集不同 plan_id
+        sorted_ranges = sorted(
+            ranges,
+            key=lambda r: r.get("created_at") or "",
+            reverse=True,
+        )
+        recent_plan_ids: set[str] = set()
+        for r in sorted_ranges:
+            plan_id = r.get("plan_id")
+            if plan_id:
+                recent_plan_ids.add(plan_id)
+            if len(recent_plan_ids) >= recent_video_count:
+                break
+
+        result[model.id] = len(recent_plan_ids)
+
+    # 未找到的素材计为 0
+    for aid in asset_ids:
+        if aid not in result:
+            result[aid] = 0
+
+    return result

apps/api/app/services/edit_plan_service.py

@@ -9,6 +9,12 @@ from __future__ import annotations
 import logging
 from typing import Any, Dict, List, Optional
 
+from app.services.asset_segment_tracker import (
+    REUSE_RATIO_LIMIT,
+    get_used_segments,
+    make_reuse_callback,
+    record_used_segments,
+)
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl import (
@@ -400,7 +406,7 @@ class EditPlanService:
                 order = clip_item.get("order") or i
                 clip = EditPlanClip.create(
                     plan_id=plan_id,
-                    clip_type="main",
+                    clip_type=clip_item.get("clip_type", "main"),
                     order=order,
                     asset_id=clip_item.get("asset_id", ""),
                     start_time=clip_item.get("start_time", 0.0),
@@ -453,6 +459,116 @@ class EditPlanService:
             logger.exception("事务性替换片段失败: plan_id=%s", plan_id)
             raise
 
+    def clone_plan_for_variant(
+        self,
+        source_plan_id: str,
+        *,
+        created_by_user_id: str = "",
+        name_suffix: str = "变体",
+        reuse_tracker: Optional[dict] = None,
+    ) -> EditPlan:
+        """为批量/多预览场景克隆一份独立 plan，片段起点全部重算（受控随机/复用）。
+
+        复制源 plan 的模板归属、config 与片段结构（asset_id / duration / clip_type /
+        order 不变），每个片段重新调用 ``_calc_random_start_time``：读取素材持久化的
+        历史已用区间避让，耗尽时受控复用（use_count<3、最久未用），从而保证 N 条
+        成片片段区间互不相同，且复用占比受控。
+
+        - 不替换/不修改源 plan，源 plan 保留用户手动编辑结果。
+        - 片段区间记录（record_used_segments）随新片段写入素材 metadata，与新 plan
+          同事务；复用历史区间时由 record 自动累加 use_count。
+        - 克隆的 clips 复用区间累计时长写入 reuse_tracker（可选），供调用方统计占比。
+
+        Raises:
+            ValueError: 源 plan 不存在或无可用片段。
+        """
+        from packages.adapters.sqlalchemy_impl.models import AssetModel
+        from packages.domain.plan_generator_utils import _calc_random_start_time
+
+        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，无法克隆变体: {source_plan_id}")
+
+        # 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 ""),
+        )
+
+        # 素材时长映射（O(N) 单查）
+        asset_ids = list({c.asset_id for c in clips if c.asset_id})
+        db = self._clip_repo.session
+        durations: dict[str, float] = {}
+        if asset_ids:
+            for m in db.query(AssetModel).filter(AssetModel.id.in_(asset_ids)).all():
+                durations[m.id] = float(getattr(m, "duration", 0.0) or 0.0)
+
+        used_segments = get_used_segments(db, asset_ids)
+        reused: dict[str, float] = reuse_tracker if reuse_tracker is not None else {}
+        asset_assigned: dict[str, float] = {}
+        # 回调内部预判复用后占比超 15% 则拒绝复用（calc 返回 None → 保留原起点）
+        reuse_cb = make_reuse_callback(db, durations, reused, assigned_tracker=asset_assigned)
+
+        clips_data: list[dict] = []
+        for i, c in enumerate(clips):
+            aid = c.asset_id
+            dur = float(c.duration or 0.0)
+            total = durations.get(aid, 0.0)
+            if aid and total > 0 and dur > 0:
+                # 复用占比闸门：本片段尚未分配，检查当前已复用占比
+                # reused / assigned 是否超 15%，超则该素材不再分配（保留原起点）；
+                # assigned=0（首个片段）放行，reused=0 时不误拦正常分配
+                assigned = asset_assigned.get(aid, 0.0)
+                eff_dur = min(dur, total)
+                reused_amt = reused.get(aid, 0.0)
+                ratio_blocked = assigned > 0 and reused_amt / assigned > REUSE_RATIO_LIMIT
+                start = None
+                if not ratio_blocked:
+                    start = _calc_random_start_time(aid, eff_dur, durations, used_segments, on_exhausted=reuse_cb)
+                if start is None:
+                    start = float(c.start_time or 0.0)
+                asset_assigned[aid] = assigned + eff_dur
+                used_segments.setdefault(aid, []).append((start, start + eff_dur))
+                record_used_segments(db, aid, start, start + eff_dur, new_plan.id)
+            else:
+                start = float(c.start_time or 0.0)
+
+            clips_data.append(
+                {
+                    "order": c.order if c.order is not None else i,
+                    "asset_id": aid,
+                    "start_time": start,
+                    "duration": dur,
+                    "clip_type": c.clip_type,
+                }
+            )
+
+        # 事务性写入新 plan 的片段（内部统一 commit/rollback）
+        self.replace_all_clips_transactional(new_plan.id, clips_data)
+        logger.info(
+            "克隆变体 plan: source=%s new=%s clips=%d",
+            source_plan_id,
+            new_plan.id,
+            len(clips_data),
+        )
+        return new_plan
+
     # ── 片段分割与合并 ──────────────────────────────────────────────────────
 
     def split_clip(self, clip_id: str, split_time: float) -> Dict[str, Any]:

apps/api/app/services/plan_generator_service.py

@@ -13,6 +13,7 @@
 from __future__ import annotations
 
 import logging
+import random
 from typing import Any, List
 
 from sqlalchemy.orm import Session
@@ -29,9 +30,11 @@ from packages.domain.editing_mode import EditingMode
 from packages.domain.plan_generator_utils import (
     create_clips_from_configs,
     distribute_assets,
+    extract_scene_points_from_metadata,
     generate_default_clips,
     map_clip_types_for_mode,
 )
+from packages.domain.smart_match import SCORE_RANDOM_NOISE_MAX, score_asset
 from packages.domain.template_clip_config import TemplateClipConfig
 
 logger = logging.getLogger(__name__)
@@ -118,9 +121,9 @@ class PlanGeneratorService:
 
         # 4. 按 editing_mode 分配素材
         if asset_ids:
-            # 如果是随机预览模式，获取素材时长信息
+            # 获取素材时长信息，用于随机起始时间
             asset_durations = None
-            if random_preview and self._asset_repo:
+            if self._asset_repo:
                 asset_durations = self._fetch_asset_durations(asset_ids)
             self._distribute_assets(
                 clips,
@@ -218,16 +221,69 @@ class PlanGeneratorService:
     ) -> None:
         """按 editing_mode 将素材分配到 clips（就地修改，未持久化）.
 
-        委托给 plan_generator_utils.distribute_assets 纯函数。
+        先用 smart_match 评分对素材排序（高分优先），再委托给
+        plan_generator_utils.distribute_assets 纯函数完成分配。
         """
+        # 预览随机模式：素材顺序已 shuffle，纯随机起点即可，不读 DB 评分/缓存
+        asset_scene_points: dict[str, list[float]] = {}
+        if not random_selection:
+            # 正式生成：smart_match 评分排序（高分优先）+ 场景切换点缓存
+            if self._asset_repo:
+                asset_ids = self._sort_assets_by_smart_score(asset_ids)
+                # 读取素材 metadata 中的场景切换点缓存（后台 SceneChange 检测写入）：
+                # 有缓存的素材片段起点从随机镜头段选取，无缓存走随机起点兜底
+                asset_scene_points = self._fetch_asset_scene_points(asset_ids)
+
+                # 正式生成也随机重排片段顺序（降重，默认开启无开关）
+                # smart_match 决定选哪些素材，shuffle 只改变分配到 clips 的顺序
+                asset_ids = list(asset_ids)  # 复制避免修改调用方原列表
+                random.shuffle(asset_ids)
+
         distribute_assets(
             clips,
             asset_ids,
             editing_mode,
             random_selection=random_selection,
             asset_durations=asset_durations,
+            asset_scene_points=asset_scene_points,
         )
 
+    def _fetch_asset_scene_points(self, asset_ids: List[str]) -> dict[str, list[float]]:
+        """从素材 metadata 读取场景切换点缓存（无缓存的素材不包含在结果中）。"""
+        points_map: dict[str, list[float]] = {}
+        if not self._asset_repo:
+            return points_map
+        for asset_id in asset_ids:
+            asset = self._asset_repo.get(asset_id)
+            if asset:
+                points = extract_scene_points_from_metadata(getattr(asset, "metadata", None))
+                if points:
+                    points_map[asset_id] = points
+        return points_map
+
+    def _sort_assets_by_smart_score(self, asset_ids: List[str]) -> List[str]:
+        """按 smart_match 综合评分降序排列素材 ID（注入随机噪声）。
+
+        评分高的素材（质量好、时长合适、新鲜、使用次数少）倾向排在前面；
+        排序时给每个素材的得分注入 0~SCORE_RANDOM_NOISE_MAX 的随机噪声，
+        使得分接近的素材排名每次浮动，避免一键生成反复选出相同素材组合，
+        从素材组合层面降低成片查重率。分差大于噪声上限时排名保持稳定。
+        """
+        scored: list[tuple[str, float]] = []
+        for asset_id in asset_ids:
+            asset = self._asset_repo.get(asset_id)
+            if asset:
+                score, _ = score_asset(asset)
+                scored.append((asset_id, score))
+            else:
+                scored.append((asset_id, 0.0))
+        # 评分 + 随机噪声后按降序排列
+        scored.sort(
+            key=lambda x: x[1] + random.uniform(0.0, SCORE_RANDOM_NOISE_MAX),
+            reverse=True,
+        )
+        return [aid for aid, _ in scored]
+
     def _fetch_asset_durations(self, asset_ids: List[str]) -> dict[str, float]:
         """从数据库获取素材时长信息.
 

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

@@ -52,7 +52,7 @@ type AssetListResponse = {
 test.describe("Core generation flow", () => {
   test.describe.configure({ timeout: 360_000 })
 
-  test("walks through 7-step wizard and starts generation", async ({ page, request }) => {
+  test("walks through 6-step wizard and starts generation", async ({ page, request }) => {
     test.setTimeout(360_000)
 
     await routeBrowserApiToTestApi(page)
@@ -185,15 +185,17 @@ test.describe("Core generation flow", () => {
     await expect(page.locator(".xx-choice-item.selected")).toBeVisible()
     await page.getByRole("button", { name: "下一步" }).click()
 
-    // Step 2: select material
+    // Step 2: select material (card grid UI)
     await expect(page.getByRole("heading", { name: /选择素材/ })).toBeVisible()
     const librarySelect = page.locator("select").first()
     await librarySelect.selectOption({ label: libraryName })
-    const materialLabel = page.getByText(sourceFileName).locator("..")
-    await expect(materialLabel.locator("input[type='checkbox']")).toBeVisible({
-      timeout: 10_000,
-    })
-    await materialLabel.locator("input[type='checkbox']").check()
+    // 新 UI: 素材以 9:16 竖屏卡片展示，点击卡片选中
+    // 注意：卡片中心是播放按钮（stopPropagation 会阻止选中），所以点击左上角避开
+    const materialCard = page.getByTestId("material-card").filter({ hasText: sourceFileName })
+    await expect(materialCard).toBeVisible({ timeout: 10_000 })
+    await materialCard.click({ position: { x: 15, y: 15 } })
+    // 验证选中：卡片应出现勾选标记（用 testid 定位，避免 ✓ 字符文本匹配不稳定）
+    await expect(materialCard.getByTestId("material-card-check")).toBeVisible({ timeout: 5_000 })
     await page.getByRole("button", { name: "下一步" }).click()
 
     // Step 3: voice (可选步骤，新注册用户无配音素材，直接跳过)
@@ -212,18 +214,14 @@ test.describe("Core generation flow", () => {
 
     const titleText = `E2E Test ${suffix}`
     await titleInput.fill(titleText)
-    await page.getByRole("button", { name: "下一步" }).click()
 
-    // Step 5: preview — 前端实时预览架构改造，无需后端生成预览
-    await expect(page.getByRole("heading", { name: /预览设置/ })).toBeVisible({ timeout: 15000 })
-    await page.getByRole("button", { name: "下一步" }).click()
-
-    // Step 6: cover (默认 AI 智能选帧模式，直接下一步)
-    await expect(page.getByRole("heading", { name: /选择封面/ })).toBeVisible({ timeout: 15000 })
-    await page.getByRole("button", { name: "下一步" }).click()
-
-    // Step 7: confirm and generate
-    await expect(page.getByRole("heading", { name: /确认生成/ })).toBeVisible()
+    // Step 4（标题+实时预览）：确认生成按钮已移到标题页，点击直接创建最终渲染任务
+    // 等待前端实时预览就绪：未就绪时右侧 FrontendPreviewPlayer 显示「准备预览素材...」占位，
+    // 就绪（previewReady：素材已解析 + 模板已选中）后占位消失；否则按钮会被校验拦截弹 warning
+    await page
+      .getByText("准备预览素材")
+      .waitFor({ state: "detached", timeout: 30_000 })
+      .catch(() => {})
 
     // Wait for generation API to be called
     // 前端直接创建生成任务：POST /generation/tasks
@@ -236,10 +234,10 @@ test.describe("Core generation flow", () => {
       { timeout: 30_000 },
     )
 
-    // Click generate button
-    await page.locator(".xx-btn-primary").filter({ hasText: "确认生成" }).first().click()
+    // 点击「确认生成视频」
+    await page.locator(".xx-btn-primary").filter({ hasText: "确认生成视频" }).first().click()
 
-    // Verify generation was triggered successfully
+    // Verify generation was triggered
     const genResp = await generatePromise
     if (!genResp.ok()) {
       const body = await genResp.text()
@@ -256,19 +254,42 @@ 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,
+      })
+
+      // 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）")
+      }
     } 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
+      await page
+        .getByText(/生成失败|重新生成/)
+        .isVisible({ timeout: 15_000 })
+        .catch(() => false)
     }
 
-    // Generation may fail in test env (no worker), that's OK
-    // Just verify the flow started - check page shows generation-related UI
-    await page
-      .getByText(/生成中|生成完成|生成失败/)
-      .isVisible({ timeout: 15_000 })
-      .catch(() => false)
-    // If we see progress or result, great; if not, flow still reached the end
-    // which is sufficient for an E2E smoke test
-
     // Verify product library page loads (smoke: just verify page renders)
     await page.goto("/app/products")
     await expect(page).toHaveURL(/\/app\/products/)

apps/web/e2e/test_asset.spec.ts

@@ -37,14 +37,33 @@ async function loginWithRetry(
   })
 }
 
+async function registerWithRetry(
+  request: APIRequestContext,
+  email: string,
+  username: string,
+  password: string,
+  displayName: string,
+  maxRetries = 2,
+) {
+  for (let i = 0; i <= maxRetries; i++) {
+    const response = await request.post(`${apiBase}/auth/register`, {
+      data: { email, password, username, display_name: displayName },
+    })
+    if (response.status() !== 429) return response
+    console.log(`[register] 触发限流，等待 65s 后重试 (${i + 1}/${maxRetries})`)
+    await new Promise((r) => setTimeout(r, 65000))
+  }
+  return request.post(`${apiBase}/auth/register`, {
+    data: { email, password, username, display_name: displayName },
+  })
+}
+
 /** 注册并登录，返回 { headers, email, username, userId } */
 async function createAuthedUser(request: APIRequestContext, label: string) {
   const email = uniqueEmail(label)
   const username = uniqueUsername(label)
 
-  const reg = await request.post(`${apiBase}/auth/register`, {
-    data: { email, password: PASSWORD, username, display_name: `E2E ${label}` },
-  })
+  const reg = await registerWithRetry(request, email, username, PASSWORD, `E2E ${label}`)
   expect(reg.ok(), `注册应成功: ${await reg.text()}`).toBeTruthy()
   const regData = await reg.json()
 

apps/web/package-lock.json

@@ -1848,10 +1848,9 @@
     },
     "node_modules/@testing-library/dom": {
       "version": "10.4.1",
-      "resolved": "https://registry.npmmirror.com/@testing-library/dom/-/dom-10.4.1.tgz",
+      "resolved": "https://registry.npmjs.org/@testing-library/dom/-/dom-10.4.1.tgz",
       "integrity": "sha512-o4PXJQidqJl82ckFaXUeoAW+XysPLauYI43Abki5hABd853iMhitooc6znOnczgbTYmEP6U6/y1ZyKAIsvMKGg==",
       "dev": true,
-      "license": "MIT",
       "peer": true,
       "dependencies": {
         "@babel/code-frame": "^7.10.4",
@@ -1938,10 +1937,9 @@
     },
     "node_modules/@types/aria-query": {
       "version": "5.0.4",
-      "resolved": "https://registry.npmmirror.com/@types/aria-query/-/aria-query-5.0.4.tgz",
+      "resolved": "https://registry.npmjs.org/@types/aria-query/-/aria-query-5.0.4.tgz",
       "integrity": "sha512-rfT93uj5s0PRL7EzccGMs3brplhcrghnDoV26NqKhCAS1hVo+WdNsPvE/yb6ilfr5hi2MEk6d5EWJTKdxg8jVw==",
       "dev": true,
-      "license": "MIT",
       "peer": true
     },
     "node_modules/@types/babel__core": {
@@ -3113,10 +3111,9 @@
     },
     "node_modules/dom-accessibility-api": {
       "version": "0.5.16",
-      "resolved": "https://registry.npmmirror.com/dom-accessibility-api/-/dom-accessibility-api-0.5.16.tgz",
+      "resolved": "https://registry.npmjs.org/dom-accessibility-api/-/dom-accessibility-api-0.5.16.tgz",
       "integrity": "sha512-X7BJ2yElsnOJ30pZF4uIIDfBEVgF4XEBxL9Bxhy6dnrm5hkzqmsWHGTiHqRiITNhMyFLyAiWndIJP7Z1NTteDg==",
       "dev": true,
-      "license": "MIT",
       "peer": true
     },
     "node_modules/dunder-proto": {
@@ -4457,10 +4454,9 @@
     },
     "node_modules/lz-string": {
       "version": "1.5.0",
-      "resolved": "https://registry.npmmirror.com/lz-string/-/lz-string-1.5.0.tgz",
+      "resolved": "https://registry.npmjs.org/lz-string/-/lz-string-1.5.0.tgz",
       "integrity": "sha512-h5bgJWpxJNswbU7qCrV0tIKQCaS3blPDrqKWx+QxzuzL1zGUzij9XCWLrSLsJPu5t+eWA/ycetzYAO5IOMcWAQ==",
       "dev": true,
-      "license": "MIT",
       "peer": true,
       "bin": {
         "lz-string": "bin/bin.js"
@@ -5008,10 +5004,9 @@
     },
     "node_modules/pretty-format": {
       "version": "27.5.1",
-      "resolved": "https://registry.npmmirror.com/pretty-format/-/pretty-format-27.5.1.tgz",
+      "resolved": "https://registry.npmjs.org/pretty-format/-/pretty-format-27.5.1.tgz",
       "integrity": "sha512-Qb1gy5OrP5+zDf2Bvnzdl3jsTf1qXVMazbvCoKhtKqVs4/YK4ozX4gKQJJVyNe+cajNPn0KoC0MC3FUmaHWEmQ==",
       "dev": true,
-      "license": "MIT",
       "peer": true,
       "dependencies": {
         "ansi-regex": "^5.0.1",
@@ -5024,10 +5019,9 @@
     },
     "node_modules/pretty-format/node_modules/ansi-styles": {
       "version": "5.2.0",
-      "resolved": "https://registry.npmmirror.com/ansi-styles/-/ansi-styles-5.2.0.tgz",
+      "resolved": "https://registry.npmjs.org/ansi-styles/-/ansi-styles-5.2.0.tgz",
       "integrity": "sha512-Cxwpt2SfTzTtXcfOlzGEee8O+c+MmUgGrNiBcXnuWxuFJHe6a5Hz7qwhwe5OgaSYI0IJvkLqWX1ASG+cJOkEiA==",
       "dev": true,
-      "license": "MIT",
       "peer": true,
       "engines": {
         "node": ">=10"
@@ -5735,10 +5729,9 @@
     },
     "node_modules/react-is": {
       "version": "17.0.2",
-      "resolved": "https://registry.npmmirror.com/react-is/-/react-is-17.0.2.tgz",
+      "resolved": "https://registry.npmjs.org/react-is/-/react-is-17.0.2.tgz",
       "integrity": "sha512-w2GsyukL62IJnlaff/nRegPQR94C/XXamvMWmSHRJ4y7Ts/4ocGRmTHvOs8PSE6pB3dWOrD/nueuU5sduBsQ4w==",
       "dev": true,
-      "license": "MIT",
       "peer": true
     },
     "node_modules/react-refresh": {

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

@@ -52,12 +52,40 @@ export const getAssetsByKind = async (
 /**
  * 智能匹配素材（后端 AI 选素材）
  * 调用后端 smart-match 端点，由后端根据素材库内容智能选择素材
+ *
+ * 后端返回 items 元素兼容两种结构（过渡期）：
+ * - 扁平结构：AssetItem 本身（id 在顶层）
+ * - 包装结构：{ asset: AssetItem, score, breakdown }（id 需从 .asset 取）
+ * 这里统一归一化为 AssetItem[]，调用方无需关心包装层。
  */
-export const smartMatchAssets = async (libraryId: string): Promise<{ items: AssetItem[] }> => {
-  const response = await apiClient.post("/assets/smart-match", {
-    library_id: libraryId,
-  })
-  return response.data
+export interface SmartMatchResult {
+  items: AssetItem[]
+}
+
+interface SmartMatchWrappedItem {
+  asset?: AssetItem
+  id?: string
+  score?: number
+  breakdown?: unknown
+}
+
+export const smartMatchAssets = async (
+  libraryId: string,
+  limit?: number,
+): Promise<SmartMatchResult> => {
+  const payload: Record<string, unknown> = { library_id: libraryId }
+  if (limit && limit > 0) payload.limit = limit
+  const response = await apiClient.post("/assets/smart-match", payload)
+  const rawItems: SmartMatchWrappedItem[] = response.data?.items ?? []
+  const items = rawItems
+    .map((it) =>
+      // 包装结构 { asset: {...} } 优先解包；否则视其本身为扁平 AssetItem
+      it?.asset && typeof it.asset === "object" && "id" in it.asset
+        ? it.asset
+        : (it as unknown as AssetItem),
+    )
+    .filter((it): it is AssetItem => !!it && typeof it.id === "string" && it.id.length > 0)
+  return { items }
 }
 
 /** 更新素材（名称、metadata 等） */

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

@@ -2,12 +2,17 @@
  * 素材诊断 API
  */
 import apiClient from "../client"
+import { getOrCreateDefaultProject } from "../projects"
 import type { AssetDiagnosis } from "./types"
 
 /** 获取素材诊断信息（可选 asset_id 查单素材，否则全局诊断） */
-export const getAssetDiagnosis = async (assetId?: string): Promise<AssetDiagnosis> => {
+export const getAssetDiagnosis = async (
+  assetId?: string,
+  projectId?: string,
+): Promise<AssetDiagnosis> => {
+  const pid = projectId ?? (await getOrCreateDefaultProject()).id
   const params: Record<string, string> = {}
   if (assetId) params.asset_id = assetId
-  const response = await apiClient.get("/asset-diagnosis", { params })
+  const response = await apiClient.get(`/projects/${pid}/asset-diagnosis`, { params })
   return response.data
 }

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

@@ -20,6 +20,10 @@ export type {
 // 素材诊断
 export { getAssetDiagnosis } from "./diagnosis"
 
+// 素材余量/可用性判断
+export { isAssetUsable } from "./usage"
+export type { AssetUsageLike } from "./usage"
+
 // 素材库
 export {
   getAssetLibraries,
@@ -39,7 +43,13 @@ export {
 } from "./assets"
 
 // 上传
-export { prepareDirectUpload, completeDirectUpload, uploadAssetDirect } from "./upload"
+export {
+  prepareDirectUpload,
+  completeDirectUpload,
+  uploadAssetDirect,
+  prepareDirectUploadHandle,
+  type DirectUploadHandle,
+} from "./upload"
 
 // 任务
 export { getIngestJob, submitClassificationJob, getClassificationJob } from "./jobs"

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

@@ -40,6 +40,10 @@ export interface AssetItem {
   thumbnail_url?: string
   /** 时长（秒），视频/音频素材由后端从 metadata 提取到顶层 */
   duration?: number
+  /** 已切片段占用时长占比（0~1，后端片段重复率控制机制返回；字段缺失视为未统计） */
+  used_ratio?: number | null
+  /** 是否已彻底用尽（无新区间且历史区间复用次数均达上限）；false 的素材不参与生成选片 */
+  usable?: boolean | null
   status?: string
   classification_status?: AssetClassificationStatus | null
   quality_score?: number | null
@@ -129,6 +133,12 @@ export interface DirectUploadPrepareResult {
   expires_at: string
   fields: Record<string, string>
   max_size_bytes: number
+  /**
+   * prepare 阶段预创建的素材记录 id（后端改造后返回：status=uploading）。
+   * 前端拿到后立即刷新列表，卡片以「上传中」态出现在素材网格中。
+   * 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
 }
 
 /** 直传完成确认返回 */
@@ -136,4 +146,8 @@ export interface DirectUploadCompleteResult {
   storage_key: string
   ingest_job_id: string
   url: string
+  /** 同库已存在相同 file_hash 的素材时为 true，ingest_job_id 为空 */
+  duplicated?: boolean
+  /** duplicated 为 true 时返回已存在素材的 id */
+  asset_id?: string
 }

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

@@ -27,28 +27,18 @@ export const completeDirectUpload = async (data: {
   return response.data
 }
 
-/** 直传上传（大文件推荐），支持可选进度回调 */
-export const uploadAssetDirect = async (data: {
-  file: File
-  library_id: string
-  onProgress?: (percent: number) => void
-}): Promise<DirectUploadCompleteResult> => {
-  const project = await getOrCreateDefaultProject()
+/** 直传 OSS 的底层传输（POST 表单到 OSS），带进度回调 */
+const putToOSS = (
+  prepared: DirectUploadPrepareResult,
+  file: File,
+  onProgress?: (percent: number) => void,
+): Promise<void> =>
+  new Promise<void>((resolve, reject) => {
+    const directForm = new FormData()
+    Object.entries(prepared.fields).forEach(([key, value]) => directForm.append(key, value))
+    directForm.append("file", file)
 
-  const prepared = await prepareDirectUpload({
-    project_id: project.id,
-    library_id: data.library_id,
-    filename: data.file.name,
-    content_type: data.file.type || "application/octet-stream",
-    file_size: data.file.size,
-  })
-
-  const directForm = new FormData()
-  Object.entries(prepared.fields).forEach(([key, value]) => directForm.append(key, value))
-  directForm.append("file", data.file)
-
-  // 使用 XMLHttpRequest 以获取上传进度 + 超时控制 + 详细错误诊断
-  await new Promise<void>((resolve, reject) => {
+    // 使用 XMLHttpRequest 以获取上传进度 + 超时控制 + 详细错误诊断
     const xhr = new XMLHttpRequest()
     xhr.open(prepared.method, prepared.upload_url)
 
@@ -56,8 +46,8 @@ export const uploadAssetDirect = async (data: {
     xhr.timeout = 10 * 60 * 1000
 
     xhr.upload.onprogress = (e) => {
-      if (e.lengthComputable && data.onProgress) {
-        data.onProgress(Math.round((e.loaded / e.total) * 100))
+      if (e.lengthComputable && onProgress) {
+        onProgress(Math.round((e.loaded / e.total) * 100))
       }
     }
     xhr.onload = () => {
@@ -102,9 +92,53 @@ export const uploadAssetDirect = async (data: {
     xhr.send(directForm)
   })
 
-  return completeDirectUpload({
+/** 单个文件的上传阶段信息（供批量上传队列做状态绑定） */
+export interface DirectUploadHandle {
+  /** prepare 返回（含可能的预建 asset_id） */
+  prepared: DirectUploadPrepareResult
+  /** 直传 OSS（可重复调用用于重试） */
+  transfer: (onProgress?: (percent: number) => void) => Promise<void>
+  /** 直传完成后调用 complete 确认入库 */
+  complete: () => Promise<DirectUploadCompleteResult>
+}
+
+/**
+ * 准备一次直传：调 prepare 拿到签名表单（后端可能同时预建 uploading 态 asset），
+ * 返回分段执行的 handle，调用方自行控制 transfer/complete 时机（便于队列并发与重试）。
+ */
+export const prepareDirectUploadHandle = async (data: {
+  file: File
+  library_id: string
+}): Promise<DirectUploadHandle> => {
+  const project = await getOrCreateDefaultProject()
+
+  const prepared = await prepareDirectUpload({
     project_id: project.id,
     library_id: data.library_id,
-    storage_key: prepared.storage_key,
+    filename: data.file.name,
+    content_type: data.file.type || "application/octet-stream",
+    file_size: data.file.size,
   })
+
+  return {
+    prepared,
+    transfer: (onProgress) => putToOSS(prepared, data.file, onProgress),
+    complete: () =>
+      completeDirectUpload({
+        project_id: project.id,
+        library_id: data.library_id,
+        storage_key: prepared.storage_key,
+      }),
+  }
+}
+
+/** 直传上传（大文件推荐），支持可选进度回调；一次性完成 prepare→transfer→complete */
+export const uploadAssetDirect = async (data: {
+  file: File
+  library_id: string
+  onProgress?: (percent: number) => void
+}): Promise<DirectUploadCompleteResult> => {
+  const handle = await prepareDirectUploadHandle({ file: data.file, library_id: data.library_id })
+  await handle.transfer(data.onProgress)
+  return handle.complete()
 }

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

@@ -0,0 +1,18 @@
+/**
+ * 素材余量/可用性判断
+ * 后端片段重复率控制机制（任意两条成片画面重复率 ≤15%）上线后，
+ * 素材列表会附加 usable / used_ratio 字段。字段未上线前一律按可用处理。
+ */
+
+/** 仅依赖素材余量相关字段的最小结构，api 层与 pages 层 AssetItem 均可传入 */
+export interface AssetUsageLike {
+  usable?: boolean | null
+  used_ratio?: number | null
+}
+
+/**
+ * 素材是否仍可参与生成选片。
+ * usable === false 表示已彻底用尽（无新区间且复用次数全部达上限）；
+ * 字段缺失（undefined/null）时降级为可用，保证后端字段上线前零影响。
+ */
+export const isAssetUsable = (asset: AssetUsageLike): boolean => asset.usable !== false

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

@@ -12,7 +12,14 @@ export interface GenerateCoverTitleConfig {
 }
 
 export interface GenerateCoverRequest {
-  asset_ids: string[]
+  /**
+   * 封面源视频标识（二选一）：
+   * - generated_video_id：确认生成任务产出的最终视频 ID
+   * - video_url：最终视频 URL（兜底）
+   * 后端根据此标识定位最终成片文件并抽帧，MediaKit 选帧逻辑不变
+   */
+  generated_video_id?: string
+  video_url?: string
   cover_type?: "ai_frame" | "manual" | "upload" | "ai_regenerate"
   frame_time?: number
   /** 标题样式，用于在封面上叠加标题文字 */
@@ -31,7 +38,7 @@ export interface GenerateCoverResponse {
   }
 }
 
-/** AI 生成封面 — 从预览视频中抽帧 */
+/** AI 生成封面 — 从最终成片中抽帧（MediaKit 选帧） */
 export async function generateCover(
   templateId: string,
   data: GenerateCoverRequest,

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

@@ -71,4 +71,5 @@ export interface VideoItem {
   generation_params: Record<string, unknown>
   download_url: string
   generated_at: string
+  duplicate_rate?: number
 }

apps/web/src/api/products/utils.ts

@@ -29,7 +29,6 @@ export function mapVideoToProductItem(video: VideoItem): ProductItem {
     // 后端字段名为 generated_at，映射为 created_at 供前端统一使用
     created_at: video.generated_at,
     updated_at: video.generated_at,
-    // 后端 /videos 接口暂无 duplicate_rate 字段
-    duplicate_rate: undefined,
+    duplicate_rate: video.duplicate_rate,
   }
 }

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

@@ -90,10 +90,21 @@ export async function createClipsFromAssets(
   templateId: string,
   assetIds: string[],
   clipType = "main",
+  requiredClipsCount?: number,
+  opts?: { signal?: AbortSignal },
 ): Promise<ClipsFromAssetsResponse> {
+  const body: Record<string, unknown> = {
+    asset_ids: assetIds,
+    clip_type: clipType,
+  }
+  if (requiredClipsCount !== undefined) {
+    body.required_clips_count = requiredClipsCount
+  }
+  // from-assets 后端会调用 MediaKit 智能选片（最长 60s），单独延长超时
   const response = await apiClient.post<ClipsFromAssetsResponse>(
     `/templates/${templateId}/editor/clips/from-assets`,
-    { asset_ids: assetIds, clip_type: clipType },
+    body,
+    { timeout: 60000, signal: opts?.signal },
   )
   return response.data
 }

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

@@ -114,8 +114,6 @@ export interface EditPlanConfig {
   auto_subtitles?: boolean
   /** 是否启用 BGM */
   bgm?: boolean
-  /** 生成数量 */
-  generate_count?: number
   /** 素材模式 */
   material_mode?: string
   /** 前端标题设置（Step4 自动保存，与 title_config 字段分离，不影响后端渲染） */

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

@@ -28,4 +28,5 @@ export {
   deleteTTSJob,
   getTtsVoices,
   previewTts,
+  extractVideoVoice,
 } from "./jobs"

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

@@ -70,3 +70,56 @@ export const previewTts = async (data: TTSPreviewRequest): Promise<TTSPreviewRes
   const response = await apiClient.post<TTSPreviewResponse>("/tts/preview", data)
   return response.data
 }
+
+/**
+ * 从视频中提取配音（上传视频 → 后端提取人声 → 保存到配音素材库）
+ * 支持 mp4/mov/webm 格式
+ */
+export const extractVideoVoice = async (
+  file: File,
+  onProgress?: (percent: number) => void,
+): Promise<{ asset_id: string; duration: number }> => {
+  const formData = new FormData()
+  formData.append("file", file)
+
+  return new Promise((resolve, reject) => {
+    const xhr = new XMLHttpRequest()
+    xhr.open("POST", "/api/v1/tts/extract-video-voice")
+
+    // 携带认证 token（从 localStorage 获取，与 apiClient 拦截器一致）
+    const token = localStorage.getItem("access_token")
+    if (token) {
+      xhr.setRequestHeader("Authorization", `Bearer ${token}`)
+    }
+
+    xhr.timeout = 10 * 60 * 1000 // 10 分钟超时
+
+    xhr.upload.onprogress = (e) => {
+      if (e.lengthComputable && onProgress) {
+        onProgress(Math.round((e.loaded / e.total) * 100))
+      }
+    }
+
+    xhr.onload = () => {
+      if (xhr.status >= 200 && xhr.status < 300) {
+        try {
+          resolve(JSON.parse(xhr.responseText))
+        } catch {
+          reject(new Error("服务器返回数据解析失败"))
+        }
+      } else {
+        try {
+          const err = JSON.parse(xhr.responseText)
+          reject(new Error(err.detail || err.message || `提取失败: HTTP ${xhr.status}`))
+        } catch {
+          reject(new Error(`提取失败: HTTP ${xhr.status}`))
+        }
+      }
+    }
+
+    xhr.onerror = () => reject(new Error("网络错误，请检查网络连接"))
+    xhr.ontimeout = () => reject(new Error("上传超时（10分钟），请检查网络或尝试更小的文件"))
+
+    xhr.send(formData)
+  })
+}

apps/web/src/api/voice-clone/clones.ts

@@ -44,14 +44,19 @@ export const getVoiceCloneDetail = async (id: string): Promise<VoiceCloneProfile
   return response.data
 }
 
-/** 创建克隆音色 */
+/** 创建克隆音色（audio_url 与 asset_id 二选一） */
 export const createVoiceClone = async (
   data: CreateVoiceCloneRequest,
 ): Promise<VoiceCloneProfile> => {
   const payload: CreateVoiceCloneRequestFull = {
     name: data.name,
     description: data.description,
-    source_audio_url: data.audio_url,
+  }
+  // 从配音素材选择克隆：直接传 asset_id，后端用素材 OSS 路径克隆
+  if (data.asset_id) {
+    payload.asset_id = data.asset_id
+  } else {
+    payload.source_audio_url = data.audio_url
   }
   const response = await apiClient.post<VoiceCloneProfile>("/voice-clones", payload)
   return response.data

apps/web/src/api/voice-clone/types.ts

@@ -22,10 +22,13 @@ export interface VoiceClone {
   updated_at: string
 }
 
-/** 创建克隆请求（前端简化版） */
+/** 创建克隆请求（前端简化版：audio_url 与 asset_id 二选一） */
 export interface CreateVoiceCloneRequest {
   name: string
-  audio_url: string
+  /** 录音/文件上传后的音频 URL（与 asset_id 二选一） */
+  audio_url?: string
+  /** 从配音素材选择时直接传素材 ID，后端用素材 OSS 路径克隆（与 audio_url 二选一） */
+  asset_id?: string
   description?: string
 }
 
@@ -72,11 +75,13 @@ export interface VoiceCloneStatusResponse {
   retry_count: number
 }
 
-/** 后端创建克隆请求（完整版） */
+/** 后端创建克隆请求（完整版：source_audio_url 与 asset_id 二选一） */
 export interface CreateVoiceCloneRequestFull {
   name: string
   description?: string
-  source_audio_url: string
+  source_audio_url?: string
+  /** 从配音素材选择克隆时传素材 ID */
+  asset_id?: string
   voice_model?: string
   language?: string
   gender?: string

apps/web/src/components/asset-selector/AssetCard.tsx

@@ -59,8 +59,23 @@ const AssetCard: React.FC<AssetCardProps> = ({
     >
       {/* 缩略图 */}
       <div className="as-card-thumb">
-        {asset.thumbnail_url ? (
-          <img src={asset.thumbnail_url} alt={asset.name} loading="lazy" />
+        {asset.thumbnail_url && !asset.thumbnail_url.match(/\.(mp4|mov|avi|webm|mkv)(\?|$)/i) ? (
+          <img
+            src={asset.thumbnail_url}
+            alt={asset.name}
+            loading="lazy"
+            onError={(e) => {
+              // 图片加载失败时降级显示类型图标
+              const parent = (e.target as HTMLElement).parentElement
+              if (parent) {
+                ;(e.target as HTMLElement).style.display = "none"
+                const icon = document.createElement("span")
+                icon.className = "as-card-thumb-icon"
+                icon.textContent = MATERIAL_TYPE_ICONS[asset.type] || "🎬"
+                parent.appendChild(icon)
+              }
+            }}
+          />
         ) : (
           <span className="as-card-thumb-icon">{MATERIAL_TYPE_ICONS[asset.type]}</span>
         )}

apps/web/src/components/voice/CloneModal/clone-modal.css

@@ -149,53 +149,29 @@
 
 /* ── 上传区域 ───────────────────────────────────────────── */
 
-.xx-clonemodal-upload-zone {
-  border: 2px dashed var(--xx-color-border, #e5e7eb);
+/* ── 素材选择空态 ─────────────────────────────────────────── */
+
+.xx-clonemodal-asset-empty {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+  padding: 14px 16px;
+  border: 1px dashed var(--xx-color-border, #e5e7eb);
   border-radius: var(--radius-sm);
-  padding: 28px 20px;
-  text-align: center;
-  cursor: pointer;
-  transition: all 0.2s ease;
   background: var(--xx-color-bg-secondary, #f9fafb);
 }
 
-.xx-clonemodal-upload-zone:hover {
-  border-color: var(--xx-color-primary, #6366f1);
-  background: rgba(99, 102, 241, 0.03);
-}
-
-.xx-clonemodal-upload-zone--active {
-  border-color: var(--xx-color-primary, #6366f1);
-  background: rgba(99, 102, 241, 0.06);
-}
-
-.xx-clonemodal-upload-zone--has-file {
-  border-style: solid;
-  border-color: var(--xx-color-primary, #6366f1);
-  background: rgba(99, 102, 241, 0.04);
-}
-
-.xx-clonemodal-upload-icon {
-  font-size: 32px;
-  margin-bottom: 8px;
-}
-
-.xx-clonemodal-upload-title {
-  font-size: 14px;
-  font-weight: 500;
-  color: var(--xx-color-text, #111827);
-  margin: 0 0 4px;
-  word-break: break-all;
-}
-
-.xx-clonemodal-upload-hint {
-  font-size: 12px;
-  color: var(--xx-color-text-secondary, #6b7280);
+.xx-clonemodal-asset-empty-text {
   margin: 0;
+  font-size: 13px;
+  color: var(--xx-color-text-secondary, #6b7280);
 }
 
 /* ── 错误提示 ───────────────────────────────────────────── */
 
+/* ── 错误提示 ───────────────────────────────────────────── */
+
 .xx-clonemodal-error {
   display: flex;
   align-items: center;

apps/web/src/components/voice/CloneModal/constants.ts

@@ -7,14 +7,5 @@ export const PROGRESS_STEPS: ProgressStep[] = [
   { key: "done", label: "完成", icon: "✅" },
 ]
 
-/** 支持的音频扩展名 */
-export const ACCEPTED_EXTENSIONS = ["mp3", "wav", "m4a", "webm"]
-
-/** 文件选择器 accept 属性 */
-export const ACCEPTED_MIME = ".mp3,.wav,.m4a,.webm,audio/mpeg,audio/wav,audio/mp4,audio/webm"
-
-/** 最大文件大小：10MB */
-export const MAX_FILE_SIZE = 10 * 1024 * 1024
-
 /** 最长录制时长：5 分钟（秒） */
 export const MAX_RECORD_SECONDS = 5 * 60

apps/web/src/components/voice/CloneModal/index.tsx

@@ -1,10 +1,12 @@
 import React, { useState, useCallback, useRef, useEffect } from "react"
+import { useNavigate } from "react-router-dom"
+import { useQuery } from "@tanstack/react-query"
 import { Modal, Button } from "@/components/ui"
 import { createVoiceClone, toVoiceClone } from "@/api/voice-clone"
-import { uploadAssetDirect, ensureDefaultLibrary } from "@/api/assets"
+import { uploadAssetDirect, ensureDefaultLibrary, getAssetsByKind } from "@/api/assets"
 import { getOrCreateDefaultProject } from "@/api/projects"
-import { PROGRESS_STEPS, ACCEPTED_MIME } from "./constants"
-import { validateFile } from "./utils"
+import { PROGRESS_STEPS } from "./constants"
+import { formatRecordTime } from "./utils"
 import { useAudioRecorder } from "./hooks/useAudioRecorder"
 import type { CloneModalProps, ModalPhase } from "./types"
 import "./clone-modal.css"
@@ -18,15 +20,21 @@ const getExtensionFromMime = (mime: string): string => {
   return "webm"
 }
 
+/** 格式化素材时长（秒 → mm:ss） */
+const formatAssetDuration = (seconds?: number): string => {
+  if (!seconds || seconds <= 0) return "--:--"
+  return formatRecordTime(Math.round(seconds))
+}
+
 const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) => {
+  const navigate = useNavigate()
   const [phase, setPhase] = useState<ModalPhase>("input")
   const [voiceName, setVoiceName] = useState("")
   const [voiceDescription, setVoiceDescription] = useState("")
-  const [selectedFile, setSelectedFile] = useState<File | null>(null)
-  const [dragActive, setDragActive] = useState(false)
+  /** 从配音素材选择的素材 ID */
+  const [selectedAssetId, setSelectedAssetId] = useState<string>("")
   const [errorMessage, setErrorMessage] = useState("")
 
-  const fileInputRef = useRef<HTMLInputElement>(null)
   const timerRef = useRef<ReturnType<typeof setTimeout> | null>(null)
   /** 默认音色名称计数器（组件级 ref，避免多实例串号） */
   const cloneCounterRef = useRef(1)
@@ -34,6 +42,14 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
   const isMountedRef = useRef(true)
   const isSubmittingRef = useRef(false)
 
+  /* ── 配音素材列表（「从配音素材选择」；弹窗打开时才发请求） ────── */
+  const { data: voiceAssets, isLoading: assetsLoading } = useQuery({
+    queryKey: ["assets", "voice", "clone-modal"],
+    queryFn: () => getAssetsByKind("voice", { limit: 100 }),
+    enabled: open,
+    staleTime: 30_000,
+  })
+
   /* ── 录音 Hook ──────────────────────────────────── */
   const {
     isRecording,
@@ -55,10 +71,9 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
     setPhase("input")
     setVoiceName(getNextDefaultName())
     setVoiceDescription("")
-    setSelectedFile(null)
-    setDragActive(false)
-    if (isSubmittingRef.current) return
-    isSubmittingRef.current = true
+    setSelectedAssetId("")
+    // 注意：resetState 不得触碰 isSubmittingRef——提交锁仅属于 handleSubmit；
+    // 此前在此上锁且无复位路径，弹窗打开即死锁
     setErrorMessage("")
     resetRecorder()
   }, [getNextDefaultName, resetRecorder])
@@ -85,65 +100,26 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
     }
   }, [])
 
-  /* ── 文件上传 ──────────────────────────────────── */
+  /* ── 素材/录音互斥：选择素材时清掉录音，开始录音时清掉素材选择 ── */
 
-  const handleUploadClick = () => {
-    fileInputRef.current?.click()
-  }
-
-  const handleFileChange = (e: React.ChangeEvent<HTMLInputElement>) => {
-    const file = e.target.files?.[0]
-    if (file) {
-      const error = validateFile(file)
-      if (error) {
-        setErrorMessage(error)
-        setSelectedFile(null)
-      } else {
-        if (isSubmittingRef.current) return
-        isSubmittingRef.current = true
-        setErrorMessage("")
-        setSelectedFile(file)
-        resetRecorder()
-      }
-    }
-    e.target.value = ""
-  }
-
-  /* ── 拖拽 ──────────────────────────────────────── */
-
-  const handleDrag = (e: React.DragEvent) => {
-    e.preventDefault()
-    e.stopPropagation()
-    if (e.type === "dragenter" || e.type === "dragover") {
-      setDragActive(true)
-    } else if (e.type === "dragleave") {
-      setDragActive(false)
+  const handleSelectAsset = (assetId: string) => {
+    setSelectedAssetId(assetId)
+    if (assetId) {
+      resetRecorder()
     }
   }
 
-  const handleDrop = (e: React.DragEvent) => {
-    e.preventDefault()
-    e.stopPropagation()
-    setDragActive(false)
-    const file = e.dataTransfer.files?.[0]
-    if (file) {
-      const error = validateFile(file)
-      if (error) {
-        setErrorMessage(error)
-        setSelectedFile(null)
-      } else {
-        if (isSubmittingRef.current) return
-        isSubmittingRef.current = true
-        setErrorMessage("")
-        setSelectedFile(file)
-        resetRecorder()
-      }
+  const handleToggleRecord = () => {
+    // 开始录音会清掉已选素材；停止录音保留录音结果
+    if (!isRecording) {
+      setSelectedAssetId("")
     }
+    toggleRecord()
   }
 
   /* ── 计算属性 ──────────────────────────────────── */
 
-  const hasAudio = selectedFile !== null || recordedBlob !== null
+  const hasAudio = selectedAssetId !== "" || recordedBlob !== null
   const isProcessing = phase === "uploading" || phase === "cloning"
   const canSubmit = hasAudio && !isProcessing
 
@@ -158,7 +134,7 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
       return
     }
     if (!hasAudio) {
-      setErrorMessage("请上传音频文件或录制一段声音")
+      setErrorMessage("请从配音素材选择一段音频，或直接录制声音")
       return
     }
 
@@ -167,20 +143,37 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
     setErrorMessage("")
 
     try {
-      // 阶段 1：上传音频
+      // 路径 A：从配音素材选择 → 无需上传，直接克隆
+      if (selectedAssetId) {
+        setPhase("cloning")
+        const result = await createVoiceClone({
+          name,
+          description: voiceDescription.trim() || undefined,
+          asset_id: selectedAssetId,
+        })
+
+        if (!isMountedRef.current) return
+
+        isSubmittingRef.current = false
+        setPhase("done")
+        timerRef.current = setTimeout(() => {
+          if (isMountedRef.current) {
+            onSuccess?.(toVoiceClone(result))
+            handleClose()
+          }
+        }, 2000)
+        return
+      }
+
+      // 路径 B：录音 → 先上传为配音素材，再克隆
       setPhase("uploading")
 
-      let fileToUpload: File
-      if (selectedFile) {
-        fileToUpload = selectedFile
-      } else {
-        // 使用浏览器实际生成的 MIME 类型，避免跨浏览器格式不匹配
-        const mimeType = recordedBlob?.type || "audio/webm"
-        const ext = getExtensionFromMime(mimeType)
-        fileToUpload = new File([recordedBlob!], `recorded-${Date.now()}.${ext}`, {
-          type: mimeType,
-        })
-      }
+      // 使用浏览器实际生成的 MIME 类型，避免跨浏览器格式不匹配
+      const mimeType = recordedBlob?.type || "audio/webm"
+      const ext = getExtensionFromMime(mimeType)
+      const fileToUpload = new File([recordedBlob!], `recorded-${Date.now()}.${ext}`, {
+        type: mimeType,
+      })
 
       // 获取默认项目和素材库
       const project = await getOrCreateDefaultProject()
@@ -226,6 +219,8 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
     }
   }
 
+  const hasAssets = (voiceAssets?.length ?? 0) > 0
+
   return (
     <Modal
       open={open}
@@ -244,7 +239,7 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
           <div className="xx-clonemodal-steps">
             <div className="xx-clonemodal-step xx-clonemodal-step--active">
               <div className="xx-clonemodal-step-number">1</div>
-              <span className="xx-clonemodal-step-label">上传/录制音频</span>
+              <span className="xx-clonemodal-step-label">选择/录制音频</span>
             </div>
             <div className="xx-clonemodal-step-connector" />
             <div className="xx-clonemodal-step">
@@ -274,30 +269,42 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
             <div className="xx-clonemodal-char-count">{voiceName.length}/20</div>
           </div>
 
-          {/* 上传区域 */}
+          {/* 从配音素材选择 */}
           <div className="xx-clonemodal-field">
-            <label className="xx-clonemodal-label">上传音频</label>
-            <div
-              className={`xx-clonemodal-upload-zone${dragActive ? " xx-clonemodal-upload-zone--active" : ""}${selectedFile ? " xx-clonemodal-upload-zone--has-file" : ""}`}
-              onClick={handleUploadClick}
-              onDragEnter={handleDrag}
-              onDragOver={handleDrag}
-              onDragLeave={handleDrag}
-              onDrop={handleDrop}
-            >
-              <div className="xx-clonemodal-upload-icon">{selectedFile ? "📄" : "🎵"}</div>
-              <p className="xx-clonemodal-upload-title">
-                {selectedFile ? selectedFile.name : "拖拽音频文件到此处，或点击上传"}
-              </p>
-              <p className="xx-clonemodal-upload-hint">支持 MP3、WAV、M4A 格式，最大 10MB</p>
-              <input
-                ref={fileInputRef}
-                type="file"
-                accept={ACCEPTED_MIME}
-                style={{ display: "none" }}
-                onChange={handleFileChange}
-              />
-            </div>
+            <label className="xx-clonemodal-label">从配音素材选择</label>
+            {hasAssets ? (
+              <select
+                className="xx-clonemodal-input"
+                value={selectedAssetId}
+                onChange={(e) => handleSelectAsset(e.target.value)}
+                disabled={assetsLoading}
+              >
+                <option value="">{assetsLoading ? "素材加载中…" : "请选择已上传的配音素材"}</option>
+                {voiceAssets!.map((asset) => (
+                  <option key={asset.id} value={asset.id}>
+                    {asset.name}（{formatAssetDuration(asset.duration)}）
+                  </option>
+                ))}
+              </select>
+            ) : (
+              <div className="xx-clonemodal-asset-empty">
+                <p className="xx-clonemodal-asset-empty-text">
+                  {assetsLoading ? "素材加载中…" : "请先在配音库上传素材"}
+                </p>
+                {!assetsLoading && (
+                  <Button
+                    buttonType="ghost"
+                    buttonSize="sm"
+                    onClick={() => {
+                      handleClose()
+                      navigate("/app/voices?tab=material&upload=1")
+                    }}
+                  >
+                    去配音库上传
+                  </Button>
+                )}
+              </div>
+            )}
           </div>
 
           {/* 或分隔 */}
@@ -332,7 +339,7 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
               <button
                 type="button"
                 className={`xx-clonemodal-record-btn${isRecording ? " xx-clonemodal-record-btn--recording" : ""}`}
-                onClick={toggleRecord}
+                onClick={handleToggleRecord}
                 title={isRecording ? "停止录制" : "开始录制"}
               >
                 {isRecording ? "⏹" : "🎙️"}
@@ -365,7 +372,7 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
           {/* 提示 */}
           <div className="xx-clonemodal-tip">
             <span className="xx-clonemodal-tip-icon">💡</span>
-            <span>建议上传 10 秒 ~ 3 分钟的清晰语音，环境安静、语速均匀效果最佳</span>
+            <span>建议使用 10 秒 ~ 3 分钟的清晰语音，环境安静、语速均匀效果最佳</span>
           </div>
 
           {/* 底部按钮 */}
@@ -415,18 +422,18 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
           {/* 当前阶段描述 */}
           <div className="xx-clonemodal-progress-info">
             {phase === "uploading" && (
-              <>
+              <div>
                 <div className="xx-clonemodal-progress-spinner" />
-                <p className="xx-clonemodal-progress-text">正在上传音频文件…</p>
-                <p className="xx-clonemodal-progress-sub">请稍候，正在将音频上传至服务器</p>
-              </>
+                <p className="xx-clonemodal-progress-text">正在上传录音…</p>
+                <p className="xx-clonemodal-progress-sub">请稍候，正在将录音上传至服务器</p>
+              </div>
             )}
             {phase === "cloning" && (
-              <>
+              <div>
                 <div className="xx-clonemodal-progress-spinner xx-clonemodal-progress-spinner--cloning" />
                 <p className="xx-clonemodal-progress-text">AI 正在克隆你的声音…</p>
                 <p className="xx-clonemodal-progress-sub">正在分析声音特征，生成专属音色模型</p>
-              </>
+              </div>
             )}
           </div>
         </div>

apps/web/src/components/voice/CloneModal/utils.ts

@@ -1,20 +1,3 @@
-import { ACCEPTED_EXTENSIONS, MAX_FILE_SIZE } from "./constants"
-
-/**
- * 验证音频文件
- * @returns 错误信息，null 表示验证通过
- */
-export const validateFile = (file: File): string | null => {
-  const ext = file.name.split(".").pop()?.toLowerCase()
-  if (!ext || !ACCEPTED_EXTENSIONS.includes(ext)) {
-    return "不支持的音频格式，请上传 MP3、WAV 或 M4A 文件"
-  }
-  if (file.size > MAX_FILE_SIZE) {
-    return "文件大小超过 10MB，请压缩后重试"
-  }
-  return null
-}
-
 /** 格式化录制时间 mm:ss */
 export const formatRecordTime = (seconds: number): string => {
   const m = Math.floor(seconds / 60)

apps/web/src/components/voice/clone-modal/InputView.tsx

@@ -1,142 +0,0 @@
-import React from "react"
-import { Button } from "@/components/ui"
-import UploadZone from "./UploadZone"
-import RecordArea from "./RecordArea"
-import StepIndicator from "./StepIndicator"
-import { MAX_VOICE_NAME_LENGTH, MAX_VOICE_DESC_LENGTH } from "../constants/cloneModal"
-
-interface InputViewProps {
-  voiceName: string
-  voiceDescription: string
-  selectedFile: File | null
-  dragActive: boolean
-  isRecording: boolean
-  recordTime: number
-  recordedBlob: Blob | null
-  errorMessage: string
-  canSubmit: boolean
-  onVoiceNameChange: (value: string) => void
-  onVoiceDescChange: (value: string) => void
-  onDragActiveChange: (active: boolean) => void
-  onFileSelect: (file: File | null, error: string) => void
-  onRecordToggle: () => void
-  onClose: () => void
-  onSubmit: () => void
-}
-
-const INPUT_STEPS = ["上传/录制音频", "填写信息", "提交克隆"]
-
-const InputView: React.FC<InputViewProps> = ({
-  voiceName,
-  voiceDescription,
-  selectedFile,
-  dragActive,
-  isRecording,
-  recordTime,
-  recordedBlob,
-  errorMessage,
-  canSubmit,
-  onVoiceNameChange,
-  onVoiceDescChange,
-  onDragActiveChange,
-  onFileSelect,
-  onRecordToggle,
-  onClose,
-  onSubmit,
-}) => {
-  return (
-    <div className="xx-clonemodal-body">
-      {/* 步骤引导 */}
-      <StepIndicator currentStep={0} steps={INPUT_STEPS} />
-
-      {/* 音色名称 */}
-      <div className="xx-clonemodal-field">
-        <label className="xx-clonemodal-label">
-          音色名称 <span className="xx-clonemodal-required">*</span>
-        </label>
-        <input
-          type="text"
-          className="xx-clonemodal-input"
-          value={voiceName}
-          onChange={(e) => onVoiceNameChange(e.target.value)}
-          placeholder="输入音色名称（2-20字符）"
-          maxLength={MAX_VOICE_NAME_LENGTH}
-        />
-        <div className="xx-clonemodal-char-count">
-          {voiceName.length}/{MAX_VOICE_NAME_LENGTH}
-        </div>
-      </div>
-
-      {/* 上传区域 */}
-      <div className="xx-clonemodal-field">
-        <label className="xx-clonemodal-label">上传音频</label>
-        <UploadZone
-          selectedFile={selectedFile}
-          dragActive={dragActive}
-          onDragActiveChange={onDragActiveChange}
-          onFileSelect={onFileSelect}
-        />
-      </div>
-
-      {/* 或分隔 */}
-      <div className="xx-clonemodal-divider">
-        <div className="xx-clonemodal-divider-line" />
-        <span className="xx-clonemodal-divider-text">或</span>
-        <div className="xx-clonemodal-divider-line" />
-      </div>
-
-      {/* 录制区域 */}
-      <div className="xx-clonemodal-field">
-        <label className="xx-clonemodal-label">直接录制</label>
-        <RecordArea
-          isRecording={isRecording}
-          recordTime={recordTime}
-          recordedBlob={recordedBlob}
-          onRecordToggle={onRecordToggle}
-        />
-      </div>
-
-      {/* 音色描述 */}
-      <div className="xx-clonemodal-field">
-        <label className="xx-clonemodal-label">音色描述</label>
-        <textarea
-          className="xx-clonemodal-textarea"
-          value={voiceDescription}
-          onChange={(e) => onVoiceDescChange(e.target.value)}
-          placeholder="可选，描述这个音色的特点（最多100字符）"
-          maxLength={MAX_VOICE_DESC_LENGTH}
-          rows={3}
-        />
-        <div className="xx-clonemodal-char-count">
-          {voiceDescription.length}/{MAX_VOICE_DESC_LENGTH}
-        </div>
-      </div>
-
-      {/* 错误提示 */}
-      {errorMessage && (
-        <div className="xx-clonemodal-error">
-          <span className="xx-clonemodal-error-icon">⚠️</span>
-          <span>{errorMessage}</span>
-        </div>
-      )}
-
-      {/* 提示 */}
-      <div className="xx-clonemodal-tip">
-        <span className="xx-clonemodal-tip-icon">💡</span>
-        <span>建议上传 10 秒 ~ 3 分钟的清晰语音，环境安静、语速均匀效果最佳</span>
-      </div>
-
-      {/* 底部按钮 */}
-      <div className="xx-clonemodal-footer">
-        <Button buttonType="ghost" onClick={onClose}>
-          取消
-        </Button>
-        <Button buttonType="primary" disabled={!canSubmit} onClick={onSubmit}>
-          🎤 开始克隆
-        </Button>
-      </div>
-    </div>
-  )
-}
-
-export default InputView

apps/web/src/components/voice/clone-modal/ProgressView.tsx

@@ -1,92 +0,0 @@
-import React from "react"
-import { PROGRESS_STEPS } from "../constants/cloneModal"
-import type { ProgressStep } from "../types/cloneModal"
-import type { ModalPhase } from "../types/cloneModal"
-
-interface ProgressViewProps {
-  phase: ModalPhase
-}
-
-const getProgressIndex = (phase: ModalPhase): number => {
-  switch (phase) {
-    case "uploading":
-      return 0
-    case "cloning":
-      return 1
-    case "done":
-      return 2
-    default:
-      return -1
-  }
-}
-
-const ProgressView: React.FC<ProgressViewProps> = ({ phase }) => {
-  const progressIndex = getProgressIndex(phase)
-  const isDone = phase === "done"
-
-  return (
-    <div className="xx-clonemodal-progress-body">
-      {/* 步骤指示器 */}
-      <div className="xx-clonemodal-steps-progress">
-        {PROGRESS_STEPS.map((step: ProgressStep, idx: number) => {
-          const isActive = idx === progressIndex && !isDone
-          const stepDone = idx < progressIndex || isDone
-          const stepClass = [
-            "xx-clonemodal-step-progress",
-            isActive ? "xx-clonemodal-step-progress--active" : "",
-            stepDone ? "xx-clonemodal-step-progress--done" : "",
-          ]
-            .filter(Boolean)
-            .join(" ")
-
-          return (
-            <React.Fragment key={step.key}>
-              {idx > 0 && (
-                <div
-                  className={`xx-clonemodal-step-connector${stepDone ? " xx-clonemodal-step-connector--done" : ""}`}
-                />
-              )}
-              <div className={stepClass}>
-                <div className="xx-clonemodal-step-icon">{stepDone ? "✓" : step.icon}</div>
-                <span className="xx-clonemodal-step-label">{step.label}</span>
-              </div>
-            </React.Fragment>
-          )
-        })}
-      </div>
-
-      {/* 完成阶段 */}
-      {isDone && (
-        <div className="xx-clonemodal-success">
-          <div className="xx-clonemodal-success-icon">🎉</div>
-          <h3 className="xx-clonemodal-success-title">克隆已提交</h3>
-          <p className="xx-clonemodal-success-desc">
-            音色正在生成中，完成后将出现在「我的克隆」列表中
-          </p>
-        </div>
-      )}
-
-      {/* 进行中阶段 */}
-      {!isDone && (
-        <div className="xx-clonemodal-progress-info">
-          {phase === "uploading" && (
-            <>
-              <div className="xx-clonemodal-progress-spinner" />
-              <p className="xx-clonemodal-progress-text">正在上传音频文件…</p>
-              <p className="xx-clonemodal-progress-sub">请稍候，正在将音频上传至服务器</p>
-            </>
-          )}
-          {phase === "cloning" && (
-            <>
-              <div className="xx-clonemodal-progress-spinner xx-clonemodal-progress-spinner--cloning" />
-              <p className="xx-clonemodal-progress-text">AI 正在克隆你的声音…</p>
-              <p className="xx-clonemodal-progress-sub">正在分析声音特征，生成专属音色模型</p>
-            </>
-          )}
-        </div>
-      )}
-    </div>
-  )
-}
-
-export default ProgressView

apps/web/src/components/voice/clone-modal/RecordArea.tsx

@@ -1,49 +0,0 @@
-import React from "react"
-import { formatRecordTime } from "../utils/cloneModal"
-
-interface RecordAreaProps {
-  isRecording: boolean
-  recordTime: number
-  recordedBlob: Blob | null
-  onRecordToggle: () => void
-}
-
-const RecordArea: React.FC<RecordAreaProps> = ({
-  isRecording,
-  recordTime,
-  recordedBlob,
-  onRecordToggle,
-}) => {
-  const getHintText = () => {
-    if (isRecording) return `录制中 ${formatRecordTime(recordTime)}`
-    if (recordedBlob) return `已录制 ${formatRecordTime(recordTime)}`
-    return "点击按钮开始录制（最长 5 分钟）"
-  }
-
-  return (
-    <div className="xx-clonemodal-record-area">
-      <div className="xx-clonemodal-record-info">
-        <p className="xx-clonemodal-record-hint">{getHintText()}</p>
-        {isRecording && (
-          <div className="xx-clonemodal-record-wave">
-            <span className="xx-clonemodal-record-wave-bar" />
-            <span className="xx-clonemodal-record-wave-bar" />
-            <span className="xx-clonemodal-record-wave-bar" />
-            <span className="xx-clonemodal-record-wave-bar" />
-            <span className="xx-clonemodal-record-wave-bar" />
-          </div>
-        )}
-      </div>
-      <button
-        type="button"
-        className={`xx-clonemodal-record-btn${isRecording ? " xx-clonemodal-record-btn--recording" : ""}`}
-        onClick={onRecordToggle}
-        title={isRecording ? "停止录制" : "开始录制"}
-      >
-        {isRecording ? "⏹" : "🎙️"}
-      </button>
-    </div>
-  )
-}
-
-export default RecordArea

apps/web/src/components/voice/clone-modal/StepIndicator.tsx

@@ -1,30 +0,0 @@
-import React from "react"
-
-interface StepIndicatorProps {
-  currentStep: number
-  steps: string[]
-}
-
-/**
- * 输入阶段顶部的步骤引导（数字步骤）
- */
-const StepIndicator: React.FC<StepIndicatorProps> = ({ currentStep, steps }) => {
-  return (
-    <div className="xx-clonemodal-steps">
-      {steps.map((label, idx) => {
-        const isActive = idx <= currentStep
-        return (
-          <React.Fragment key={idx}>
-            {idx > 0 && <div className="xx-clonemodal-step-connector" />}
-            <div className={`xx-clonemodal-step${isActive ? " xx-clonemodal-step--active" : ""}`}>
-              <div className="xx-clonemodal-step-number">{idx + 1}</div>
-              <span className="xx-clonemodal-step-label">{label}</span>
-            </div>
-          </React.Fragment>
-        )
-      })}
-    </div>
-  )
-}
-
-export default StepIndicator

apps/web/src/components/voice/clone-modal/UploadZone.tsx

@@ -1,79 +0,0 @@
-import React, { useRef } from "react"
-import { ACCEPTED_MIME } from "../constants/cloneModal"
-import { validateFile } from "../utils/cloneModal"
-
-interface UploadZoneProps {
-  selectedFile: File | null
-  dragActive: boolean
-  onDragActiveChange: (active: boolean) => void
-  onFileSelect: (file: File | null, error: string) => void
-}
-
-const UploadZone: React.FC<UploadZoneProps> = ({
-  selectedFile,
-  dragActive,
-  onDragActiveChange,
-  onFileSelect,
-}) => {
-  const fileInputRef = useRef<HTMLInputElement>(null)
-
-  const handleUploadClick = () => {
-    fileInputRef.current?.click()
-  }
-
-  const handleFileChange = (e: React.ChangeEvent<HTMLInputElement>) => {
-    const file = e.target.files?.[0]
-    if (file) {
-      const error = validateFile(file)
-      onFileSelect(error ? null : file, error || "")
-    }
-    e.target.value = ""
-  }
-
-  const handleDrag = (e: React.DragEvent) => {
-    e.preventDefault()
-    e.stopPropagation()
-    if (e.type === "dragenter" || e.type === "dragover") {
-      onDragActiveChange(true)
-    } else if (e.type === "dragleave") {
-      onDragActiveChange(false)
-    }
-  }
-
-  const handleDrop = (e: React.DragEvent) => {
-    e.preventDefault()
-    e.stopPropagation()
-    onDragActiveChange(false)
-    const file = e.dataTransfer.files?.[0]
-    if (file) {
-      const error = validateFile(file)
-      onFileSelect(error ? null : file, error || "")
-    }
-  }
-
-  return (
-    <div
-      className={`xx-clonemodal-upload-zone${dragActive ? " xx-clonemodal-upload-zone--active" : ""}${selectedFile ? " xx-clonemodal-upload-zone--has-file" : ""}`}
-      onClick={handleUploadClick}
-      onDragEnter={handleDrag}
-      onDragOver={handleDrag}
-      onDragLeave={handleDrag}
-      onDrop={handleDrop}
-    >
-      <div className="xx-clonemodal-upload-icon">{selectedFile ? "📄" : "🎵"}</div>
-      <p className="xx-clonemodal-upload-title">
-        {selectedFile ? selectedFile.name : "拖拽音频文件到此处，或点击上传"}
-      </p>
-      <p className="xx-clonemodal-upload-hint">支持 MP3、WAV、M4A 格式，最大 10MB</p>
-      <input
-        ref={fileInputRef}
-        type="file"
-        accept={ACCEPTED_MIME}
-        style={{ display: "none" }}
-        onChange={handleFileChange}
-      />
-    </div>
-  )
-}
-
-export default UploadZone

apps/web/src/components/voice/constants/cloneModal.ts

@@ -1,29 +0,0 @@
-import type { ProgressStep } from "../types/cloneModal"
-
-/** 进度阶段配置 */
-export const PROGRESS_STEPS: ProgressStep[] = [
-  { key: "uploading", label: "上传中", icon: "📤" },
-  { key: "cloning", label: "克隆中", icon: "🧬" },
-  { key: "done", label: "完成", icon: "✅" },
-]
-
-/** 支持的音频扩展名 */
-export const ACCEPTED_EXTENSIONS = ["mp3", "wav", "m4a"]
-
-/** input accept 属性值 */
-export const ACCEPTED_MIME = ".mp3,.wav,.m4a,audio/mpeg,audio/wav,audio/mp4"
-
-/** 最大文件大小：10MB */
-export const MAX_FILE_SIZE = 10 * 1024 * 1024
-
-/** 最长录制时长（秒）：5 分钟 */
-export const MAX_RECORD_SECONDS = 5 * 60
-
-/** 音色名称最小长度 */
-export const MIN_VOICE_NAME_LENGTH = 2
-
-/** 音色名称最大长度 */
-export const MAX_VOICE_NAME_LENGTH = 20
-
-/** 音色描述最大长度 */
-export const MAX_VOICE_DESC_LENGTH = 100

apps/web/src/components/voice/hooks/useAudioRecorder.ts

@@ -1,119 +0,0 @@
-import { useState, useRef, useCallback, useEffect } from "react"
-import { MAX_RECORD_SECONDS } from "../constants/cloneModal"
-
-interface UseAudioRecorderReturn {
-  isRecording: boolean
-  recordTime: number
-  recordedBlob: Blob | null
-  toggleRecording: () => void
-  resetRecording: () => void
-}
-
-/**
- * 录音 Hook —— 封装 MediaRecorder 录音逻辑
- */
-const useAudioRecorder = (): UseAudioRecorderReturn => {
-  const [isRecording, setIsRecording] = useState(false)
-  const [recordTime, setRecordTime] = useState(0)
-  const [recordedBlob, setRecordedBlob] = useState<Blob | null>(null)
-
-  const recordTimerRef = useRef<ReturnType<typeof setInterval> | null>(null)
-  const mediaRecorderRef = useRef<MediaRecorder | null>(null)
-  const audioChunksRef = useRef<Blob[]>([])
-
-  const stopRecording = useCallback(() => {
-    setIsRecording(false)
-    if (recordTimerRef.current) {
-      clearInterval(recordTimerRef.current)
-      recordTimerRef.current = null
-    }
-    if (mediaRecorderRef.current && mediaRecorderRef.current.state !== "inactive") {
-      mediaRecorderRef.current.stop()
-    }
-  }, [])
-
-  const startRecording = useCallback(async () => {
-    try {
-      const stream = await navigator.mediaDevices.getUserMedia({ audio: true })
-      const mediaRecorder = new MediaRecorder(stream)
-      mediaRecorderRef.current = mediaRecorder
-      audioChunksRef.current = []
-
-      mediaRecorder.ondataavailable = (event) => {
-        if (event.data.size > 0) {
-          audioChunksRef.current.push(event.data)
-        }
-      }
-
-      mediaRecorder.onstop = () => {
-        const blob = new Blob(audioChunksRef.current, { type: "audio/webm" })
-        setRecordedBlob(blob)
-        stream.getTracks().forEach((track) => track.stop())
-      }
-
-      mediaRecorder.start()
-      setIsRecording(true)
-      setRecordTime(0)
-      setRecordedBlob(null)
-
-      recordTimerRef.current = setInterval(() => {
-        setRecordTime((prev) => {
-          const next = prev + 1
-          if (next >= MAX_RECORD_SECONDS) {
-            setTimeout(() => {
-              stopRecording()
-            }, 0)
-            return MAX_RECORD_SECONDS
-          }
-          return next
-        })
-      }, 1000)
-    } catch {
-      // 错误由调用方通过其他机制提示
-      setIsRecording(false)
-    }
-  }, [stopRecording])
-
-  const toggleRecording = useCallback(() => {
-    if (isRecording) {
-      stopRecording()
-    } else {
-      startRecording()
-    }
-  }, [isRecording, startRecording, stopRecording])
-
-  const resetRecording = useCallback(() => {
-    setIsRecording(false)
-    setRecordTime(0)
-    setRecordedBlob(null)
-    audioChunksRef.current = []
-    if (recordTimerRef.current) {
-      clearInterval(recordTimerRef.current)
-      recordTimerRef.current = null
-    }
-    if (mediaRecorderRef.current && mediaRecorderRef.current.state !== "inactive") {
-      mediaRecorderRef.current.stop()
-    }
-    mediaRecorderRef.current = null
-  }, [])
-
-  // 卸载时清理
-  useEffect(() => {
-    return () => {
-      if (recordTimerRef.current) clearInterval(recordTimerRef.current)
-      if (mediaRecorderRef.current && mediaRecorderRef.current.state !== "inactive") {
-        mediaRecorderRef.current.stop()
-      }
-    }
-  }, [])
-
-  return {
-    isRecording,
-    recordTime,
-    recordedBlob,
-    toggleRecording,
-    resetRecording,
-  }
-}
-
-export default useAudioRecorder

apps/web/src/components/voice/hooks/useCloneFormState.ts

@@ -1,134 +0,0 @@
-import { useState, useRef, useCallback, useEffect } from "react"
-import type { ModalPhase } from "../types/cloneModal"
-import { MIN_VOICE_NAME_LENGTH, MAX_VOICE_NAME_LENGTH } from "../constants/cloneModal"
-import useAudioRecorder from "./useAudioRecorder"
-
-/**
- * 克隆弹窗表单状态 Hook
- * 管理表单字段、录音、文件选择、验证逻辑
- */
-export function useCloneFormState({ open, onClose }: { open: boolean; onClose: () => void }) {
-  const [phase, setPhase] = useState<ModalPhase>("input")
-  const [voiceName, setVoiceName] = useState("")
-  const [voiceDescription, setVoiceDescription] = useState("")
-  const [selectedFile, setSelectedFile] = useState<File | null>(null)
-  const [dragActive, setDragActive] = useState(false)
-  const [errorMessage, setErrorMessage] = useState("")
-
-  const { isRecording, recordTime, recordedBlob, toggleRecording, resetRecording } =
-    useAudioRecorder()
-
-  /** 默认音色名称计数器 */
-  const cloneCounterRef = useRef(1)
-
-  const getNextDefaultName = useCallback((): string => {
-    const name = `我的声音 ${cloneCounterRef.current}`
-    cloneCounterRef.current += 1
-    return name
-  }, [])
-
-  const hasAudio = selectedFile !== null || recordedBlob !== null
-
-  const canSubmit =
-    voiceName.trim().length >= MIN_VOICE_NAME_LENGTH &&
-    voiceName.trim().length <= MAX_VOICE_NAME_LENGTH &&
-    hasAudio
-
-  const isProcessing = phase === "uploading" || phase === "cloning"
-
-  /** 重置弹窗状态 */
-  const resetState = useCallback(() => {
-    setPhase("input")
-    setVoiceName(getNextDefaultName())
-    setVoiceDescription("")
-    setSelectedFile(null)
-    setDragActive(false)
-    setErrorMessage("")
-    resetRecording()
-  }, [getNextDefaultName, resetRecording])
-
-  /** 关闭弹窗 */
-  const handleClose = useCallback(() => {
-    resetState()
-    onClose()
-  }, [resetState, onClose])
-
-  /** 弹窗打开时重置状态 */
-  useEffect(() => {
-    if (open) {
-      resetState()
-    }
-  }, [open, resetState])
-
-  /** 选择文件（来自上传或拖拽） */
-  const handleFileSelect = useCallback(
-    (file: File | null, error: string) => {
-      if (error) {
-        setErrorMessage(error)
-        setSelectedFile(null)
-      } else {
-        setErrorMessage("")
-        setSelectedFile(file)
-        // 清除录音
-        resetRecording()
-      }
-    },
-    [resetRecording],
-  )
-
-  /** 录音切换 */
-  const handleRecordToggle = useCallback(() => {
-    setErrorMessage("")
-    if (isRecording) {
-      toggleRecording()
-    } else {
-      // 开始录制前清除已选文件
-      setSelectedFile(null)
-      toggleRecording()
-    }
-  }, [isRecording, toggleRecording])
-
-  /** 表单验证 */
-  const validateForm = useCallback((): string | null => {
-    const name = voiceName.trim()
-    if (!name) {
-      return "请输入音色名称"
-    }
-    if (name.length < MIN_VOICE_NAME_LENGTH || name.length > MAX_VOICE_NAME_LENGTH) {
-      return `音色名称需在 ${MIN_VOICE_NAME_LENGTH}-${MAX_VOICE_NAME_LENGTH} 个字符之间`
-    }
-    if (!hasAudio) {
-      return "请上传音频文件或录制一段声音"
-    }
-    return null
-  }, [voiceName, hasAudio])
-
-  return {
-    // 状态
-    phase,
-    setPhase,
-    voiceName,
-    setVoiceName,
-    voiceDescription,
-    setVoiceDescription,
-    selectedFile,
-    dragActive,
-    setDragActive,
-    errorMessage,
-    setErrorMessage,
-    // 录音
-    isRecording,
-    recordTime,
-    recordedBlob,
-    // 计算属性
-    hasAudio,
-    canSubmit,
-    isProcessing,
-    // handlers
-    handleFileSelect,
-    handleRecordToggle,
-    handleClose,
-    validateForm,
-    resetState,
-  }
-}

apps/web/src/components/voice/hooks/useCloneModal.ts

@@ -1,46 +0,0 @@
-import type { CloneModalProps } from "../types/cloneModal"
-import { useCloneFormState } from "./useCloneFormState"
-import { useCloneSubmit } from "./useCloneSubmit"
-
-/**
- * 音色克隆弹窗主业务 Hook
- * 组合表单状态 + 提交流程两个子 Hook
- */
-const useCloneModal = ({ open, onClose, onSuccess }: CloneModalProps) => {
-  const formState = useCloneFormState({ open, onClose })
-
-  const { handleSubmit } = useCloneSubmit({
-    voiceName: formState.voiceName,
-    voiceDescription: formState.voiceDescription,
-    selectedFile: formState.selectedFile,
-    recordedBlob: formState.recordedBlob,
-    setPhase: formState.setPhase,
-    setErrorMessage: formState.setErrorMessage,
-    validateForm: formState.validateForm,
-    onSuccess,
-    onClose: formState.handleClose,
-  })
-
-  return {
-    phase: formState.phase,
-    voiceName: formState.voiceName,
-    voiceDescription: formState.voiceDescription,
-    selectedFile: formState.selectedFile,
-    dragActive: formState.dragActive,
-    errorMessage: formState.errorMessage,
-    isRecording: formState.isRecording,
-    recordTime: formState.recordTime,
-    recordedBlob: formState.recordedBlob,
-    canSubmit: formState.canSubmit,
-    isProcessing: formState.isProcessing,
-    setVoiceName: formState.setVoiceName,
-    setVoiceDescription: formState.setVoiceDescription,
-    setDragActive: formState.setDragActive,
-    handleFileSelect: formState.handleFileSelect,
-    handleRecordToggle: formState.handleRecordToggle,
-    handleClose: formState.handleClose,
-    handleSubmit,
-  }
-}
-
-export default useCloneModal

apps/web/src/components/voice/hooks/useCloneSubmit.ts

@@ -1,108 +0,0 @@
-import { useRef, useCallback, useEffect } from "react"
-import { createVoiceClone, toVoiceClone } from "@/api/voice-clone"
-import { uploadAssetDirect, ensureDefaultLibrary } from "@/api/assets"
-import { getOrCreateDefaultProject } from "@/api/projects"
-import type { VoiceClone } from "@/api/voice-clone"
-
-interface UseCloneSubmitOptions {
-  voiceName: string
-  voiceDescription: string
-  selectedFile: File | null
-  recordedBlob: Blob | null
-  setPhase: (phase: "input" | "uploading" | "cloning" | "done") => void
-  setErrorMessage: (msg: string) => void
-  validateForm: () => string | null
-  onSuccess?: (clone: VoiceClone) => void
-  onClose: () => void
-}
-
-/**
- * 克隆提交流程 Hook
- * 封装上传 + 克隆 + 完成的三阶段流程
- */
-export function useCloneSubmit({
-  voiceName,
-  voiceDescription,
-  selectedFile,
-  recordedBlob,
-  setPhase,
-  setErrorMessage,
-  validateForm,
-  onSuccess,
-  onClose,
-}: UseCloneSubmitOptions) {
-  const timerRef = useRef<ReturnType<typeof setTimeout> | null>(null)
-
-  /** 组件卸载时清理定时器 */
-  useEffect(() => {
-    return () => {
-      if (timerRef.current) clearTimeout(timerRef.current)
-    }
-  }, [])
-
-  const handleSubmit = useCallback(async () => {
-    const formError = validateForm()
-    if (formError) {
-      setErrorMessage(formError)
-      return
-    }
-
-    setErrorMessage("")
-
-    try {
-      // 阶段 1：上传音频
-      setPhase("uploading")
-
-      let fileToUpload: File
-      if (selectedFile) {
-        fileToUpload = selectedFile
-      } else {
-        fileToUpload = new File([recordedBlob!], `recorded-${Date.now()}.webm`, {
-          type: "audio/webm",
-        })
-      }
-
-      // 获取默认项目和素材库
-      const project = await getOrCreateDefaultProject()
-      const library = await ensureDefaultLibrary({ project_id: project.id, kind: "voice" })
-
-      // 直传到 OSS
-      const uploadResult = await uploadAssetDirect({
-        file: fileToUpload,
-        library_id: library.id,
-      })
-
-      // 阶段 2：克隆
-      setPhase("cloning")
-      const result = await createVoiceClone({
-        name: voiceName.trim(),
-        description: voiceDescription.trim() || undefined,
-        audio_url: uploadResult.url,
-      })
-
-      // 阶段 3：完成
-      setPhase("done")
-
-      // 2秒后自动关闭
-      timerRef.current = setTimeout(() => {
-        onSuccess?.(toVoiceClone(result))
-        onClose()
-      }, 2000)
-    } catch (err) {
-      setPhase("input")
-      setErrorMessage(err instanceof Error ? err.message : "克隆失败，请重试")
-    }
-  }, [
-    validateForm,
-    selectedFile,
-    recordedBlob,
-    voiceName,
-    voiceDescription,
-    setPhase,
-    setErrorMessage,
-    onSuccess,
-    onClose,
-  ])
-
-  return { handleSubmit }
-}

apps/web/src/components/voice/types/cloneModal.ts

@@ -1,20 +0,0 @@
-import type { VoiceClone } from "@/api/voice-clone"
-
-/** 弹窗阶段 */
-export type ModalPhase = "input" | "uploading" | "cloning" | "done"
-
-export interface CloneModalProps {
-  /** 弹窗是否可见 */
-  open: boolean
-  /** 关闭弹窗回调 */
-  onClose: () => void
-  /** 克隆成功回调（返回新创建的音色） */
-  onSuccess?: (voice: VoiceClone) => void
-}
-
-/** 进度步骤项 */
-export interface ProgressStep {
-  key: string
-  label: string
-  icon: string
-}

apps/web/src/components/voice/utils/cloneModal.ts

@@ -1,25 +0,0 @@
-import { ACCEPTED_EXTENSIONS, MAX_FILE_SIZE } from "../constants/cloneModal"
-
-/**
- * 格式化录制时间 mm:ss
- */
-export const formatRecordTime = (seconds: number): string => {
-  const m = Math.floor(seconds / 60)
-  const s = seconds % 60
-  return `${m.toString().padStart(2, "0")}:${s.toString().padStart(2, "0")}`
-}
-
-/**
- * 验证上传的音频文件
- * @returns 错误信息，null 表示验证通过
- */
-export const validateFile = (file: File): string | null => {
-  const ext = file.name.split(".").pop()?.toLowerCase()
-  if (!ext || !ACCEPTED_EXTENSIONS.includes(ext)) {
-    return "不支持的音频格式，请上传 MP3、WAV 或 M4A 文件"
-  }
-  if (file.size > MAX_FILE_SIZE) {
-    return "文件大小超过 10MB，请压缩后重试"
-  }
-  return null
-}

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

@@ -20,6 +20,7 @@ import LibrarySidebar from "@/pages/assets/components/LibrarySidebar"
 import AssetFilterBar from "@/pages/assets/components/AssetFilterBar"
 import BatchOperationBar from "@/pages/assets/components/BatchOperationBar"
 import AssetUploadZone from "@/pages/assets/components/AssetUploadZone"
+import UploadQueuePanel from "@/pages/assets/components/UploadQueuePanel"
 import AssetGridSection from "@/pages/assets/components/AssetGridSection"
 import AssetModals from "@/pages/assets/components/AssetModals"
 import { useAssetsData } from "@/pages/assets/hooks/useAssetsData"
@@ -69,7 +70,30 @@ const AssetLibrary: React.FC = () => {
   })
 
   /* ── 上传 ── */
-  const { uploading, uploadProgress, handleUpload } = useAssetUpload({ effectiveLibId })
+  const {
+    uploadItems,
+    enqueueUploads,
+    retryUpload,
+    removeUpload,
+    clearFinished,
+    uploading,
+    activeCount,
+    pendingCount,
+  } = useAssetUpload({ effectiveLibId })
+
+  /* ── 上传中 asset_id → 进度/状态映射，合并进网格卡片展示真实进度 ── */
+  const uploadProgressMap = React.useMemo(() => {
+    const map = new Map<string, { progress: number; uploading: boolean }>()
+    for (const it of uploadItems) {
+      if (it.assetId && (it.status === "uploading" || it.status === "ingesting")) {
+        map.set(it.assetId, {
+          progress: it.status === "ingesting" ? 100 : it.progress,
+          uploading: it.status === "uploading",
+        })
+      }
+    }
+    return map
+  }, [uploadItems])
 
   /* ── 选中态管理 ── */
   const { selectedIds, setSelectedIds, toggleSelect, selectAll, deselectAll } = useAssetSelection({
@@ -144,8 +168,17 @@ const AssetLibrary: React.FC = () => {
           {/* 上传区域 */}
           <AssetUploadZone
             uploading={uploading}
-            uploadProgress={uploadProgress}
-            onUpload={handleUpload}
+            activeCount={activeCount}
+            pendingCount={pendingCount}
+            onUpload={enqueueUploads}
+          />
+
+          {/* 上传队列：独立进度 + 失败重试/移除 */}
+          <UploadQueuePanel
+            items={uploadItems}
+            onRetry={retryUpload}
+            onRemove={removeUpload}
+            onClearFinished={clearFinished}
           />
 
           {/* 筛选栏 */}
@@ -180,6 +213,7 @@ const AssetLibrary: React.FC = () => {
             assets={filteredAssets}
             selectedIds={selectedIds}
             diagnosingId={diagnosingId}
+            uploadProgressMap={uploadProgressMap}
             onRetry={refetchAssets}
             onToggleSelect={toggleSelect}
             onDiagnose={handleDiagnose}
@@ -191,8 +225,6 @@ const AssetLibrary: React.FC = () => {
 
       {/* ─── 弹窗集合 ─── */}
       <AssetModals
-        uploading={uploading}
-        uploadProgress={uploadProgress}
         createModalOpen={createModalOpen}
         onCreateModalCancel={() => setCreateModalOpen(false)}
         onCreateModalOk={handleCreateLibrary}

apps/web/src/pages/assets/assets.css

@@ -147,44 +147,52 @@
 /* ============================================================
    上传区域
    ============================================================ */
-.xx-asset-upload-zone {
-  border: 2px dashed var(--border-color);
-  border-radius: var(--radius-lg);
-  padding: var(--space-2xl) var(--space-xl);
-  text-align: center;
-  background: var(--bg-secondary);
-  cursor: pointer;
+.xx-asset-upload-entry {
+  display: flex;
+  align-items: center;
+  gap: var(--space-md);
+  flex-wrap: wrap;
+  padding: var(--space-sm) var(--space-md);
+  border: 1px dashed transparent;
+  border-radius: var(--radius-md);
   transition: var(--transition-all);
 }
 
-.xx-asset-upload-zone:hover {
+.xx-asset-upload-entry-dragover {
   border-color: var(--primary-color);
   background: var(--primary-soft);
 }
 
-.xx-asset-upload-zone:active {
-  border-style: solid;
-  transform: scale(0.99);
-  box-shadow: inset 0 2px 4px rgba(0, 0, 0, 0.06);
-}
-
-.xx-asset-upload-icon {
-  font-size: 40px;
-  margin-bottom: var(--space-sm);
-  color: var(--primary-color);
-}
-
-.xx-asset-upload-text {
-  font-size: var(--font-size-base) !important;
-  color: var(--text-primary) !important;
-  margin: 0 0 var(--space-xs) !important;
+.xx-asset-upload-btn {
+  display: inline-flex;
+  align-items: center;
+  gap: var(--space-xs);
+  padding: 6px 16px;
+  font-size: var(--font-size-sm);
   font-weight: var(--font-weight-medium);
+  color: var(--text-inverse);
+  background: var(--primary-color);
+  border: none;
+  border-radius: var(--radius-sm);
+  cursor: pointer;
+  transition: var(--transition-all);
+  white-space: nowrap;
 }
 
-.xx-asset-upload-hint {
-  font-size: var(--font-size-sm) !important;
-  color: var(--text-tertiary) !important;
-  margin: 0 !important;
+.xx-asset-upload-btn:hover {
+  opacity: 0.9;
+}
+
+.xx-asset-upload-btn:active {
+  transform: scale(0.98);
+}
+
+.xx-asset-upload-status {
+  display: inline-flex;
+  align-items: center;
+  gap: var(--space-xs);
+  font-size: var(--font-size-xs);
+  color: var(--text-tertiary);
 }
 
 /* ============================================================
@@ -215,7 +223,7 @@
    ============================================================ */
 .xx-asset-grid {
   display: grid;
-  grid-template-columns: repeat(4, 1fr);
+  grid-template-columns: repeat(6, 1fr);
   gap: 14px;
 }
 
@@ -234,7 +242,7 @@
 .xx-asset-card:hover {
   border-color: var(--primary-color);
   box-shadow: var(--shadow-sm);
-  transform: translateY(-2px);
+  transform: translateY(-1px);
 }
 
 .xx-asset-card:active {
@@ -244,7 +252,7 @@
 
 /* 缩略图 */
 .xx-asset-thumb {
-  aspect-ratio: 9 / 16;
+  aspect-ratio: 3 / 4;
   position: relative;
   overflow: hidden;
   display: grid;
@@ -260,22 +268,22 @@
 }
 
 .xx-asset-thumb-placeholder {
-  font-size: var(--font-size-3xl);
+  font-size: var(--font-size-xl);
   opacity: 0.7;
 }
 
 /* 播放按钮 */
 .xx-asset-play {
   position: absolute;
-  width: 40px;
-  height: 40px;
+  width: 32px;
+  height: 32px;
   border-radius: var(--radius-full);
   background: rgba(255, 255, 255, 0.25);
   backdrop-filter: blur(4px);
   display: grid;
   place-items: center;
   color: var(--text-inverse);
-  font-size: var(--font-size-md);
+  font-size: var(--font-size-sm);
   transition: var(--transition-all);
 }
 
@@ -332,8 +340,8 @@
   position: absolute;
   bottom: var(--space-sm, 8px);
   right: var(--space-sm, 8px);
-  width: 28px;
-  height: 28px;
+  width: 24px;
+  height: 24px;
   border-radius: var(--radius-full, 999px);
   background: rgba(0, 0, 0, 0.45);
   backdrop-filter: blur(4px);
@@ -380,12 +388,12 @@
 
 /* 卡片信息 */
 .xx-asset-info {
-  padding: 12px;
+  padding: 8px;
 }
 
 .xx-asset-name {
-  margin: 0 0 6px;
-  font-size: var(--font-size-sm);
+  margin: 0 0 4px;
+  font-size: var(--font-size-xs);
   font-weight: var(--font-weight-semibold);
   color: var(--text-primary);
   white-space: nowrap;
@@ -397,9 +405,67 @@
   display: flex;
   justify-content: space-between;
   align-items: center;
+  gap: var(--space-xs);
   font-size: var(--font-size-xs);
   color: var(--text-secondary);
-  margin-bottom: var(--space-sm);
+  margin-bottom: 4px;
+  min-width: 0;
+}
+
+/* 状态标签行：标签过长省略 */
+.xx-asset-meta-status {
+  display: inline-flex;
+  align-items: center;
+  min-width: 0;
+}
+
+.xx-asset-meta-status .xx-status-pill {
+  max-width: 100%;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.xx-asset-meta-duration {
+  flex-shrink: 0;
+  color: var(--text-tertiary);
+  font-variant-numeric: tabular-nums;
+}
+
+/* 余量标签独占一行 */
+.xx-asset-meta-usage {
+  justify-content: flex-start;
+  margin-bottom: var(--space-xs);
+}
+
+/* 视频素材余量角标（仅状态展示，不影响卡片操作） */
+.xx-asset-usage-badge {
+  display: inline-flex;
+  align-items: center;
+  padding: 1px 6px;
+  border-radius: var(--radius-full);
+  font-size: 10px;
+  font-weight: var(--font-weight-medium);
+  line-height: 1.5;
+  white-space: nowrap;
+}
+
+/* 已用尽：红色实心 */
+.xx-asset-usage-badge-exhausted {
+  background: var(--error-color);
+  color: var(--text-inverse);
+}
+
+/* 即将用尽：红色软底 */
+.xx-asset-usage-badge-warning {
+  background: var(--error-soft);
+  color: var(--error-color);
+}
+
+/* 已用 xx%：橙色软底 */
+.xx-asset-usage-badge-ratio {
+  background: var(--warning-soft);
+  color: var(--warning-color);
 }
 
 /* 诊断按钮 */
@@ -443,12 +509,16 @@
 .xx-status-pill {
   display: inline-flex;
   align-items: center;
-  gap: var(--space-xs);
-  padding: var(--space-xxs) var(--space-sm);
+  gap: 2px;
+  padding: 1px 6px;
   border-radius: var(--radius-full);
-  font-size: var(--font-size-xs);
+  font-size: 10px;
   font-weight: var(--font-weight-medium);
   line-height: 1.5;
+  white-space: nowrap;
+  max-width: 100%;
+  overflow: hidden;
+  text-overflow: ellipsis;
 }
 
 .xx-status-pill-ok {
@@ -592,7 +662,7 @@
    ============================================================ */
 .xx-assets-skeleton-grid {
   display: grid;
-  grid-template-columns: repeat(4, 1fr);
+  grid-template-columns: repeat(6, 1fr);
   gap: 14px;
 }
 
@@ -607,6 +677,10 @@
   .xx-asset-grid {
     grid-template-columns: repeat(3, 1fr);
   }
+
+  .xx-assets-skeleton-grid {
+    grid-template-columns: repeat(3, 1fr);
+  }
 }
 
 @media (max-width: 1200px) {
@@ -630,6 +704,10 @@
   .xx-asset-grid {
     grid-template-columns: repeat(3, 1fr);
   }
+
+  .xx-assets-skeleton-grid {
+    grid-template-columns: repeat(3, 1fr);
+  }
 }
 
 @media (max-width: 768px) {
@@ -641,6 +719,10 @@
     grid-template-columns: repeat(2, 1fr);
   }
 
+  .xx-assets-skeleton-grid {
+    grid-template-columns: repeat(2, 1fr);
+  }
+
   .xx-assets-filters {
     flex-direction: column;
     align-items: stretch;
@@ -660,43 +742,132 @@
   .xx-asset-grid {
     grid-template-columns: 1fr;
   }
+
+  .xx-assets-skeleton-grid {
+    grid-template-columns: 1fr;
+  }
 }
 
-/* ─── 上传进度弹窗 ─── */
-.xx-upload-progress-modal .ant-modal-content {
-  padding: 24px 16px 20px;
-  border-radius: 16px;
+/* ─── 上传队列面板 ─── */
+.xx-upload-queue {
+  margin-top: 12px;
+  border: 1px solid var(--border-primary, #e5e7eb);
+  border-radius: 12px;
+  background: #fff;
+  overflow: hidden;
 }
 
-.xx-upload-progress-body {
+.xx-upload-queue-header {
   display: flex;
-  flex-direction: column;
+  justify-content: space-between;
   align-items: center;
-  gap: 16px;
-  padding: 8px 0;
+  padding: 10px 14px;
+  border-bottom: 1px solid var(--border-primary, #eef2f7);
+  background: #f8fafc;
 }
 
-.xx-upload-progress-ring {
-  display: block;
-}
-
-.xx-upload-progress-text {
-  display: flex;
-  flex-direction: column;
-  align-items: center;
-  gap: 4px;
-}
-
-.xx-upload-progress-pct {
-  font-size: 22px;
-  font-weight: 700;
-  color: var(--primary-color, #6366f1);
-  line-height: 1;
-}
-
-.xx-upload-progress-label {
+.xx-upload-queue-title {
   font-size: 13px;
-  color: var(--text-secondary, #6b7280);
+  font-weight: 600;
+  color: var(--text-primary, #1e293b);
+}
+
+.xx-upload-queue-list {
+  max-height: 240px;
+  overflow-y: auto;
+}
+
+.xx-upload-queue-item {
+  display: flex;
+  align-items: flex-start;
+  gap: 10px;
+  padding: 10px 14px;
+}
+
+.xx-upload-queue-item + .xx-upload-queue-item {
+  border-top: 1px solid var(--border-primary, #f1f5f9);
+}
+
+.xx-upload-queue-icon {
+  padding-top: 2px;
+  font-size: 15px;
+}
+
+.xx-upload-queue-body {
+  flex: 1;
+  min-width: 0;
+}
+
+.xx-upload-queue-name {
+  font-size: 13px;
+  color: var(--text-primary, #1e293b);
+  white-space: nowrap;
+  overflow: hidden;
+  text-overflow: ellipsis;
+}
+
+.xx-upload-queue-progress {
+  margin-top: 6px;
+  height: 4px;
+  border-radius: 2px;
+  background: var(--border-primary, #e5e7eb);
+  overflow: hidden;
+}
+
+.xx-upload-queue-progress-bar {
+  height: 100%;
+  border-radius: 2px;
+  background: var(--primary-color, #6366f1);
+  transition: width 0.25s ease;
+}
+
+.xx-upload-queue-status {
+  margin-top: 4px;
+  font-size: 12px;
+  color: var(--text-tertiary, #94a3b8);
+}
+
+.xx-upload-queue-error .xx-upload-queue-status {
+  color: #ef4444;
+}
+
+.xx-upload-queue-actions {
+  display: flex;
+  gap: 6px;
+}
+
+.xx-upload-queue-btn {
+  border: none;
+  background: none;
+  cursor: pointer;
+  font-size: 13px;
+  color: var(--text-tertiary, #94a3b8);
+  padding: 2px 4px;
+}
+
+.xx-upload-queue-btn:hover {
+  color: var(--primary-color, #6366f1);
+}
+
+/* ─── 素材卡片上传中遮罩进度条 ─── */
+.xx-asset-thumb-uploading {
+  flex-direction: column;
+  gap: 8px;
+}
+
+.xx-asset-upload-bar {
+  width: 70%;
+  height: 4px;
+  border-radius: 2px;
+  background: rgba(255, 255, 255, 0.35);
+  overflow: hidden;
+}
+
+.xx-asset-upload-bar-inner {
+  height: 100%;
+  border-radius: 2px;
+  background: #fff;
+  transition: width 0.25s ease;
 }
 
 /* ─── 批量打标签弹窗 ─── */

apps/web/src/pages/assets/components/AssetCard.tsx

@@ -8,7 +8,7 @@ import {
   CloseCircleOutlined,
 } from "@ant-design/icons"
 import { Popconfirm } from "antd"
-import type { AssetItem } from "@/pages/assets/types"
+import { getUsageBadge, type AssetItem } from "@/pages/assets/types"
 import { thumbGradient } from "@/pages/assets/utils/asset"
 import { kindIcon } from "@/pages/assets/utils/kindIcon"
 import { StatusPill } from "./AssetSkeleton"
@@ -20,6 +20,8 @@ export interface AssetCardProps {
   asset: AssetItem
   selected: boolean
   diagnosing?: boolean
+  /** 上传中实时进度（仅 uploading 态有值；ingesting 后由后端状态接管） */
+  uploadProgress?: { progress: number; uploading: boolean }
   onToggle: () => void
   onDiagnose: () => void
   onPlay: () => void
@@ -30,99 +32,128 @@ const AssetCard: React.FC<AssetCardProps> = ({
   asset,
   selected,
   diagnosing,
+  uploadProgress,
   onToggle,
   onDiagnose,
   onPlay,
   onDelete,
-}) => (
-  <div className={`xx-asset-card${selected ? " xx-asset-card-selected" : ""}`} onClick={onToggle}>
-    {/* 缩略图区 */}
-    <div className="xx-asset-thumb" style={{ background: thumbGradient(asset.kind) }}>
-      {asset.thumbUrl ? (
-        <img src={asset.thumbUrl} alt={asset.name} />
-      ) : (
-        <span className="xx-asset-thumb-placeholder">
-          {asset.loading ? <LoadingOutlined /> : kindIcon(asset.kind)}
-        </span>
-      )}
+}) => {
+  const isUploading = !!uploadProgress?.uploading
+  // 视频素材余量角标（已用尽/即将用尽/已用 xx%）；非视频或字段缺失返回 null
+  const usageBadge = getUsageBadge(asset)
+  return (
+    <div className={`xx-asset-card${selected ? " xx-asset-card-selected" : ""}`} onClick={onToggle}>
+      {/* 缩略图区 */}
+      <div className="xx-asset-thumb" style={{ background: thumbGradient(asset.kind) }}>
+        {asset.thumbUrl ? (
+          <img src={asset.thumbUrl} alt={asset.name} />
+        ) : (
+          <span className="xx-asset-thumb-placeholder">
+            {asset.loading ? <LoadingOutlined /> : kindIcon(asset.kind)}
+          </span>
+        )}
 
-      {/* 处理中遮罩 */}
-      {asset.loading && (
-        <div className="xx-asset-thumb-overlay xx-asset-thumb-processing">
-          <LoadingOutlined />
-          <span>处理中</span>
+        {/* 上传中遮罩：真实进度百分比 + 进度条 */}
+        {isUploading && (
+          <div className="xx-asset-thumb-overlay xx-asset-thumb-uploading">
+            <LoadingOutlined />
+            <span>上传中 {uploadProgress?.progress ?? 0}%</span>
+            <div className="xx-asset-upload-bar">
+              <div
+                className="xx-asset-upload-bar-inner"
+                style={{ width: `${uploadProgress?.progress ?? 0}%` }}
+              />
+            </div>
+          </div>
+        )}
+
+        {/* 转码/处理中遮罩 */}
+        {asset.loading && !isUploading && (
+          <div className="xx-asset-thumb-overlay xx-asset-thumb-processing">
+            <LoadingOutlined />
+            <span>转码处理中</span>
+          </div>
+        )}
+
+        {/* 失败状态标识 */}
+        {asset.status === "bad" && asset.statusLabel === "处理失败" && (
+          <div className="xx-asset-thumb-overlay xx-asset-thumb-failed">
+            <CloseCircleOutlined />
+            <span>处理失败</span>
+          </div>
+        )}
+
+        {/* 视频/配音类显示播放按钮（处理中/失败不显示） */}
+        {asset.kind === "video" && !asset.loading && asset.status !== "bad" && (
+          <span
+            className="xx-asset-play"
+            onClick={(e) => {
+              e.stopPropagation()
+              onPlay()
+            }}
+          >
+            <PlayCircleOutlined />
+          </span>
+        )}
+
+        {/* 删除按钮 */}
+        <Popconfirm
+          title="确认删除"
+          description="删除后不可恢复，确定要删除这个素材吗？"
+          onConfirm={(e) => {
+            e?.stopPropagation()
+            onDelete()
+          }}
+          onCancel={(e) => e?.stopPropagation()}
+          okText="删除"
+          cancelText="取消"
+          okButtonProps={{ danger: true }}
+        >
+          <span className="xx-asset-delete" onClick={(e) => e.stopPropagation()}>
+            <DeleteOutlined />
+          </span>
+        </Popconfirm>
+
+        {/* 选中态勾选 */}
+        {selected && (
+          <span className="xx-asset-check">
+            <CheckOutlined />
+          </span>
+        )}
+      </div>
+
+      {/* 信息区 */}
+      <div className="xx-asset-info">
+        <p className="xx-asset-name" title={asset.name}>
+          {asset.name}
+        </p>
+        <div className="xx-asset-meta">
+          <span className="xx-asset-meta-status">
+            <StatusPill status={asset.status} label={asset.statusLabel} />
+          </span>
+          {asset.duration && <span className="xx-asset-meta-duration">{asset.duration}</span>}
         </div>
-      )}
-
-      {/* 失败状态标识 */}
-      {asset.status === "bad" && asset.statusLabel === "处理失败" && (
-        <div className="xx-asset-thumb-overlay xx-asset-thumb-failed">
-          <CloseCircleOutlined />
-          <span>处理失败</span>
-        </div>
-      )}
-
-      {/* 视频/配音类显示播放按钮（处理中/失败不显示） */}
-      {asset.kind === "video" && !asset.loading && asset.status !== "bad" && (
-        <span
-          className="xx-asset-play"
+        {usageBadge && (
+          <div className="xx-asset-meta xx-asset-meta-usage">
+            <span className={`xx-asset-usage-badge xx-asset-usage-badge-${usageBadge.variant}`}>
+              {usageBadge.label}
+            </span>
+          </div>
+        )}
+        <button
+          className={`xx-asset-diagnose-btn${diagnosing ? " xx-asset-diagnose-btn-loading" : ""}`}
+          disabled={diagnosing || asset.loading || asset.status === "bad"}
           onClick={(e) => {
             e.stopPropagation()
-            onPlay()
+            onDiagnose()
           }}
         >
-          <PlayCircleOutlined />
-        </span>
-      )}
-
-      {/* 删除按钮 */}
-      <Popconfirm
-        title="确认删除"
-        description="删除后不可恢复，确定要删除这个素材吗？"
-        onConfirm={(e) => {
-          e?.stopPropagation()
-          onDelete()
-        }}
-        onCancel={(e) => e?.stopPropagation()}
-        okText="删除"
-        cancelText="取消"
-        okButtonProps={{ danger: true }}
-      >
-        <span className="xx-asset-delete" onClick={(e) => e.stopPropagation()}>
-          <DeleteOutlined />
-        </span>
-      </Popconfirm>
-
-      {/* 选中态勾选 */}
-      {selected && (
-        <span className="xx-asset-check">
-          <CheckOutlined />
-        </span>
-      )}
-    </div>
-
-    {/* 信息区 */}
-    <div className="xx-asset-info">
-      <p className="xx-asset-name" title={asset.name}>
-        {asset.name}
-      </p>
-      <div className="xx-asset-meta">
-        <StatusPill status={asset.status} label={asset.statusLabel} />
-        {asset.duration && <span>{asset.duration}</span>}
+          {diagnosing ? <LoadingOutlined /> : <ExperimentOutlined />}
+          {diagnosing ? "诊断中..." : "诊断"}
+        </button>
       </div>
-      <button
-        className={`xx-asset-diagnose-btn${diagnosing ? " xx-asset-diagnose-btn-loading" : ""}`}
-        disabled={diagnosing || asset.loading || asset.status === "bad"}
-        onClick={(e) => {
-          e.stopPropagation()
-          onDiagnose()
-        }}
-      >
-        {diagnosing ? <LoadingOutlined /> : <ExperimentOutlined />}
-        {diagnosing ? "诊断中..." : "诊断"}
-      </button>
     </div>
-  </div>
-)
+  )
+}
 
 export default AssetCard

apps/web/src/pages/assets/components/AssetGridSection.tsx

@@ -8,6 +8,9 @@ import type { AssetItem } from "../types"
 import AssetCard from "./AssetCard"
 import { SkeletonCard } from "./AssetSkeleton"
 
+/** 上传中素材的实时进度（asset_id → 进度信息），由上传队列合并到卡片 */
+export type UploadProgressMap = Map<string, { progress: number; uploading: boolean }>
+
 export interface AssetGridSectionProps {
   loading: boolean
   error: boolean
@@ -15,6 +18,7 @@ export interface AssetGridSectionProps {
   assets: AssetItem[]
   selectedIds: Set<string>
   diagnosingId: string | null
+  uploadProgressMap?: UploadProgressMap
   onRetry?: () => void
   onToggleSelect: (id: string) => void
   onDiagnose: (asset: AssetItem) => void
@@ -29,6 +33,7 @@ export const AssetGridSection: React.FC<AssetGridSectionProps> = ({
   assets,
   selectedIds,
   diagnosingId,
+  uploadProgressMap,
   onRetry,
   onToggleSelect,
   onDiagnose,
@@ -70,6 +75,7 @@ export const AssetGridSection: React.FC<AssetGridSectionProps> = ({
             asset={asset}
             selected={selectedIds.has(asset.id)}
             diagnosing={diagnosingId === asset.id}
+            uploadProgress={uploadProgressMap?.get(asset.id)}
             onToggle={() => onToggleSelect(asset.id)}
             onDiagnose={() => onDiagnose(asset)}
             onPlay={() => onPlay(asset)}

apps/web/src/pages/assets/components/AssetModals.tsx

@@ -11,12 +11,9 @@ import BatchTagModal from "./BatchTagModal"
 import BatchClassifyModal from "./BatchClassifyModal"
 import BatchMarkModal from "./BatchMarkModal"
 import ResultDrawer from "./ResultDrawer"
-import UploadProgressModal from "./UploadProgressModal"
 
 export interface AssetModalsProps {
   /* 上传进度 */
-  uploading: boolean
-  uploadProgress: number
 
   /* 新建视频库 */
   createModalOpen: boolean
@@ -68,8 +65,6 @@ export interface AssetModalsProps {
 }
 
 export const AssetModals: React.FC<AssetModalsProps> = ({
-  uploading,
-  uploadProgress,
   createModalOpen,
   onCreateModalCancel,
   onCreateModalOk,
@@ -109,9 +104,6 @@ export const AssetModals: React.FC<AssetModalsProps> = ({
 }) => {
   return (
     <>
-      {/* 上传进度弹窗 */}
-      <UploadProgressModal open={uploading} progress={uploadProgress} />
-
       {/* 新建视频库弹窗 */}
       <CreateLibraryModal
         open={createModalOpen}

apps/web/src/pages/assets/components/AssetUploadZone.tsx

@@ -1,37 +1,91 @@
 /**
- * AssetLibrary 上传拖拽区域
+ * AssetLibrary 上传入口（紧凑按钮模式）
+ * - 点击按钮打开文件选择（多选），多文件入队由 useAssetUpload 队列控制（最多 3 路直传）
+ * - 拖拽文件到内容区任意位置同样触发上传（不再占用大面积虚线框）
  */
-import React from "react"
-import { Upload } from "antd"
-import { InboxOutlined } from "@ant-design/icons"
+import React, { useRef, useState } from "react"
+import { PlusOutlined, CloudUploadOutlined } from "@ant-design/icons"
 
 export interface AssetUploadZoneProps {
   uploading: boolean
-  uploadProgress: number
-  onUpload: (file: File) => void
+  activeCount: number
+  pendingCount: number
+  onUpload: (files: File[]) => void
 }
 
-export const AssetUploadZone: React.FC<AssetUploadZoneProps> = ({ uploading, onUpload }) => {
+export const AssetUploadZone: React.FC<AssetUploadZoneProps> = ({
+  uploading,
+  activeCount,
+  pendingCount,
+  onUpload,
+}) => {
+  const inputRef = useRef<HTMLInputElement>(null)
+  // dragenter/dragleave 在经过子元素时会成对触发，用计数器避免高亮闪烁；
+  // 计数器归零（拖拽真正离开容器）才取消高亮
+  const dragDepthRef = useRef(0)
+  const [dragOver, setDragOver] = useState(false)
+
+  const pickFiles = (list: FileList | null) => {
+    if (!list || list.length === 0) return
+    onUpload(Array.from(list))
+  }
+
   return (
-    <Upload.Dragger
-      beforeUpload={(file) => {
-        onUpload(file as File)
-        return false
+    <div
+      className={`xx-asset-upload-entry${dragOver ? " xx-asset-upload-entry-dragover" : ""}`}
+      onDragEnter={(e) => {
+        e.preventDefault()
+        dragDepthRef.current += 1
+        setDragOver(true)
+      }}
+      onDragOver={(e) => {
+        e.preventDefault()
+      }}
+      onDragLeave={(e) => {
+        e.preventDefault()
+        dragDepthRef.current = Math.max(0, dragDepthRef.current - 1)
+        if (dragDepthRef.current === 0) {
+          setDragOver(false)
+        }
+      }}
+      onDrop={(e) => {
+        e.preventDefault()
+        dragDepthRef.current = 0
+        setDragOver(false)
+        pickFiles(e.dataTransfer.files)
       }}
-      showUploadList={false}
-      multiple
-      accept="video/*,image/*"
     >
-      <div className="xx-asset-upload-zone">
-        <p className="xx-asset-upload-icon">
-          <InboxOutlined />
-        </p>
-        <p className="xx-asset-upload-text">
-          {uploading ? "上传中..." : "点击或拖拽文件到此区域上传"}
-        </p>
-        <p className="xx-asset-upload-hint">支持视频、图片，单文件不超过 2GB</p>
-      </div>
-    </Upload.Dragger>
+      <button
+        type="button"
+        className="xx-asset-upload-btn"
+        onClick={() => inputRef.current?.click()}
+      >
+        <PlusOutlined />
+        上传素材
+      </button>
+      <span className="xx-asset-upload-status">
+        {uploading ? (
+          <>
+            <CloudUploadOutlined />
+            上传中…（进行 {activeCount} 个{pendingCount > 0 ? `，排队 ${pendingCount} 个` : ""}）
+          </>
+        ) : (
+          "视频、图片均可，单文件不超过 2GB；也可直接拖拽文件到此区域"
+        )}
+      </span>
+      <input
+        ref={inputRef}
+        type="file"
+        multiple
+        accept="video/*,image/*"
+        style={{ display: "none" }}
+        onChange={(e) => {
+          pickFiles(e.target.files)
+          // 允许连续选择同一文件
+          e.target.value = ""
+        }}
+      />
+    </div>
   )
 }
 

apps/web/src/pages/assets/components/UploadProgressModal.tsx

@@ -1,56 +0,0 @@
-import React from "react"
-import { Modal as AntModal } from "antd"
-
-/* ============================================================
- * UploadProgressModal — 上传进度弹窗（圆形动画 + 百分比）
- * ============================================================ */
-export interface UploadProgressModalProps {
-  open: boolean
-  progress: number
-}
-
-const UploadProgressModal: React.FC<UploadProgressModalProps> = ({ open, progress }) => (
-  <AntModal
-    open={open}
-    footer={null}
-    closable={false}
-    centered
-    width={260}
-    maskClosable={false}
-    className="xx-upload-progress-modal"
-  >
-    <div className="xx-upload-progress-body">
-      <svg className="xx-upload-progress-ring" viewBox="0 0 120 120" width={120} height={120}>
-        {/* 背景圆环 */}
-        <circle
-          cx="60"
-          cy="60"
-          r="52"
-          fill="none"
-          stroke="var(--border-primary, #e5e7eb)"
-          strokeWidth="8"
-        />
-        {/* 进度圆弧 */}
-        <circle
-          cx="60"
-          cy="60"
-          r="52"
-          fill="none"
-          stroke="var(--primary-color, #6366f1)"
-          strokeWidth="8"
-          strokeLinecap="round"
-          strokeDasharray={`${2 * Math.PI * 52}`}
-          strokeDashoffset={`${2 * Math.PI * 52 * (1 - progress / 100)}`}
-          transform="rotate(-90 60 60)"
-          style={{ transition: "stroke-dashoffset 0.3s ease" }}
-        />
-      </svg>
-      <div className="xx-upload-progress-text">
-        <span className="xx-upload-progress-pct">{progress}%</span>
-        <span className="xx-upload-progress-label">上传中…</span>
-      </div>
-    </div>
-  </AntModal>
-)
-
-export default UploadProgressModal

apps/web/src/pages/assets/components/UploadQueuePanel.tsx

@@ -0,0 +1,117 @@
+/**
+ * 上传队列面板
+ * 展示批量上传中每个文件的独立状态/进度；失败可重试、可移除、可清空已完成。
+ * 上传中的素材卡片同时也会出现在素材网格（后端 prepare 预建 asset），
+ * 此面板用于展示真实传输进度与失败重试入口。
+ */
+import React from "react"
+import {
+  LoadingOutlined,
+  CheckCircleFilled,
+  CloseCircleFilled,
+  ReloadOutlined,
+  CloseOutlined,
+} from "@ant-design/icons"
+import type { UploadItem } from "../hooks/useAssetUpload"
+
+export interface UploadQueuePanelProps {
+  items: UploadItem[]
+  onRetry: (tempId: string) => void
+  onRemove: (tempId: string) => void
+  onClearFinished: () => void
+}
+
+const STATUS_TEXT: Record<UploadItem["status"], string> = {
+  preparing: "排队中…",
+  uploading: "上传中",
+  ingesting: "转码中…",
+  done: "已完成",
+  error: "上传失败",
+}
+
+const UploadQueuePanel: React.FC<UploadQueuePanelProps> = ({
+  items,
+  onRetry,
+  onRemove,
+  onClearFinished,
+}) => {
+  if (items.length === 0) return null
+  const finishedCount = items.filter((it) => it.status === "done").length
+
+  return (
+    <div className="xx-upload-queue">
+      <div className="xx-upload-queue-header">
+        <span className="xx-upload-queue-title">
+          上传任务（{items.length}
+          {finishedCount > 0 ? `，已完成 ${finishedCount}` : ""}）
+        </span>
+        {finishedCount > 0 && (
+          <button type="button" className="xx-link-btn" onClick={onClearFinished}>
+            清空已完成
+          </button>
+        )}
+      </div>
+      <div className="xx-upload-queue-list">
+        {items.map((it) => {
+          const isActive = it.status === "preparing" || it.status === "uploading"
+          const showProgress = it.status === "uploading" || it.status === "ingesting"
+          return (
+            <div key={it.tempId} className={`xx-upload-queue-item xx-upload-queue-${it.status}`}>
+              <span className="xx-upload-queue-icon">
+                {it.status === "done" || it.duplicated ? (
+                  <CheckCircleFilled style={{ color: "#22c55e" }} />
+                ) : it.status === "error" ? (
+                  <CloseCircleFilled style={{ color: "#ef4444" }} />
+                ) : (
+                  <LoadingOutlined style={{ color: "var(--primary-color)" }} />
+                )}
+              </span>
+              <div className="xx-upload-queue-body">
+                <div className="xx-upload-queue-name" title={it.fileName}>
+                  {it.fileName}
+                </div>
+                {showProgress ? (
+                  <div className="xx-upload-queue-progress">
+                    <div
+                      className="xx-upload-queue-progress-bar"
+                      style={{ width: `${it.status === "ingesting" ? 100 : it.progress}%` }}
+                    />
+                  </div>
+                ) : null}
+                <div className="xx-upload-queue-status">
+                  {it.duplicated ? "素材已存在，已跳过" : STATUS_TEXT[it.status]}
+                  {it.status === "uploading" ? ` ${it.progress}%` : ""}
+                  {it.status === "error" && it.error ? `：${it.error}` : ""}
+                </div>
+              </div>
+              <span className="xx-upload-queue-actions">
+                {it.status === "error" && (
+                  <button
+                    type="button"
+                    className="xx-upload-queue-btn"
+                    title="重试"
+                    onClick={() => onRetry(it.tempId)}
+                  >
+                    <ReloadOutlined />
+                  </button>
+                )}
+                {(it.status === "error" || it.status === "done") && !isActive && (
+                  <button
+                    type="button"
+                    className="xx-upload-queue-btn"
+                    title="移除"
+                    onClick={() => onRemove(it.tempId)}
+                  >
+                    <CloseOutlined />
+                  </button>
+                )}
+              </span>
+            </div>
+          )
+        })}
+      </div>
+    </div>
+  )
+}
+
+export default UploadQueuePanel

apps/web/src/pages/assets/hooks/useAssetUpload.ts

@@ -1,65 +1,196 @@
-import { useState, useCallback } from "react"
+import { useState, useCallback, useRef, useEffect } from "react"
 import { useQueryClient } from "@tanstack/react-query"
 import { message } from "antd"
-import { uploadAssetDirect } from "@/api/assets"
-import { MAX_FILE_SIZE, LARGE_FILE_THRESHOLD } from "../constants"
+import { prepareDirectUploadHandle, type DirectUploadHandle } from "@/api/assets"
+import { MAX_FILE_SIZE } from "../constants"
 
-/**
- * 素材上传 Hook
- * 封装上传状态、进度管理和上传逻辑
- */
-interface UseAssetUploadProps {
-  effectiveLibId: string
+/** 单文件上传状态机 */
+export type UploadItemStatus = "preparing" | "uploading" | "ingesting" | "done" | "error"
+
+export interface UploadItem {
+  /** 前端临时 id（prepare 前无 asset_id 时用） */
+  tempId: string
+  file: File
+  fileName: string
+  /** 进度 0~100（仅直传阶段有真实进度） */
+  progress: number
+  status: UploadItemStatus
+  /** 后端 prepare 预建的 asset id（旧后端可能为空） */
+  assetId?: string
+  /** 去重命中：complete 返回 duplicated，标记完成但不产生新素材 */
+  duplicated?: boolean
+  error?: string
 }
 
-export function useAssetUpload({ effectiveLibId }: UseAssetUploadProps) {
+/** 批量直传最大并发数，避免多文件瓜分上行带宽 */
+const MAX_CONCURRENT = 3
+
+/**
+ * 素材批量上传 Hook
+ * - prepare 阶段后端预建 status=uploading 的 asset，前端拿到 asset_id 立即刷新列表
+ * - OSS 直传并发限制为 3，其余排队；每个文件独立进度/状态
+ * - complete 后素材进入转码（ingesting/processing），由列表轮询反映
+ * - 失败卡片支持重试/移除
+ */
+export function useAssetUpload({ effectiveLibId }: { effectiveLibId: string }) {
   const queryClient = useQueryClient()
 
-  const [uploading, setUploading] = useState(false)
-  const [uploadProgress, setUploadProgress] = useState(0)
+  const [items, setItems] = useState<UploadItem[]>([])
+  const itemsRef = useRef<UploadItem[]>([])
+  itemsRef.current = items
 
-  const handleUpload = useCallback(
-    async (file: File) => {
-      if (file.size > MAX_FILE_SIZE) {
-        message.error(`文件 "${file.name}" 超过 2GB 限制`)
-        return
+  const updateItem = useCallback((tempId: string, patch: Partial<UploadItem>) => {
+    setItems((prev) => prev.map((it) => (it.tempId === tempId ? { ...it, ...patch } : it)))
+  }, [])
+
+  /** 刷新素材列表（prepare 后/complete 后调用，让卡片即时出现/流转） */
+  const refreshList = useCallback(() => {
+    // 使用 refetchQueries 强制立即重新获取，避免 staleTime 导致延迟
+    if (effectiveLibId) {
+      queryClient.refetchQueries({ queryKey: ["assets", effectiveLibId] })
+    }
+    queryClient.invalidateQueries({ queryKey: ["asset-libraries"] })
+  }, [queryClient, effectiveLibId])
+
+  /** 执行单个文件的完整上传流程（prepare→transfer→complete） */
+  const runUpload = useCallback(
+    async (item: UploadItem, handle?: DirectUploadHandle) => {
+      try {
+        // 1. prepare（重试时复用已准备的 handle 也行，但签名可能过期，重新 prepare 最稳）
+        const h =
+          handle ??
+          (await prepareDirectUploadHandle({ file: item.file, library_id: effectiveLibId }))
+        if (h.prepared.asset_id) {
+          updateItem(item.tempId, {
+            status: "uploading",
+            assetId: h.prepared.asset_id,
+            progress: 0,
+          })
+          // 预建 asset 已入库，立即刷新让「上传中」卡片出现在网格
+          refreshList()
+        } else {
+          updateItem(item.tempId, { status: "uploading", progress: 0 })
+        }
+
+        // 2. OSS 直传（真实进度）
+        await h.transfer((pct) => updateItem(item.tempId, { progress: pct }))
+
+        // 3. complete：后端创建 ingest job，素材进入转码
+        updateItem(item.tempId, { status: "ingesting", progress: 100 })
+        const result = await h.complete()
+        refreshList()
+
+        if (result.duplicated) {
+          updateItem(item.tempId, { status: "done", duplicated: true, assetId: result.asset_id })
+          message.info(`"${item.fileName}" 与素材库已有内容相同，已跳过`)
+        } else {
+          updateItem(item.tempId, { status: "done" })
+          message.success(`"${item.fileName}" 上传完成，正在转码处理`)
+        }
+      } catch (err: unknown) {
+        const detail = err instanceof Error ? err.message : "上传失败"
+        console.error("[useAssetUpload] 上传失败:", item.fileName, err)
+        updateItem(item.tempId, { status: "error", error: detail })
+        message.error(`"${item.fileName}" 上传失败：${detail}`)
       }
+    },
+    [effectiveLibId, refreshList, updateItem],
+  )
+
+  /**
+   * 队列调度：把并发槽塞满（同时在途的 prepare+transfer 不超过 MAX_CONCURRENT）。
+   * runUpload 在 await prepare 期间 state 仍是 preparing，多个并发 pump 若只看 state
+   * 会重复认领同一项，因此用 claimedRef 记录已被认领的 tempId。
+   */
+  const inFlightRef = useRef(0)
+  const claimedRef = useRef<Set<string>>(new Set())
+  const pumpRef = useRef<() => void>(() => {})
+  pumpRef.current = () => {
+    while (inFlightRef.current < MAX_CONCURRENT) {
+      const next = itemsRef.current.find(
+        (it) => it.status === "preparing" && !claimedRef.current.has(it.tempId),
+      )
+      if (!next) return
+      claimedRef.current.add(next.tempId)
+      inFlightRef.current += 1
+      void runUpload(next).finally(() => {
+        inFlightRef.current -= 1
+        claimedRef.current.delete(next.tempId)
+        // 一个任务结束（成功/失败）后继续拉起排队任务
+        setTimeout(() => pumpRef.current(), 0)
+      })
+    }
+  }
+
+  useEffect(() => {
+    pumpRef.current()
+  }, [items])
+
+  /** 入队一个或多个文件 */
+  const enqueueUploads = useCallback(
+    (files: File[]) => {
       if (!effectiveLibId) {
         message.warning("请先选择或创建一个视频库")
         return
       }
-
-      setUploading(true)
-      setUploadProgress(0)
-      try {
-        if (file.size > LARGE_FILE_THRESHOLD) {
-          message.info(`大文件 "${file.name}" 将使用直传上传`)
+      const valid: File[] = []
+      for (const file of files) {
+        if (file.size > MAX_FILE_SIZE) {
+          message.error(`文件 "${file.name}" 超过 2GB 限制`)
+          continue
         }
-        await uploadAssetDirect({
-          file,
-          library_id: effectiveLibId,
-          onProgress: (pct) => setUploadProgress(pct),
-        })
-        message.success(`"${file.name}" 上传成功`)
-        queryClient.invalidateQueries({ queryKey: ["assets"] })
-        queryClient.invalidateQueries({ queryKey: ["asset-libraries"] })
-      } catch (err: unknown) {
-        const detail = err instanceof Error ? err.message : ""
-        console.error("[handleUpload] 上传失败:", err)
-        message.error(`"${file.name}" 上传失败${detail ? `：${detail}` : ""}`)
-        // 错误时延迟关闭弹窗，让用户能看到错误提示
-        await new Promise((r) => setTimeout(r, 1500))
-      } finally {
-        setUploading(false)
-        setUploadProgress(0)
+        valid.push(file)
       }
+      if (valid.length === 0) return
+
+      const newItems: UploadItem[] = valid.map((file, idx) => ({
+        tempId: `${Date.now()}-${idx}-${Math.random().toString(36).slice(2, 8)}`,
+        file,
+        fileName: file.name,
+        progress: 0,
+        status: "preparing",
+      }))
+      setItems((prev) => [...prev, ...newItems])
     },
-    [effectiveLibId, queryClient],
+    [effectiveLibId],
   )
 
+  /** 重试失败任务 */
+  const retryUpload = useCallback(
+    (tempId: string) => {
+      const target = itemsRef.current.find((it) => it.tempId === tempId)
+      if (!target) return
+      updateItem(tempId, { status: "preparing", progress: 0, error: undefined })
+      // 状态更新后由 useEffect 触发 pump
+    },
+    [updateItem],
+  )
+
+  /** 从上传列表移除（已进入转码的由素材网格管理；这里只移除上传面板记录） */
+  const removeUpload = useCallback((tempId: string) => {
+    setItems((prev) => prev.filter((it) => it.tempId !== tempId))
+  }, [])
+
+  /** 清空已完成/去重记录 */
+  const clearFinished = useCallback(() => {
+    setItems((prev) => prev.filter((it) => it.status !== "done"))
+  }, [])
+
+  const activeCount = items.filter(
+    (it) => it.status === "preparing" || it.status === "uploading",
+  ).length
+  const pendingCount = items.filter((it) => it.status === "preparing").length
+  const hasActive = activeCount > 0 || items.some((it) => it.status === "ingesting")
+
   return {
-    uploading,
-    uploadProgress,
-    handleUpload,
+    uploadItems: items,
+    enqueueUploads,
+    retryUpload,
+    removeUpload,
+    clearFinished,
+    /** 是否有进行中的上传（用于上传区文案） */
+    uploading: hasActive,
+    activeCount,
+    pendingCount,
   }
 }

apps/web/src/pages/assets/hooks/useAssetsData.ts

@@ -3,9 +3,11 @@ import { useQuery } from "@tanstack/react-query"
 import {
   getAssetLibraries,
   getAssets,
+  ensureDefaultLibrary,
   type AssetLibraryItem,
   type AssetItem as ApiAssetItem,
 } from "@/api/assets"
+import { getOrCreateDefaultProject } from "@/api/projects"
 import { mapLibrary, mapAsset, type AssetItem, type LibraryItem } from "../types"
 
 /**
@@ -16,7 +18,18 @@ export function useAssetsData() {
   /* ── 视频库列表查询 ── */
   const { data: apiLibraries = [], isLoading: libLoading } = useQuery<AssetLibraryItem[], Error>({
     queryKey: ["asset-libraries"],
-    queryFn: getAssetLibraries,
+    queryFn: async () => {
+      const libs = await getAssetLibraries()
+      // 如果没有 video 类型的库，自动创建默认视频素材库（与 useVoiceMaterials 保持一致）
+      const hasVideoLib = libs.some((lib) => lib.kind === "video")
+      if (!hasVideoLib) {
+        const project = await getOrCreateDefaultProject()
+        await ensureDefaultLibrary({ project_id: project.id, kind: "video" })
+        // 创建后重新拉取最新列表
+        return getAssetLibraries()
+      }
+      return libs
+    },
     staleTime: 60_000,
   })
 
@@ -45,11 +58,21 @@ export function useAssetsData() {
     queryKey: ["assets", effectiveLibId],
     queryFn: () =>
       getAssets(effectiveLibId, {
-        // 拉取所有非删除状态的素材，让用户上传后立刻能看到"处理中"的素材
+        // 拉取所有非删除状态的素材，让用户上传后立刻能看到"上传中/处理中"的素材
         status: "ready,uploading,ingesting,processing,pending,error,failed",
       }),
     enabled: !!effectiveLibId,
     staleTime: 30_000,
+    // 列表中存在上传中/转码中素材时每 3s 轮询；全部就绪后自动停止
+    refetchInterval: (query) => {
+      const data = query.state.data as { items: ApiAssetItem[] } | undefined
+      const items = data?.items ?? []
+      const processing = items.some((a) => {
+        const st = a.status ?? ""
+        return st === "uploading" || st === "ingesting" || st === "processing" || st === "pending"
+      })
+      return processing ? 3000 : false
+    },
   })
 
   const assets: AssetItem[] = useMemo(

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

@@ -27,6 +27,36 @@ export interface AssetItem {
   duration?: string
   size: number
   createdAt: string
+  /** 已切片段占用时长占比（0~1），后端字段缺失时为 undefined */
+  usedRatio?: number
+  /** 是否已彻底用尽（false 的素材不参与生成选片），字段缺失时视为可用 */
+  usable?: boolean
+}
+
+/** 素材余量角标状态（仅视频素材） */
+export interface UsageBadge {
+  /** 角标文案 */
+  label: string
+  /** 样式变体：exhausted=红色实心，warning=红色软底，ratio=橙色软底 */
+  variant: "exhausted" | "warning" | "ratio"
+}
+
+/**
+ * 根据后端余量字段计算视频素材的余量角标；
+ * 非视频、字段缺失或已用占比 <50% 时不显示（返回 null）。
+ */
+export const getUsageBadge = (asset: {
+  kind?: AssetKind
+  usable?: boolean
+  usedRatio?: number
+}): UsageBadge | null => {
+  if (asset.kind && asset.kind !== "video") return null
+  if (asset.usable === false) return { label: "已用尽", variant: "exhausted" }
+  const ratio = asset.usedRatio
+  if (ratio == null) return null
+  if (ratio >= 0.85) return { label: "即将用尽", variant: "warning" }
+  if (ratio >= 0.5) return { label: `已用 ${Math.round(ratio * 100)}%`, variant: "ratio" }
+  return null
 }
 
 /** 根据 mime_type 推断前端 AssetKind */
@@ -111,5 +141,7 @@ export const mapAsset = (item: ApiAssetItem): AssetItem => {
     duration: metadata.duration != null ? formatDuration(metadata.duration as number) : undefined,
     size: item.file_size ? +(item.file_size / (1024 * 1024)).toFixed(1) : 0,
     createdAt: item.created_at ? new Date(item.created_at).toISOString().slice(0, 10) : "—",
+    usedRatio: item.used_ratio ?? undefined,
+    usable: item.usable ?? undefined,
   }
 }

apps/web/src/pages/editing-planner/EditingPlanner.tsx

@@ -15,8 +15,6 @@ import React, { useState } from "react"
 import { useSearchParams } from "react-router-dom"
 import { MODE_LABELS } from "@/api/editing-planner"
 import { MODE_LIST } from "./constants"
-import type { MediaAsset } from "@/api/template-editor"
-
 import MediaPanel from "./components/MediaPanel"
 import PreviewPlayer from "./components/PreviewPlayer"
 import TimelinePanel from "./components/TimelinePanel"
@@ -79,14 +77,6 @@ const EditingPlanner: React.FC = () => {
   /* ── 右侧栏 Tab ── */
   const [rightTab, setRightTab] = useState<"properties" | "clips">("properties")
 
-  /* ── 素材库 ── */
-  const [mediaAssets, setMediaAssets] = useState<MediaAsset[]>([])
-  const [selectedAssetIds, setSelectedAssetIds] = useState<string[]>([])
-
-  const handleAssetSelect = (ids: string[]) => {
-    setSelectedAssetIds(ids)
-  }
-
   /* ── 配音素材 ── */
   const {
     voiceMaterials,
@@ -113,7 +103,6 @@ const EditingPlanner: React.FC = () => {
     resetClips,
     setClips,
     setSelectedClipId: clipOps.setSelectedClipId,
-    setMediaAssets,
     setTitleConfig,
     setSubtitleSettings,
     setBgmSettings,
@@ -164,9 +153,6 @@ const EditingPlanner: React.FC = () => {
           onLoadTemplate={tpl.handleLoadTemplate}
           onSearchChange={tpl.setSearchQuery}
           onFilterChange={tpl.setCurrentFilter}
-          mediaAssets={mediaAssets}
-          onAssetSelect={handleAssetSelect}
-          selectedAssetIds={selectedAssetIds}
         />
 
         {/* 中栏 flex-1 */}

apps/web/src/pages/editing-planner/components/MediaPanel.tsx

@@ -1,12 +1,10 @@
 /**
- * 左侧面板 — V8 原型 1:1 还原
- * Tab 切换：模板列表 + 素材库
+ * 左侧面板 — 模板列表
+ * 模板编辑器只负责定义模板规则（片段数量、时长范围），不承载素材管理。
  */
-import React, { useState } from "react"
+import React from "react"
 import type { EditingTemplate } from "@/api/editing-planner"
 import { MODE_LABELS } from "@/api/editing-planner"
-import type { MediaAsset } from "@/api/template-editor"
-import AssetSelector from "@/components/asset-selector/AssetSelector"
 
 interface MediaPanelProps {
   templates: EditingTemplate[]
@@ -18,10 +16,6 @@ interface MediaPanelProps {
   onLoadTemplate: (id: string) => void
   onSearchChange: (q: string) => void
   onFilterChange: (f: string) => void
-  // 素材相关
-  mediaAssets?: MediaAsset[]
-  onAssetSelect?: (ids: string[]) => void
-  selectedAssetIds?: string[]
 }
 
 const MediaPanel: React.FC<MediaPanelProps> = ({
@@ -34,113 +28,73 @@ const MediaPanel: React.FC<MediaPanelProps> = ({
   onLoadTemplate,
   onSearchChange,
   onFilterChange,
-  mediaAssets = [],
-  onAssetSelect,
-  selectedAssetIds = [],
 }) => {
-  const [activeTab, setActiveTab] = useState<"templates" | "assets">("templates")
-
   return (
     <div className="ep-left-panel">
-      {/* Tab 切换 */}
-      <div className="ep-left-tabs">
-        <button
-          className={`ep-left-tab ${activeTab === "templates" ? "active" : ""}`}
-          onClick={() => setActiveTab("templates")}
-        >
-          📋 模板
-        </button>
-        <button
-          className={`ep-left-tab ${activeTab === "assets" ? "active" : ""}`}
-          onClick={() => setActiveTab("assets")}
-        >
-          📁 素材
-        </button>
+      {/* 搜索 */}
+      <div className="ep-search-wrap ep-media-panel-inner">
+        <span className="ep-search-icon">🔍</span>
+        <input
+          className="ep-search-input"
+          placeholder="搜索模板..."
+          value={searchQuery}
+          onChange={(e) => onSearchChange(e.target.value)}
+        />
       </div>
 
-      {/* 模板 Tab */}
-      {activeTab === "templates" && (
-        <>
-          {/* 搜索 */}
-          <div className="ep-search-wrap ep-media-panel-inner">
-            <span className="ep-search-icon">🔍</span>
-            <input
-              className="ep-search-input"
-              placeholder="搜索模板..."
-              value={searchQuery}
-              onChange={(e) => onSearchChange(e.target.value)}
-            />
-          </div>
+      {/* Chip 分类筛选 */}
+      <div className="ep-filter-chips">
+        {filterCategories.map((cat) => (
+          <button
+            key={cat}
+            className={`ep-filter-chip ${currentFilter === cat ? "active" : ""}`}
+            onClick={() => onFilterChange(cat)}
+          >
+            {cat}
+          </button>
+        ))}
+      </div>
 
-          {/* Chip 分类筛选 */}
-          <div className="ep-filter-chips">
-            {filterCategories.map((cat) => (
-              <button
-                key={cat}
-                className={`ep-filter-chip ${currentFilter === cat ? "active" : ""}`}
-                onClick={() => onFilterChange(cat)}
-              >
-                {cat}
-              </button>
-            ))}
+      {/* 模板列表 */}
+      <div className="ep-template-list">
+        {loading ? (
+          <div className="ep-loading">
+            <span>⏳</span>
+            <span>加载中...</span>
           </div>
-
-          {/* 模板列表 */}
-          <div className="ep-template-list">
-            {loading ? (
-              <div className="ep-loading">
-                <span>⏳</span>
-                <span>加载中...</span>
+        ) : templates.length === 0 ? (
+          <div className="ep-empty">
+            <span>📭</span>
+            <span>暂无模板</span>
+          </div>
+        ) : (
+          templates.map((tpl) => (
+            <div
+              key={tpl.id}
+              className={`ep-template-card ${loadedTemplateId === tpl.id ? "active" : ""}`}
+              onClick={() => onLoadTemplate(tpl.id)}
+            >
+              <div className="ep-template-card-header">
+                <span className="ep-template-card-name">{tpl.name}</span>
+                <span className="ep-template-card-mode">{MODE_LABELS[tpl.mode]}</span>
               </div>
-            ) : templates.length === 0 ? (
-              <div className="ep-empty">
-                <span>📭</span>
-                <span>暂无模板</span>
+              <div className="ep-template-card-meta">
+                <span>⏱️ {tpl.estimated_duration}s</span>
+                <span>📐 {tpl.segments.length}片段</span>
               </div>
-            ) : (
-              templates.map((tpl) => (
-                <div
-                  key={tpl.id}
-                  className={`ep-template-card ${loadedTemplateId === tpl.id ? "active" : ""}`}
-                  onClick={() => onLoadTemplate(tpl.id)}
-                >
-                  <div className="ep-template-card-header">
-                    <span className="ep-template-card-name">{tpl.name}</span>
-                    <span className="ep-template-card-mode">{MODE_LABELS[tpl.mode]}</span>
-                  </div>
-                  <div className="ep-template-card-meta">
-                    <span>⏱️ {tpl.estimated_duration}s</span>
-                    <span>📐 {tpl.segments.length}片段</span>
-                  </div>
-                  {tpl.tags.length > 0 && (
-                    <div className="ep-template-card-tags">
-                      {tpl.tags.map((tag) => (
-                        <span key={tag} className="ep-template-tag">
-                          {tag}
-                        </span>
-                      ))}
-                    </div>
-                  )}
+              {tpl.tags.length > 0 && (
+                <div className="ep-template-card-tags">
+                  {tpl.tags.map((tag) => (
+                    <span key={tag} className="ep-template-tag">
+                      {tag}
+                    </span>
+                  ))}
                 </div>
-              ))
-            )}
-          </div>
-        </>
-      )}
-
-      {/* 素材 Tab */}
-      {activeTab === "assets" && (
-        <div className="ep-assets-tab">
-          <AssetSelector
-            assets={mediaAssets}
-            selectedIds={selectedAssetIds}
-            onSelectionChange={onAssetSelect}
-            showQualityFilter={false}
-            showBatchSelect={false}
-            compact
-          />
-        </div>
-      )}
+              )}
+            </div>
+          ))
+        )}
+      </div>
     </div>
   )
 }

apps/web/src/pages/editing-planner/hooks/template-management/useTemplateList.ts

@@ -6,16 +6,13 @@ import {
   type EditingTemplate,
   type TemplateCategory,
 } from "@/api/editing-planner"
-import { getMediaAssets, type MediaAsset } from "@/api/template-editor"
 import { FILTER_CATEGORIES } from "../../constants"
 
 /**
  * 模板列表 + 分类 + 筛选搜索
+ * 模板编辑器只负责模板规则定义，不再加载/管理业务素材。
  */
-export function useTemplateList(
-  setMediaAssets: (assets: MediaAsset[]) => void,
-  initialTemplateId: string | null,
-) {
+export function useTemplateList(initialTemplateId: string | null) {
   const [templates, setTemplates] = useState<EditingTemplate[]>([])
   const [categories, setCategories] = useState<TemplateCategory[]>([])
   const [loadingTemplates, setLoadingTemplates] = useState(false)
@@ -24,26 +21,20 @@ export function useTemplateList(
   const [loadedTemplateId, setLoadedTemplateId] = useState<string | null>(initialTemplateId)
 
   /**
-   * 并行加载模板列表、分类、素材库
-   * 三个接口无依赖关系，用 Promise.all 并发
+   * 并行加载模板列表和分类（两者无依赖关系）
    */
   const loadTemplates = useCallback(async () => {
     setLoadingTemplates(true)
     try {
-      const [tpls, cats, assets] = await Promise.all([
-        getEditingTemplates(),
-        getTemplateCategories(),
-        getMediaAssets(),
-      ])
+      const [tpls, cats] = await Promise.all([getEditingTemplates(), getTemplateCategories()])
       setTemplates(tpls)
       setCategories(cats)
-      setMediaAssets(assets)
     } catch {
       message.error("加载模板失败")
     } finally {
       setLoadingTemplates(false)
     }
-  }, [setMediaAssets])
+  }, [])
 
   useEffect(() => {
     loadTemplates()

apps/web/src/pages/editing-planner/hooks/template-management/useTemplateSave.ts

@@ -98,8 +98,8 @@ export function useTemplateSave(options: UseTemplateSaveOptions) {
         estimated_duration: totalDuration,
         segments: clips.map((c, i) => ({
           segment_order: i,
-          duration_min: Math.max(1, c.duration - 2),
-          duration_max: c.duration + 2,
+          duration_min: c.duration,
+          duration_max: c.duration,
           material_type: c.type === "voice" ? "voiceover" : "video",
           transition: c.transition
             ? { type: c.transition.type, duration: c.transition.duration }

apps/web/src/pages/editing-planner/hooks/useTemplateManagement.ts

@@ -1,6 +1,6 @@
 import { useState, useCallback, type Dispatch, type SetStateAction } from "react"
 import type { TemplateMode } from "@/api/editing-planner"
-import type { MediaAsset, TitleConfig } from "@/api/template-editor"
+import type { TitleConfig } from "@/api/template-editor"
 import type {
   ClipData,
   WatermarkConfig,
@@ -24,7 +24,6 @@ interface UseTemplateManagementParams {
   resetClips: (clips: ClipData[]) => void
   setClips: (updater: (prev: ClipData[]) => ClipData[]) => void
   setSelectedClipId: (id: string | null) => void
-  setMediaAssets: (assets: MediaAsset[]) => void
   setTitleConfig: Dispatch<SetStateAction<TitleConfig>>
   setSubtitleSettings: Dispatch<SetStateAction<SubtitleStyleConfig>>
   setBgmSettings: Dispatch<SetStateAction<BgmMixConfig>>
@@ -52,7 +51,6 @@ export const useTemplateManagement = (params: UseTemplateManagementParams) => {
     resetClips,
     setClips,
     setSelectedClipId,
-    setMediaAssets,
     setTitleConfig,
     setSubtitleSettings,
     setBgmSettings,
@@ -86,7 +84,7 @@ export const useTemplateManagement = (params: UseTemplateManagementParams) => {
     filteredTemplates,
     currentTemplate,
     loadTemplates,
-  } = useTemplateList(setMediaAssets, urlTemplateId || null)
+  } = useTemplateList(urlTemplateId || null)
 
   /* ── 保存 ── */
   const {

apps/web/src/pages/generate/GeneratePage.tsx

@@ -1,15 +1,15 @@
 /**
  * 智能剪辑页面 — 前端实时预览架构
- * 7 步向导：选择模板 → 素材 → 配音 → 标题 → 预览 → 封面 → 确认生成
- * 左右布局：左侧 generate-form + 右侧 generate-preview
+ * 6 步向导：选择模板 → 素材 → 配音 → 标题(含预览) → 确认生成 → 选择封面
  *
  * 架构：
- * - Step4+ 右侧预览面板使用 FrontendPreviewPlayer 实时播放素材片段
- * - 标题样式编辑时 CSS 层实时叠加预览，所见即所得
+ * - 步骤 4 右侧显示 FrontendPreviewPlayer 实时预览
+ * - 步骤 5 右侧内联播放生成中的/最终视频
+ * - 步骤 6 封面从最终成片中智能选帧（MediaKit）
  * - 点"确认生成"时调用 createGenerationTask 创建一次服务器渲染任务
  */
-import React, { useMemo, useState, useEffect, useRef } from "react"
-import { Modal, message } from "antd"
+import React, { useMemo, useState, useEffect, useRef, useCallback } from "react"
+import { message } from "antd"
 import { useNavigate } from "react-router-dom"
 import type { VoiceClone } from "@/api/voice-clone"
 import { useQuery } from "@tanstack/react-query"
@@ -20,9 +20,8 @@ import {
   calculateTotalVideoDuration,
   estimateTotalVideoDuration,
 } from "./utils/calculateTotalVideoDuration"
-import GenerateStepsBar from "./components/GenerateStepsBar"
-import GenerateResultPanel from "./components/GenerateResultPanel"
 import FrontendPreviewPlayer from "./components/FrontendPreviewPlayer"
+import GenerateStepsBar from "./components/GenerateStepsBar"
 import GenerateStepContent from "./components/GenerateStepContent"
 import GenerateStepActions from "./components/GenerateStepActions"
 import { useGenerateFormState } from "./hooks/useGenerateFormState"
@@ -64,8 +63,6 @@ const GeneratePage: React.FC = () => {
     presetVoices,
     cloneModalOpen,
     setCloneModalOpen,
-    generateCount,
-    setGenerateCount,
     videoRatio,
     duration,
     style,
@@ -73,14 +70,12 @@ const GeneratePage: React.FC = () => {
     bgm,
     editPlanId,
     sourceEditPlanId,
-    previewVideo,
-    setPreviewVideo,
-    previewModalOpen,
-    setPreviewModalOpen,
     previewTaskId,
     setPreviewTaskId,
     storedSourceEditPlanId,
     setStoredSourceEditPlanId,
+    serverClips,
+    setServerClips,
   } = formState
 
   /* ── 标题样式回调 ── */
@@ -89,7 +84,7 @@ const GeneratePage: React.FC = () => {
     onTitleSettingsChange: setTitleSettings,
   })
 
-  /* ── 配音预览音频（TTS 试听）── */
+  /* ── 配音素材库（TTS 试听）── */
   const { data: voiceMaterials = [] } = useQuery({
     queryKey: ["assets", "voice"],
     queryFn: () => getAssetsByKind("voice", { limit: 50 }),
@@ -99,30 +94,24 @@ const GeneratePage: React.FC = () => {
   const ttsAbortRef = useRef<AbortController | null>(null)
 
   useEffect(() => {
-    // 如果 selectedVoice 是已上传的配音素材，直接用 file_url
     const voiceAsset = voiceMaterials.find((m) => m.id === selectedVoice)
     if (voiceAsset?.file_url) {
       setPreviewVoiceAudioUrl(voiceAsset.file_url)
       return
     }
 
-    // 没有选中的 voice 或标题，跳过
     const voiceId = selectedClonedVoice || selectedVoice
     if (!voiceId || !titleSettings.title) {
       setPreviewVoiceAudioUrl(null)
       return
     }
 
-    // 预设音色 / 克隆音色 → 调 TTS 合成
     ttsAbortRef.current?.abort()
     const controller = new AbortController()
     ttsAbortRef.current = controller
     let cancelled = false
 
-    previewTts({
-      text: titleSettings.title,
-      voice_id: voiceId,
-    })
+    previewTts({ text: titleSettings.title, voice_id: voiceId })
       .then((res) => {
         if (!cancelled && res.audio_url) {
           setPreviewVoiceAudioUrl(res.audio_url)
@@ -174,7 +163,16 @@ const GeneratePage: React.FC = () => {
 
   /* ── 加载素材详情（供前端预览播放器使用 + 配音时长校验） ── */
   const previewAssetsEnabled = previewAssetIds.length > 0
-  const { assets: previewAssets } = usePreviewAssets(previewAssetIds, previewAssetsEnabled)
+  const { assets: previewAssets, ready: previewAssetsReady } = usePreviewAssets(
+    previewAssetIds,
+    previewAssetsEnabled,
+  )
+
+  /* ── 预览就绪：素材已加载，且有模板 ── */
+  const previewReady = useMemo(
+    () => previewAssetsReady && !!currentTemplate,
+    [previewAssetsReady, currentTemplate],
+  )
 
   /* ── 视频总时长计算 ── */
   const totalVideoDuration = useMemo(() => {
@@ -183,17 +181,6 @@ const GeneratePage: React.FC = () => {
     return estimateTotalVideoDuration(currentTemplate ?? undefined)
   }, [previewAssets, currentTemplate])
 
-  /* ── 步骤导航 ── */
-  const { goNext, goPrev } = useStepNavigation({
-    currentStep,
-    setCurrentStep,
-    selectedTemplate,
-    materialMode,
-    selectedMaterials,
-    smartSelectedIds,
-    titleSettings,
-  })
-
   /* ── 视频生成核心逻辑 ── */
   const {
     generating,
@@ -221,7 +208,6 @@ const GeneratePage: React.FC = () => {
     duration,
     autoSubtitles,
     bgm,
-    generateCount,
     sourceEditPlanId: storedSourceEditPlanId || sourceEditPlanId,
     previewTaskId,
     bgmConfig,
@@ -231,6 +217,38 @@ const GeneratePage: React.FC = () => {
     },
   })
 
+  /* ── 步骤4「确认生成视频」：校验标题/预览 → 创建最终渲染任务 → 成功后进入步骤5 ── */
+  const handleConfirmGenerate = useCallback(async () => {
+    if (!titleSettings.title.trim()) {
+      message.warning("请选择或输入标题")
+      return
+    }
+    if (!previewReady) {
+      message.warning("预览视频正在加载，请稍候")
+      return
+    }
+    const ok = await handleGenerate()
+    if (ok) {
+      setCurrentStep(5)
+    }
+  }, [titleSettings.title, previewReady, handleGenerate, setCurrentStep])
+
+  /* ── 步骤导航 ── */
+  const { goNext, goPrev } = useStepNavigation({
+    currentStep,
+    setCurrentStep,
+    selectedTemplate,
+    materialMode,
+    selectedMaterials,
+    smartSelectedIds,
+    titleSettings,
+    previewReady,
+    generated,
+  })
+
+  /* ── 最终成片（步骤5/6 右侧播放） ── */
+  const finalVideo = generatedVideos[0]
+
   /* ================================================================
      渲染
      ================================================================ */
@@ -241,7 +259,7 @@ const GeneratePage: React.FC = () => {
 
       <GenerateStepsBar currentStep={currentStep} onStepClick={setCurrentStep} />
 
-      <div className="xx-generate-layout">
+      <div className={`xx-generate-layout${currentStep < 4 ? " full-width" : ""}`}>
         {/* ════ 左侧：表单区 ════ */}
         <div className="xx-generate-form">
           <GenerateStepContent
@@ -267,16 +285,14 @@ const GeneratePage: React.FC = () => {
             onApplyPreset={styleUpdaters.applyPreset}
             activePreset={styleUpdaters.activePreset}
             titlePresets={styleUpdaters.titlePresets}
-            onPreviewTaskCreated={setPreviewTaskId}
-            onSourceEditPlanIdExtracted={setStoredSourceEditPlanId}
             bgm={bgm}
             bgmConfig={bgmConfig}
             coverSettings={coverSettings}
             onCoverSettingsChange={setCoverSettings}
-            duration={duration}
             selectedVoice={selectedVoice}
             onSelectedVoiceChange={setSelectedVoice}
             totalVideoDuration={totalVideoDuration}
+            onServerClipsChange={setServerClips}
             voiceMode={voiceMode}
             onVoiceModeChange={setVoiceMode}
             selectedClonedVoice={selectedClonedVoice}
@@ -286,8 +302,6 @@ const GeneratePage: React.FC = () => {
             hasProcessing={hasProcessing}
             cloneModalOpen={cloneModalOpen}
             onCloneModalOpenChange={setCloneModalOpen}
-            generateCount={generateCount}
-            onGenerateCountChange={setGenerateCount}
             generating={generating}
             generated={generated}
             generateError={generateError}
@@ -302,77 +316,67 @@ const GeneratePage: React.FC = () => {
             currentStep={currentStep}
             onPrev={goPrev}
             onNext={goNext}
-            onGenerate={handleGenerate}
+            onConfirmGenerate={handleConfirmGenerate}
             generating={generating}
             generated={generated}
             generateError={generateError}
           />
         </div>
 
-        {/* ════ 右侧：预览 + 结果 ════ */}
+        {/* ════ 右侧：步骤4实时预览，步骤5/6最终视频 ════ */}
         <div className="xx-generate-right-col">
-          {currentStep >= 4 && !!currentTemplate && (
+          {currentStep === 4 && !!currentTemplate && (
             <FrontendPreviewPlayer
               assets={previewAssets}
               template={currentTemplate}
               videoRatio={videoRatio}
               ready={previewAssets.length > 0}
+              serverClips={serverClips}
               voiceAudioUrl={previewVoiceAudioUrl || undefined}
               titleSettings={{
                 title: titleSettings.title,
                 size: titleSettings.size,
                 font: titleSettings.font,
                 color: titleSettings.color,
-                position: titleSettings.position as "top" | "center" | "bottom",
+                position: titleSettings.position as "top" | "center" | "bottom" | "custom",
                 bold: titleSettings.bold,
                 italic: titleSettings.italic,
                 stroke: titleSettings.stroke,
                 shadow: titleSettings.shadow,
+                posX: titleSettings.posX,
+                posY: titleSettings.posY,
               }}
+              onTitlePositionChange={styleUpdaters.updateTitlePosition}
             />
           )}
-          {currentStep >= 6 && (
-            <GenerateResultPanel
-              generated={generated}
-              generating={generating}
-              progress={progress}
-              generateError={generateError}
-              generatedVideos={generatedVideos}
-              onVideoPreview={(video) => {
-                setPreviewVideo(video)
-                setPreviewModalOpen(true)
-              }}
-              onDownload={handleDownload}
-              onShare={handleShare}
-              onGoToLibrary={() => navigate("/app/products")}
-            />
+          {currentStep >= 5 && generated && finalVideo && (
+            <div className="xx-inline-video-player">
+              <video
+                src={finalVideo.download_url || finalVideo.file_url}
+                controls
+                autoPlay={currentStep === 5}
+                style={{ width: "100%", maxHeight: "70vh", objectFit: "contain", borderRadius: 12 }}
+                poster={finalVideo.thumbnail_url || undefined}
+              />
+              <div style={{ display: "flex", gap: 8, marginTop: 12, justifyContent: "center" }}>
+                <button className="xx-btn xx-btn-ghost xx-btn-sm" onClick={handleDownload}>
+                  ⬇️ 下载
+                </button>
+                <button className="xx-btn xx-btn-ghost xx-btn-sm" onClick={handleShare}>
+                  🔗 分享
+                </button>
+                <button
+                  className="xx-btn xx-btn-ghost xx-btn-sm"
+                  onClick={() => navigate("/app/products")}
+                >
+                  📁 前往成片库
+                </button>
+              </div>
+            </div>
           )}
         </div>
       </div>
 
-      {/* 视频预览弹窗 */}
-      <Modal
-        className="xx-preview-modal"
-        open={previewModalOpen}
-        onCancel={() => setPreviewModalOpen(false)}
-        footer={null}
-        width="80vw"
-        centered
-        destroyOnClose
-      >
-        {previewVideo && (
-          <div className="xx-preview-modal-content">
-            <video
-              src={previewVideo.download_url || previewVideo.file_url}
-              controls
-              autoPlay
-              style={{ width: "100%", maxHeight: "70vh", objectFit: "contain" }}
-              poster={previewVideo.thumbnail_url || undefined}
-            />
-          </div>
-        )}
-      </Modal>
-
       {/* 音色克隆弹窗 */}
       <CloneModal
         open={cloneModalOpen}

apps/web/src/pages/generate/components/FrontendPreviewPlayer.tsx

@@ -16,6 +16,7 @@ import {
 } from "@ant-design/icons"
 import type { AssetItem } from "@/api/assets"
 import type { EditingTemplate } from "@/api/editing-planner"
+import type { EditPlanClip } from "@/api/template-editor"
 import { useSegmentScheduler, type PlaybackSegment } from "../hooks/useSegmentScheduler"
 import { useCanvasPlayer } from "../hooks/useCanvasPlayer"
 
@@ -24,18 +25,22 @@ interface FrontendPreviewPlayerProps {
   template: EditingTemplate | null
   videoRatio: string
   ready: boolean
+  serverClips?: EditPlanClip[]
   voiceAudioUrl?: string
   titleSettings?: {
     title: string
     size: number
     font: string
     color: string
-    position: "top" | "center" | "bottom"
+    position: "top" | "center" | "bottom" | "custom"
     bold?: boolean
     italic?: boolean
     stroke?: boolean
     shadow?: boolean
+    posX?: number | null
+    posY?: number | null
   }
+  onTitlePositionChange?: (posX: number, posY: number) => void
 }
 
 function formatTime(seconds: number): string {
@@ -50,9 +55,31 @@ function formatTime(seconds: number): string {
 function buildPlaybackSegments(
   assets: AssetItem[],
   template: EditingTemplate | null,
+  serverClips?: EditPlanClip[],
 ): PlaybackSegment[] {
   if (!assets.length) return []
 
+  // Build asset lookup map
+  const assetMap = new Map(assets.map((a) => [a.id, a]))
+
+  // 优先使用服务端 clips（含随机 start_time 和正确数量），与最终生成结果一致
+  if (serverClips && serverClips.length > 0) {
+    const segments: PlaybackSegment[] = []
+    for (const clip of serverClips) {
+      const asset = assetMap.get(clip.asset_id)
+      if (!asset) continue
+      const assetDuration = asset.duration || asset.metadata?.duration || 30
+      const startTime = clip.start_time || 0
+      const endTime = Math.min(startTime + (clip.duration || assetDuration), assetDuration)
+      const videoUrl = asset.file_url || asset.storage_key
+      segments.push({ assetId: asset.id, videoUrl, startTime, endTime, order: clip.order })
+    }
+    if (segments.length > 0) {
+      return segments.sort((a, b) => a.order - b.order)
+    }
+  }
+
+  // Fallback: 本地构建片段（与旧行为一致）
   const templateSegments = template?.segments || []
   const segments: PlaybackSegment[] = []
 
@@ -76,12 +103,127 @@ function buildPlaybackSegments(
 const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   assets,
   template,
-  videoRatio: _videoRatio,
+  videoRatio,
   ready,
+  serverClips,
   voiceAudioUrl,
   titleSettings,
+  onTitlePositionChange,
 }) => {
-  const segments = useMemo(() => buildPlaybackSegments(assets, template), [assets, template])
+  const segments = useMemo(
+    () => buildPlaybackSegments(assets, template, serverClips),
+    [assets, template, serverClips],
+  )
+
+  // ── ASS 坐标系参数（与后端 ass_subtitle_builder.py 一致） ──
+  const TITLE_MARGIN_TOP = 120
+  const TITLE_MARGIN_BOTTOM = 60
+  const TITLE_MARGIN_SIDE = 40
+  const playRes = (() => {
+    switch (videoRatio) {
+      case "16:9":
+        return { width: 1920, height: 1080 }
+      case "1:1":
+        return { width: 1080, height: 1080 }
+      case "9:16":
+      default:
+        return { width: 1080, height: 1920 }
+    }
+  })()
+  const customTitleXPct =
+    titleSettings?.posX != null && playRes.width > 0
+      ? (titleSettings.posX / playRes.width) * 100
+      : null
+  const customTitleYPct =
+    titleSettings?.posY != null && playRes.height > 0
+      ? (titleSettings.posY / playRes.height) * 100
+      : null
+
+  // ── 拖拽状态（用 ref 避免在每帧渲染中触发重渲染）──
+  const draggingTitleRef = useRef(false)
+  const titleDragRef = useRef<HTMLDivElement>(null)
+  const handleTitlePointerDown = useCallback(
+    (e: React.PointerEvent<HTMLDivElement>) => {
+      if (!onTitlePositionChange || !playerContainerRef.current) return
+      e.preventDefault()
+      e.stopPropagation()
+      ;(e.target as Element).setPointerCapture(e.pointerId)
+      draggingTitleRef.current = true
+      ;(e.currentTarget as HTMLDivElement).style.cursor = "grabbing"
+    },
+    [onTitlePositionChange],
+  )
+  const handleTitlePointerMove = useCallback((e: React.PointerEvent<HTMLDivElement>) => {
+    if (!draggingTitleRef.current || !playerContainerRef.current) return
+    e.preventDefault()
+    e.stopPropagation()
+    // 拖拽过程中直接修改 DOM，不触发 React 渲染（避免频繁重渲染导致换行）
+    if (titleDragRef.current) {
+      const rect = playerContainerRef.current.getBoundingClientRect()
+      const relX = Math.max(0, Math.min(rect.width, e.clientX - rect.left))
+      const relY = Math.max(0, Math.min(rect.height, e.clientY - rect.top))
+      const xpct = (relX / rect.width) * 100
+      const ypct = (relY / rect.height) * 100
+      titleDragRef.current.style.left = `${xpct}%`
+      titleDragRef.current.style.top = `${ypct}%`
+    }
+  }, [])
+  const handleTitlePointerUp = useCallback(
+    (e: React.PointerEvent<HTMLDivElement>) => {
+      if (!draggingTitleRef.current) return
+      draggingTitleRef.current = false
+      // 拖拽结束时才调用 onTitlePositionChange 保存最终位置
+      if (onTitlePositionChange && playerContainerRef.current) {
+        const rect = playerContainerRef.current.getBoundingClientRect()
+        const relX = Math.max(0, Math.min(rect.width, e.clientX - rect.left))
+        const relY = Math.max(0, Math.min(rect.height, e.clientY - rect.top))
+        const posX = Math.round((relX / rect.width) * playRes.width)
+        const posY = Math.round((relY / rect.height) * playRes.height)
+        onTitlePositionChange(posX, posY)
+      }
+      ;(e.currentTarget as HTMLDivElement).style.cursor = "grab"
+      try {
+        if ((e.currentTarget as Element).hasPointerCapture(e.pointerId)) {
+          ;(e.currentTarget as Element).releasePointerCapture(e.pointerId)
+        }
+      } catch {
+        /* ignore */
+      }
+    },
+    [onTitlePositionChange, playRes.width, playRes.height],
+  )
+
+  const playerContainerRef = useRef<HTMLDivElement>(null)
+  const [containerHeight, setContainerHeight] = useState(0)
+  useEffect(() => {
+    const el = playerContainerRef.current
+    if (!el) return
+    const ro = new ResizeObserver((entries) => {
+      for (const entry of entries) {
+        const h = entry.contentRect.height
+        if (h > 0) setContainerHeight(h)
+      }
+    })
+    ro.observe(el)
+    const rect = el.getBoundingClientRect()
+    if (rect.height > 0) setContainerHeight(rect.height)
+    return () => ro.disconnect()
+  }, [])
+
+  // 标题字号按容器高度与 PlayResY 的比例缩放
+  const titleFontSizePx =
+    containerHeight > 0
+      ? ((titleSettings?.size ?? 36) / playRes.height) * containerHeight
+      : (titleSettings?.size ?? 36)
+  const titleSidePct = (TITLE_MARGIN_SIDE / playRes.width) * 100
+  const titleTopPct = (TITLE_MARGIN_TOP / playRes.height) * 100
+  const titleBottomPct = (TITLE_MARGIN_BOTTOM / playRes.height) * 100
+  // 描边/阴影也要按缩放比例放大
+  const titleScale = containerHeight > 0 ? containerHeight / playRes.height : 1
+  const titleStrokeWidth = Math.max(1, 2 * titleScale)
+  const titleShadowBlur = 4 * titleScale
+  const titleShadowOffset = 2 * titleScale
+
   // 默认走原生 video 播放（浏览器硬件解码，独立线程，不阻塞 UI）
   // WebCodecs 仅在明确需要时启用（保留代码作为兜底）
   const useWebCodecs = false
@@ -96,7 +238,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
         fontSize: titleSettings.size,
         fontFamily: titleSettings.font || "思源黑体",
         color: titleSettings.color || "#ffffff",
-        position: titleSettings.position || "bottom",
+        position: titleSettings.position || "top",
         bold: titleSettings.bold,
         stroke: titleSettings.stroke,
         shadow: titleSettings.shadow,
@@ -374,6 +516,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
 
   return (
     <div
+      ref={playerContainerRef}
       style={{
         position: "relative",
         width: "100%",
@@ -434,48 +577,76 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           />
         ))}
 
-      {/* 标题CSS叠加层 — video fallback 路径也要渲染 */}
+      {/* 标题CSS叠加层 — 与后端 ASS 烧录坐标系 1:1 对齐 */}
       {titleSettings?.title && (
         <div
           style={{
             position: "absolute",
-            left: 0,
-            right: 0,
+            inset: 0,
             zIndex: 5,
-            display: "flex",
-            justifyContent: "center",
             pointerEvents: "none",
-            ...(titleSettings.position === "top"
-              ? { top: "10%" }
-              : titleSettings.position === "center"
-                ? { top: "50%", transform: "translateY(-50%)" }
-                : { bottom: "15%" }),
+            overflow: "hidden",
           }}
         >
-          <span
+          <div
             style={{
-              fontSize: titleSettings.size,
-              fontFamily: titleSettings.font || "思源黑体",
-              color: titleSettings.color || "#ffffff",
-              fontWeight: titleSettings.bold ? 700 : 400,
-              fontStyle: titleSettings.italic ? "italic" : "normal",
-              textShadow: [
-                titleSettings.shadow ? "0 2px 8px rgba(0,0,0,0.7)" : undefined,
-                titleSettings.stroke
-                  ? "1px 1px 0 rgba(0,0,0,0.5), -1px -1px 0 rgba(0,0,0,0.5), 1px -1px 0 rgba(0,0,0,0.5), -1px 1px 0 rgba(0,0,0,0.5)"
-                  : undefined,
-                "0 1px 3px rgba(0,0,0,0.4)",
-              ]
-                .filter(Boolean)
-                .join(", "),
-              maxWidth: "90%",
-              textAlign: "center",
-              lineHeight: 1.3,
-              wordBreak: "break-word",
+              position: "absolute",
+              ...(customTitleXPct != null && customTitleYPct != null
+                ? {
+                    left: `${customTitleXPct}%`,
+                    top: `${customTitleYPct}%`,
+                    transform: "translate(-50%, -50%)",
+                    textAlign: "center" as const,
+                  }
+                : {
+                    left: `${titleSidePct}%`,
+                    right: `${titleSidePct}%`,
+                    textAlign: "center" as const,
+                    ...(titleSettings.position === "top"
+                      ? { top: `${titleTopPct}%` }
+                      : titleSettings.position === "center"
+                        ? { top: "50%", transform: "translateY(-50%)" }
+                        : { bottom: `${titleBottomPct}%` }),
+                  }),
+              pointerEvents: "auto",
+              cursor: onTitlePositionChange ? "grab" : "default",
+              touchAction: "none",
+              userSelect: "none",
+              WebkitUserSelect: "none",
+              padding: "8px 12px",
+              boxShadow: "inset 0 0 0 16px transparent",
             }}
+            ref={titleDragRef}
+            onPointerDown={handleTitlePointerDown}
+            onPointerMove={handleTitlePointerMove}
+            onPointerUp={handleTitlePointerUp}
+            onPointerCancel={handleTitlePointerUp}
           >
-            {titleSettings.title}
-          </span>
+            <span
+              style={{
+                fontSize: `${titleFontSizePx}px`,
+                fontFamily: titleSettings.font || "思源黑体",
+                color: titleSettings.color || "#ffffff",
+                fontWeight: titleSettings.bold ? 700 : 400,
+                fontStyle: titleSettings.italic ? "italic" : "normal",
+                lineHeight: 1.05,
+                wordBreak: "break-word",
+                WebkitTextStroke: titleSettings.stroke
+                  ? `${titleStrokeWidth}px #000000`
+                  : undefined,
+                textShadow: titleSettings.shadow
+                  ? `${titleShadowOffset}px ${titleShadowOffset}px ${titleShadowBlur}px rgba(0,0,0,0.8)`
+                  : undefined,
+              }}
+            >
+              {titleSettings.title.split(/[/／]/).map((part, i) => (
+                <span key={i}>
+                  {i > 0 && <br />}
+                  {part}
+                </span>
+              ))}
+            </span>
+          </div>
         </div>
       )}
 

apps/web/src/pages/generate/components/GenerateStepActions.tsx

@@ -1,14 +1,19 @@
 /**
  * GeneratePage 步骤底部操作按钮
+ *
+ * 步骤 1~3：上一步 / 下一步
+ * 步骤 4（标题+预览）：上一步 / 确认生成视频（点击后直接创建最终渲染任务，成功后跳转步骤5）
+ * 步骤 5（确认生成）：上一步 / 下一步（渲染中禁用，渲染完成后可进入封面）
+ * 步骤 6（选择封面）：仅上一步
  */
 import React from "react"
-import { ThunderboltOutlined } from "@ant-design/icons"
 
 export interface GenerateStepActionsProps {
   currentStep: number
   onPrev: () => void
   onNext: () => void
-  onGenerate: () => void
+  /** 步骤4：确认生成视频（校验 + 创建渲染任务 + 成功后进入步骤5） */
+  onConfirmGenerate: () => void | Promise<void>
   generating: boolean
   generated: boolean
   generateError: string | null
@@ -18,36 +23,74 @@ export const GenerateStepActions: React.FC<GenerateStepActionsProps> = ({
   currentStep,
   onPrev,
   onNext,
-  onGenerate,
+  onConfirmGenerate,
   generating,
   generated,
   generateError,
 }) => {
+  const renderPrimaryButton = () => {
+    /* 步骤 1~3：上一步 / 下一步（必填校验由 useStepNavigation.goNext 统一处理） */
+    if (currentStep < 4) {
+      return (
+        <button className="xx-btn xx-btn-primary" onClick={onNext}>
+          下一步 →
+        </button>
+      )
+    }
+
+    /* 步骤 4：确认生成视频（触发按钮在标题页） */
+    if (currentStep === 4) {
+      if (generating) {
+        return (
+          <button className="xx-btn xx-btn-primary" disabled>
+            ⏳ 视频生成中…
+          </button>
+        )
+      }
+      if (generateError) {
+        return (
+          <button className="xx-btn xx-btn-primary" onClick={onConfirmGenerate}>
+            🔄 重新生成视频
+          </button>
+        )
+      }
+      if (generated) {
+        return (
+          <button className="xx-btn xx-btn-primary" onClick={onNext}>
+            下一步 →
+          </button>
+        )
+      }
+      return (
+        <button className="xx-btn xx-btn-primary" onClick={onConfirmGenerate}>
+          ✨ 确认生成视频
+        </button>
+      )
+    }
+
+    /* 步骤 5：渲染中禁用，完成后下一步进入封面 */
+    if (currentStep === 5) {
+      return (
+        <button
+          className="xx-btn xx-btn-primary"
+          onClick={onNext}
+          disabled={generating || !generated}
+        >
+          {generating ? "视频生成中…" : "下一步 →"}
+        </button>
+      )
+    }
+
+    /* 步骤 6（最后一步）：无主按钮 */
+    return null
+  }
+
   return (
     <div className="xx-step-actions">
       <button className="xx-btn xx-btn-ghost" onClick={onPrev} disabled={currentStep === 1}>
         ← 上一步
       </button>
-      {currentStep < 7 ? (
-        <button className="xx-btn xx-btn-primary" onClick={onNext}>
-          下一步 →
-        </button>
-      ) : (
-        <button
-          className="xx-btn xx-btn-primary"
-          onClick={onGenerate}
-          disabled={generating || (generated && !generateError)}
-        >
-          <ThunderboltOutlined />
-          {generating
-            ? "生成中…"
-            : generated && !generateError
-              ? "已生成"
-              : generateError
-                ? "🔄 重新生成"
-                : "✨ 确认生成"}
-        </button>
-      )}
+      {renderPrimaryButton()}
     </div>
   )
 }

apps/web/src/pages/generate/components/GenerateStepContent.tsx

@@ -1,12 +1,10 @@
 /**
  * GeneratePage 步骤内容渲染
- * 根据当前步骤渲染对应的 Step 组件
- * 步骤顺序：模板(1) → 素材(2) → 配音(3) → 标题(4) → 预览(5) → 封面(6) → 确认(7)
- *
- * V24: 移除 Step5 预览生成相关 props，改为纯标题样式编辑
+ * 步骤顺序（6步）：模板(1) → 素材(2) → 配音(3) → 标题(4) → 确认生成(5) → 封面(6)
  */
 import React from "react"
 import type { EditingTemplate } from "@/api/editing-planner"
+import type { EditPlanClip } from "@/api/template-editor"
 import type { PresetVoiceItem } from "@/api/voices"
 import type { VoiceClone } from "@/api/voice-clone"
 import type { CoverConfig } from "../types/cover"
@@ -14,10 +12,9 @@ import type { TitleSettings } from "../types"
 import Step1TemplateSelect from "../components/Step1TemplateSelect"
 import Step2MaterialSelect from "../components/Step2MaterialSelect"
 import Step3VoiceSelect from "../components/Step5VoiceSelect"
-import Step5GeneratePreview from "../components/Step5GeneratePreview"
 import Step4TitleSettings from "../components/Step4TitleSettings"
+import Step5ConfirmGenerate from "../components/Step7ConfirmGenerate"
 import Step6CoverSettings from "../components/Step6CoverSettings"
-import Step7ConfirmGenerate from "../components/Step7ConfirmGenerate"
 import type { GeneratedVideo } from "@/api/template-editor"
 
 export interface GenerateStepContentProps {
@@ -36,7 +33,6 @@ export interface GenerateStepContentProps {
   /* 标题 */
   titleSettings: TitleSettings
   onTitleSettingsChange: (settings: TitleSettings) => void
-  /* 标题样式回调 — Step5 样式面板使用 */
   onUpdatePosition: (position: string) => void
   onUpdateFont: (font: string) => void
   onUpdateSize: (size: number) => void
@@ -50,11 +46,11 @@ export interface GenerateStepContentProps {
   /* 封面 */
   coverSettings: CoverConfig
   onCoverSettingsChange: (settings: CoverConfig) => void
-  duration: number
   /* 配音 */
   selectedVoice: string
   onSelectedVoiceChange: (id: string) => void
   totalVideoDuration?: number
+  onServerClipsChange: (clips: EditPlanClip[]) => void
   voiceMode: "preset" | "custom" | "clone"
   onVoiceModeChange: (mode: "preset" | "custom" | "clone") => void
   selectedClonedVoice: string
@@ -65,8 +61,6 @@ export interface GenerateStepContentProps {
   cloneModalOpen: boolean
   onCloneModalOpenChange: (open: boolean) => void
   /* 生成 */
-  generateCount: number
-  onGenerateCountChange: (n: number) => void
   generating: boolean
   generated: boolean
   generateError: string | null
@@ -76,10 +70,6 @@ export interface GenerateStepContentProps {
   onDismissError: () => void
   /* 其他 */
   presetVoices: PresetVoiceItem[]
-  /** 预览任务创建回调——传递给 Step6CoverSettings */
-  onPreviewTaskCreated?: (taskId: string) => void
-  /** 从预览响应中提取到 source_edit_plan_id 时的回调 */
-  onSourceEditPlanIdExtracted?: (planId: string) => void
   /** BGM 开关 */
   bgm: boolean
   /** BGM 配置（来自模板） */
@@ -112,15 +102,13 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     titlePresets,
     coverSettings,
     onCoverSettingsChange,
-    duration,
     selectedVoice,
     onSelectedVoiceChange,
     totalVideoDuration,
+    onServerClipsChange,
     voiceMode,
     selectedClonedVoice,
     clonedVoices,
-    generateCount,
-    onGenerateCountChange,
     generating,
     generated,
     generateError,
@@ -129,10 +117,6 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     onRetry,
     onDismissError,
     presetVoices,
-    onPreviewTaskCreated,
-    onSourceEditPlanIdExtracted,
-    bgm,
-    bgmConfig,
   } = props
 
   /* 当前模板的 segments，传给 Step2 构建 clips */
@@ -159,6 +143,7 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           onSmartSelectedIdsChange={onSmartSelectedIdsChange}
           selectedTemplate={selectedTemplate}
           templateSegments={templateSegments}
+          onServerClipsChange={onServerClipsChange}
         />
       )
     case 3:
@@ -175,12 +160,6 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           titleSettings={titleSettings}
           onTitleSettingsChange={onTitleSettingsChange}
           selectedTemplate={selectedTemplate}
-        />
-      )
-    case 5:
-      return (
-        <Step5GeneratePreview
-          titleSettings={titleSettings}
           onUpdatePosition={onUpdatePosition}
           onUpdateFont={onUpdateFont}
           onUpdateSize={onUpdateSize}
@@ -193,27 +172,9 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           titlePresets={titlePresets}
         />
       )
-    case 6:
+    case 5:
       return (
-        <Step6CoverSettings
-          coverSettings={coverSettings}
-          onCoverSettingsChange={onCoverSettingsChange}
-          duration={duration}
-          assetIds={materialMode === "auto" ? smartSelectedIds : selectedMaterials}
-          selectedTemplate={selectedTemplate}
-          titleSettings={titleSettings}
-          onPreviewTaskCreated={onPreviewTaskCreated}
-          onSourceEditPlanIdExtracted={onSourceEditPlanIdExtracted}
-          voiceMode={voiceMode}
-          selectedVoice={selectedVoice}
-          selectedClonedVoice={selectedClonedVoice}
-          bgm={bgm}
-          bgmConfig={bgmConfig}
-        />
-      )
-    case 7:
-      return (
-        <Step7ConfirmGenerate
+        <Step5ConfirmGenerate
           templates={userTemplates}
           selectedTemplate={selectedTemplate}
           materialMode={materialMode}
@@ -226,8 +187,6 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           presetVoices={presetVoices}
           clonedVoices={clonedVoices}
           coverSettings={coverSettings}
-          generateCount={generateCount}
-          onGenerateCountChange={onGenerateCountChange}
           generating={generating}
           generated={generated}
           generateError={generateError}
@@ -237,6 +196,16 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           onDismissError={onDismissError}
         />
       )
+    case 6:
+      return (
+        <Step6CoverSettings
+          coverSettings={coverSettings}
+          onCoverSettingsChange={onCoverSettingsChange}
+          selectedTemplate={selectedTemplate}
+          titleSettings={titleSettings}
+          generatedVideos={generatedVideos}
+        />
+      )
     default:
       return null
   }

apps/web/src/pages/generate/components/PreviewVideoPanel.tsx

@@ -40,17 +40,33 @@ interface PreviewVideoPanelProps {
 }
 
 /* ── ASS 坐标系参数（与后端 ass_subtitle_builder.py 一致） ── */
-const ASS_VIDEO_HEIGHT = 720
-const ASS_TITLE_MARGIN_TOP = 60
-const ASS_TITLE_MARGIN_BOTTOM = 60
-const ASS_TITLE_MARGIN_SIDE = 40
+const TITLE_MARGIN_TOP = 120
+const TITLE_MARGIN_BOTTOM = 60
+const TITLE_MARGIN_SIDE = 40
 
-function getPositionStyle(position: string): React.CSSProperties {
-  const sidePercent = (ASS_TITLE_MARGIN_SIDE / 1280) * 100
+/** 根据视频比例返回后端实际渲染分辨率（PlayResX × PlayResY） */
+function getResolution(ratio: string): { width: number; height: number } {
+  switch (ratio) {
+    case "16:9":
+      return { width: 1920, height: 1080 }
+    case "1:1":
+      return { width: 1080, height: 1080 }
+    case "9:16":
+    default:
+      return { width: 1080, height: 1920 }
+  }
+}
+
+function getPositionStyle(
+  position: string,
+  playResX: number,
+  playResY: number,
+): React.CSSProperties {
+  const sidePercent = (TITLE_MARGIN_SIDE / playResX) * 100
   switch (position) {
     case "bottom":
       return {
-        bottom: `${(ASS_TITLE_MARGIN_BOTTOM / ASS_VIDEO_HEIGHT) * 100}%`,
+        bottom: `${(TITLE_MARGIN_BOTTOM / playResY) * 100}%`,
         left: `${sidePercent}%`,
         right: `${sidePercent}%`,
         textAlign: "center",
@@ -66,7 +82,7 @@ function getPositionStyle(position: string): React.CSSProperties {
     case "top":
     default:
       return {
-        top: `${(ASS_TITLE_MARGIN_TOP / ASS_VIDEO_HEIGHT) * 100}%`,
+        top: `${(TITLE_MARGIN_TOP / playResY) * 100}%`,
         left: `${sidePercent}%`,
         right: `${sidePercent}%`,
         textAlign: "center",
@@ -74,11 +90,16 @@ function getPositionStyle(position: string): React.CSSProperties {
   }
 }
 
-function buildTitleStyle(settings: TitleSettings, containerHeight: number): React.CSSProperties {
+function buildTitleStyle(
+  settings: TitleSettings,
+  containerHeight: number,
+  playResY: number,
+): React.CSSProperties {
+  // 字号按容器高度与 PlayResY 的比例缩放，不设上限（与后端一致）
   const fontSizePx =
     containerHeight > 0
-      ? (Math.min(settings.size, 96) / ASS_VIDEO_HEIGHT) * containerHeight
-      : (Math.min(settings.size, 96) / ASS_VIDEO_HEIGHT) * 400
+      ? (settings.size / playResY) * containerHeight
+      : (settings.size / playResY) * 400
 
   const base: React.CSSProperties = {
     fontFamily: getFontFamily(settings.font),
@@ -86,12 +107,10 @@ function buildTitleStyle(settings: TitleSettings, containerHeight: number): Reac
     color: settings.color || "#ffffff",
     fontWeight: settings.bold ? 700 : 400,
     fontStyle: settings.italic ? "italic" : "normal",
-    lineHeight: 1.3,
+    lineHeight: 1.05,
     wordBreak: "break-word",
     pointerEvents: "none",
     userSelect: "none",
-    paddingLeft: `${(ASS_TITLE_MARGIN_SIDE / 1280) * 100}%`,
-    paddingRight: `${(ASS_TITLE_MARGIN_SIDE / 1280) * 100}%`,
   }
   if (settings.stroke) base.WebkitTextStroke = "1px #000000"
   if (settings.shadow) base.textShadow = "2px 2px 4px rgba(0,0,0,0.8)"
@@ -99,7 +118,10 @@ function buildTitleStyle(settings: TitleSettings, containerHeight: number): Reac
 }
 
 /** CSS 标题实时预览覆盖层 */
-const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSettings }) => {
+const TitleOverlay: React.FC<{ titleSettings: TitleSettings; videoRatio: string }> = ({
+  titleSettings,
+  videoRatio,
+}) => {
   const containerRef = useRef<HTMLDivElement>(null)
   const [containerHeight, setContainerHeight] = useState(400)
 
@@ -118,12 +140,15 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
     return () => ro.disconnect()
   }, [])
 
+  const { width: playResX, height: playResY } = getResolution(videoRatio)
+
   const positionStyle = useMemo(
-    () => getPositionStyle(titleSettings.position),
-    [titleSettings.position],
+    () => getPositionStyle(titleSettings.position, playResX, playResY),
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+    [titleSettings.position, playResX, playResY],
   )
   const titleStyle = useMemo(
-    () => buildTitleStyle(titleSettings, containerHeight),
+    () => buildTitleStyle(titleSettings, containerHeight, playResY),
     // eslint-disable-next-line react-hooks/exhaustive-deps
     [
       containerHeight,
@@ -134,6 +159,7 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
       titleSettings.italic,
       titleSettings.stroke,
       titleSettings.shadow,
+      playResY,
     ],
   )
 
@@ -151,7 +177,7 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
       }}
     >
       <div style={{ ...positionStyle, ...titleStyle, position: "absolute" }}>
-        {displayTitle.split("/").map((part, i) => (
+        {displayTitle.split(/[/／]/).map((part, i) => (
           <span key={i}>
             {i > 0 && <br />}
             {part}
@@ -266,7 +292,9 @@ export const PreviewVideoPanel: React.FC<PreviewVideoPanelProps> = ({
         )}
 
         {/* 标题样式实时预览层（仅在有视频时叠加） */}
-        {isReady && titleSettings && <TitleOverlay titleSettings={titleSettings} />}
+        {isReady && titleSettings && (
+          <TitleOverlay titleSettings={titleSettings} videoRatio={videoRatio} />
+        )}
 
         {/* stale 遮罩：配置变更提示 */}
         {isStale && (

apps/web/src/pages/generate/components/Step2MaterialSelect.tsx

@@ -3,6 +3,7 @@
  */
 import React from "react"
 import type { TemplateSegment } from "@/api/templates/types"
+import type { EditPlanClip } from "@/api/template-editor"
 import { useStep2Materials } from "../hooks/useStep2Materials"
 import MaterialModeTabs from "./material/MaterialModeTabs"
 import ManualMaterialList from "./material/ManualMaterialList"
@@ -20,6 +21,8 @@ interface Step2MaterialSelectProps {
   selectedTemplate?: string
   /** 当前模板的 segments（用于构建 clips duration） */
   templateSegments?: TemplateSegment[]
+  /** 服务端 clips 创建成功后的回调 */
+  onServerClipsChange?: (clips: EditPlanClip[]) => void
 }
 
 const Step2MaterialSelect: React.FC<Step2MaterialSelectProps> = (props) => {
@@ -59,8 +62,9 @@ const Step2MaterialSelect: React.FC<Step2MaterialSelectProps> = (props) => {
           </div>
 
           <ManualMaterialList
-            materials={m.materials}
+            materials={m.selectableMaterials}
             materialsLoading={m.materialsLoading}
+            allExhausted={m.allMaterialsExhausted}
             selectedMaterials={m.selectedMaterials}
             onToggle={m.handleToggleMaterial}
           />
@@ -74,21 +78,11 @@ const Step2MaterialSelect: React.FC<Step2MaterialSelectProps> = (props) => {
             onMatch={m.handleSmartMatch}
             hasMatched={m.hasMatched}
             onRefresh={m.handleRefreshMatch}
-            materialsCount={m.materials.items.length}
+            materialsCount={m.selectableMaterials.items.length}
             loading={m.materialsLoading}
           />
 
-          <SmartMatchResults
-            matchedAssets={m.smartMatchedResults}
-            selectedIds={m.smartSelectedIds}
-            matching={m.smartMatching}
-            hasMatched={m.hasMatched}
-            onToggle={m.handleToggleSmartSelect}
-            onSelectAll={m.handleSelectAllMatched}
-            onClear={m.handleClearSmartSelect}
-            formatDuration={m.formatDuration}
-            selectedTotalDuration={m.smartSelectedTotalDuration}
-          />
+          <SmartMatchResults matching={m.smartMatching} hasMatched={m.hasMatched} />
         </div>
       )}
     </div>

apps/web/src/pages/generate/components/Step4TitleSettings.tsx

@@ -1,23 +1,50 @@
 /**
- * Step 4 标题设置组件
- * 仅包含标题文字输入 + AI 标题生成
- * 标题样式面板已迁移到 Step5（生成预览页面）
+ * Step 4 选择标题（合并原 Step4 标题输入 + Step5 标题样式面板）
+ *
+ * 左侧：标题文字输入 + AI生成标题 + 样式设置（位置/字号/字体/颜色/样式/预设）
+ * 右侧：FrontendPreviewPlayer 实时预览（由 GeneratePage 统一渲染）
  */
 import React from "react"
 import { AutoComplete } from "antd"
+import { PlayCircleOutlined } from "@ant-design/icons"
 import type { TitleSettings } from "../types"
+import { POSITION_OPTIONS, FONT_OPTIONS } from "../constants"
 import { useStep4Title } from "../hooks/useStep4Title"
 import AiTitleGenerator from "./title/AiTitleGenerator"
+import TitleStylePanel from "./title/TitleStylePanel"
 
 interface Step4TitleSettingsProps {
   titleSettings: TitleSettings
   onTitleSettingsChange: (settings: TitleSettings) => void
   /** 当前选中的模板/草稿 ID，用于自动保存 */
   selectedTemplate?: string
+  /* 标题样式回调 */
+  onUpdatePosition: (position: string) => void
+  onUpdateFont: (font: string) => void
+  onUpdateSize: (size: number) => void
+  onToggleBold: () => void
+  onToggleItalic: () => void
+  onToggleStroke: () => void
+  onToggleShadow: () => void
+  onApplyPreset: (presetKey: string) => void
+  activePreset: string | null
+  titlePresets: { key: string; label: string; previewStyle: React.CSSProperties }[]
 }
 
 const Step4TitleSettings: React.FC<Step4TitleSettingsProps> = (props) => {
   const t = useStep4Title(props)
+  const {
+    onUpdatePosition,
+    onUpdateFont,
+    onUpdateSize,
+    onToggleBold,
+    onToggleItalic,
+    onToggleStroke,
+    onToggleShadow,
+    onApplyPreset,
+    activePreset,
+    titlePresets,
+  } = props
 
   return (
     <div className="xx-form-section">
@@ -112,6 +139,42 @@ const Step4TitleSettings: React.FC<Step4TitleSettingsProps> = (props) => {
           </div>
         </>
       )}
+
+      {/* 标题样式面板（原 Step5） */}
+      <div
+        style={{
+          display: "flex",
+          alignItems: "center",
+          gap: 8,
+          padding: "10px 14px",
+          background: "rgba(59, 130, 246, 0.08)",
+          borderRadius: 8,
+          marginTop: 16,
+          marginBottom: 12,
+          border: "1px solid rgba(59, 130, 246, 0.15)",
+        }}
+      >
+        <PlayCircleOutlined style={{ fontSize: 16, color: "#3b82f6" }} />
+        <span style={{ fontSize: 12, color: "var(--text-secondary, #666)" }}>
+          右侧为实时预览，调整样式即时生效
+        </span>
+      </div>
+
+      <TitleStylePanel
+        settings={t.titleSettings}
+        onUpdatePosition={onUpdatePosition}
+        onUpdateFont={onUpdateFont}
+        onUpdateSize={onUpdateSize}
+        onToggleBold={onToggleBold}
+        onToggleItalic={onToggleItalic}
+        onToggleStroke={onToggleStroke}
+        onToggleShadow={onToggleShadow}
+        onApplyPreset={onApplyPreset}
+        activePreset={activePreset}
+        titlePresets={titlePresets}
+        POSITION_OPTIONS={POSITION_OPTIONS}
+        FONT_OPTIONS={FONT_OPTIONS}
+      />
     </div>
   )
 }

apps/web/src/pages/generate/components/Step5GeneratePreview.tsx

@@ -1,83 +0,0 @@
-/**
- * Step 5 预览设置组件
- *
- * 前端实时预览架构：
- * - 右侧面板使用 FrontendPreviewPlayer 实时播放素材片段
- * - 标题样式可实时调整，CSS 层即时叠加预览
- * - 点"确认生成"时触发一次服务器渲染
- */
-import React from "react"
-import { PlayCircleOutlined } from "@ant-design/icons"
-import { POSITION_OPTIONS, FONT_OPTIONS } from "../constants"
-import type { TitleSettings } from "../types"
-import TitleStylePanel from "./title/TitleStylePanel"
-
-interface Step5GeneratePreviewProps {
-  titleSettings: TitleSettings
-  onUpdatePosition: (position: string) => void
-  onUpdateFont: (font: string) => void
-  onUpdateSize: (size: number) => void
-  onToggleBold: () => void
-  onToggleItalic: () => void
-  onToggleStroke: () => void
-  onToggleShadow: () => void
-  onApplyPreset: (presetKey: string) => void
-  activePreset: string | null
-  titlePresets: { key: string; label: string; previewStyle: React.CSSProperties }[]
-}
-
-const Step5GeneratePreview: React.FC<Step5GeneratePreviewProps> = ({
-  titleSettings,
-  onUpdatePosition,
-  onUpdateFont,
-  onUpdateSize,
-  onToggleBold,
-  onToggleItalic,
-  onToggleStroke,
-  onToggleShadow,
-  onApplyPreset,
-  activePreset,
-  titlePresets,
-}) => {
-  return (
-    <div className="xx-form-section">
-      <h3>🎬 预览设置</h3>
-
-      <div
-        style={{
-          display: "flex",
-          alignItems: "center",
-          gap: 8,
-          padding: "12px 16px",
-          background: "rgba(59, 130, 246, 0.08)",
-          borderRadius: 8,
-          marginBottom: 16,
-          border: "1px solid rgba(59, 130, 246, 0.15)",
-        }}
-      >
-        <PlayCircleOutlined style={{ fontSize: 18, color: "#3b82f6" }} />
-        <span style={{ fontSize: 13, color: "var(--text-secondary, #666)" }}>
-          右侧为实时预览，选完素材即可播放。确认生成后服务器渲染最终视频
-        </span>
-      </div>
-
-      <TitleStylePanel
-        settings={titleSettings}
-        onUpdatePosition={onUpdatePosition}
-        onUpdateFont={onUpdateFont}
-        onUpdateSize={onUpdateSize}
-        onToggleBold={onToggleBold}
-        onToggleItalic={onToggleItalic}
-        onToggleStroke={onToggleStroke}
-        onToggleShadow={onToggleShadow}
-        onApplyPreset={onApplyPreset}
-        activePreset={activePreset}
-        titlePresets={titlePresets}
-        POSITION_OPTIONS={POSITION_OPTIONS}
-        FONT_OPTIONS={FONT_OPTIONS}
-      />
-    </div>
-  )
-}
-
-export default Step5GeneratePreview

apps/web/src/pages/generate/components/Step5VoiceSelect.tsx

@@ -27,6 +27,13 @@ const getFileSize = (item: AssetItem): number => {
   return item.file_size ?? (item.metadata?.file_size as number) ?? 0
 }
 
+/** 是否为 AI 音色（克隆/预置音色模型：无固定时长、无实体音频文件，按脚本实时合成） */
+const isAiVoice = (item: AssetItem): boolean => {
+  const duration = getDuration(item)
+  const size = getFileSize(item)
+  return (!duration || duration <= 0) && (!size || size <= 0)
+}
+
 const formatDuration = (seconds?: number): string => {
   if (!seconds || seconds <= 0) return "00:00"
   const m = Math.floor(seconds / 60)
@@ -96,10 +103,10 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
   /** 选中素材（含时长校验） */
   const handleSelect = useCallback(
     (id: string) => {
-      // 如果启用了时长校验，且配音时长不足
+      // 如果启用了时长校验，且配音时长不足（AI 音色按脚本实时合成，不参与时长校验）
       if (totalVideoDuration > 0) {
         const material = materials.find((m) => m.id === id)
-        if (material && getDuration(material) < totalVideoDuration) {
+        if (material && !isAiVoice(material) && getDuration(material) < totalVideoDuration) {
           setPendingVoiceId(id)
           setDurationWarningOpen(true)
           return
@@ -127,7 +134,7 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
 
   /** 跳转到配音库上传 */
   const handleGoToUpload = useCallback(() => {
-    navigate("/app/voices")
+    navigate("/app/voices?tab=material&upload=1")
   }, [navigate])
 
   // 加载中状态
@@ -280,25 +287,29 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
                   alignItems: "center",
                 }}
               >
-                <span style={{ display: "flex", alignItems: "center", gap: 4 }}>
-                  {formatDuration(getDuration(item))}
-                  {totalVideoDuration > 0 && getDuration(item) < Number(totalVideoDuration) && (
-                    <span
-                      style={{
-                        color: "#ff4d4f",
-                        fontSize: 11,
-                        fontWeight: 500,
-                        display: "inline-flex",
-                        alignItems: "center",
-                        gap: 2,
-                      }}
-                    >
-                      <WarningOutlined />
-                      时长不足
-                    </span>
-                  )}
-                </span>
-                <span>{formatFileSize(getFileSize(item))}</span>
+                {isAiVoice(item) ? (
+                  <span style={{ color: "#1677ff", fontWeight: 500 }}>AI 音色</span>
+                ) : (
+                  <span style={{ display: "flex", alignItems: "center", gap: 4 }}>
+                    {formatDuration(getDuration(item))}
+                    {totalVideoDuration > 0 && getDuration(item) < Number(totalVideoDuration) && (
+                      <span
+                        style={{
+                          color: "#ff4d4f",
+                          fontSize: 11,
+                          fontWeight: 500,
+                          display: "inline-flex",
+                          alignItems: "center",
+                          gap: 2,
+                        }}
+                      >
+                        <WarningOutlined />
+                        时长不足
+                      </span>
+                    )}
+                  </span>
+                )}
+                <span>{isAiVoice(item) ? "按文本合成" : formatFileSize(getFileSize(item))}</span>
               </div>
             </div>
           )

apps/web/src/pages/generate/components/Step6CoverSettings.tsx

@@ -1,6 +1,8 @@
 import React from "react"
 import { Modal, Spin } from "antd"
 import type { CoverConfig } from "../types/cover"
+import type { GeneratedVideo } from "@/api/template-editor"
+import type { TitleSettings } from "../types"
 import { useStep6Cover } from "../hooks/useStep6Cover"
 import Button from "@/components/ui/Button"
 import CoverSettingsModal from "./cover-settings/CoverSettingsModal"
@@ -9,27 +11,12 @@ import CoverEditorModal from "./cover-settings/CoverEditorModal"
 interface Step6CoverSettingsProps {
   coverSettings: CoverConfig
   onCoverSettingsChange: (settings: CoverConfig) => void
-  duration: number
-  /** 当前素材 ID 列表，用于智能封面生成 */
-  assetIds?: string[]
   /** 当前选中的模板 ID */
   selectedTemplate?: string
-  /** Step4 标题设置，用于预览视频烧录标题 & 封面叠加标题 */
-  titleSettings?: import("../types").TitleSettings
-  /** 预览任务创建回调——将 task_id 暴露给父组件供 confirmGeneration 复用 */
-  onPreviewTaskCreated?: (taskId: string) => void
-  /** 从预览响应中提取到 source_edit_plan_id 时的回调 */
-  onSourceEditPlanIdExtracted?: (planId: string) => void
-  /** 配音模式 */
-  voiceMode?: "preset" | "custom" | "clone"
-  /** 选中的配音素材 ID */
-  selectedVoice?: string
-  /** 选中的克隆音色 ID */
-  selectedClonedVoice?: string
-  /** BGM 开关 */
-  bgm?: boolean
-  /** BGM 配置 */
-  bgmConfig?: { enabled: boolean; music_id?: string }
+  /** Step4 标题设置，用于封面叠加标题 */
+  titleSettings?: TitleSettings
+  /** 确认生成步骤产出的最终视频列表 */
+  generatedVideos: GeneratedVideo[]
 }
 
 const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
@@ -37,6 +24,7 @@ const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
     coverSettings,
     generating,
     generateAutoCover,
+    finalVideo,
     showCoverSettings,
     setShowCoverSettings,
     showCoverEditor,
@@ -53,17 +41,9 @@ const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
   } = useStep6Cover({
     coverSettings: props.coverSettings,
     onCoverSettingsChange: props.onCoverSettingsChange,
-    duration: props.duration,
-    assetIds: props.assetIds,
     selectedTemplate: props.selectedTemplate,
     titleSettings: props.titleSettings,
-    onPreviewTaskCreated: props.onPreviewTaskCreated,
-    onSourceEditPlanIdExtracted: props.onSourceEditPlanIdExtracted,
-    voiceMode: props.voiceMode,
-    selectedVoice: props.selectedVoice,
-    selectedClonedVoice: props.selectedClonedVoice,
-    bgm: props.bgm,
-    bgmConfig: props.bgmConfig,
+    generatedVideos: props.generatedVideos,
   })
 
   const handleAutoGenerate = () => {
@@ -77,8 +57,25 @@ const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
     <div className="xx-form-section">
       <h3>🖼️ 选择封面</h3>
 
+      {/* 最终成片信息 */}
+      {finalVideo && (
+        <div
+          style={{
+            padding: "10px 14px",
+            background: "rgba(16, 185, 129, 0.08)",
+            borderRadius: 8,
+            marginBottom: 16,
+            border: "1px solid rgba(16, 185, 129, 0.15)",
+            fontSize: 13,
+            color: "var(--text-secondary, #666)",
+          }}
+        >
+          🎬 封面将从最终成片「{finalVideo.name}」中智能选帧
+        </div>
+      )}
+
       <div className="xx-cover-actions">
-        <Button buttonType="primary" onClick={handleAutoGenerate}>
+        <Button buttonType="primary" onClick={handleAutoGenerate} disabled={!finalVideo}>
           ✨ 自动生成封面
         </Button>
         <Button buttonType="ghost" onClick={() => setShowCoverSettings(true)}>
@@ -126,7 +123,9 @@ const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
       <Modal open={generating} closable={false} footer={null} centered>
         <div style={{ textAlign: "center", padding: "24px 0" }}>
           <Spin size="large" />
-          <p style={{ marginTop: 16, fontSize: 14, color: "#666" }}>AI 正在生成封面，请稍候...</p>
+          <p style={{ marginTop: 16, fontSize: 14, color: "#666" }}>
+            AI 正在从最终成片选帧，请稍候...
+          </p>
         </div>
       </Modal>
     </div>

apps/web/src/pages/generate/components/Step7ConfirmGenerate.tsx

@@ -24,8 +24,6 @@ interface Step7ConfirmGenerateProps {
   presetVoices: PresetVoiceItem[]
   clonedVoices: VoiceClone[]
   coverSettings: CoverConfig
-  generateCount: number
-  onGenerateCountChange: (count: number) => void
   generating: boolean
   generated: boolean
   generateError: string | null
@@ -42,9 +40,6 @@ const Step7ConfirmGenerate: React.FC<Step7ConfirmGenerateProps> = (props) => {
     title,
     voiceName,
     coverSummary,
-    generateCount,
-    handleDecrement,
-    handleIncrement,
     generating,
     generated,
     generateError,
@@ -65,10 +60,6 @@ const Step7ConfirmGenerate: React.FC<Step7ConfirmGenerateProps> = (props) => {
         title={title}
         voiceName={voiceName}
         coverSummary={coverSummary}
-        generateCount={generateCount}
-        generating={generating}
-        onDecrement={handleDecrement}
-        onIncrement={handleIncrement}
       />
       <GenerationStatus
         generating={generating}

apps/web/src/pages/generate/components/material/ManualMaterialList.tsx

@@ -1,7 +1,8 @@
 /**
- * 手动选择素材列表
+ * 手动选择素材列表 — 竖屏 9:16 卡片网格
+ * 交互：默认显示封面，点击播放按钮播放，播放中隐藏按钮，点击视频区域暂停
  */
-import React, { useRef, useCallback } from "react"
+import React, { useRef, useState, useCallback } from "react"
 import { Typography } from "antd"
 import type { AssetItem } from "@/api/assets"
 
@@ -10,191 +11,274 @@ const { Text } = Typography
 interface ManualMaterialListProps {
   materials: { items: AssetItem[]; total: number }
   materialsLoading: boolean
+  /** 库内有素材但全部已用尽（usable === false），用于区分空状态文案 */
+  allExhausted?: boolean
   selectedMaterials: string[]
   onToggle: (materialId: string) => void
 }
 
+/** 秒数格式化为 mm:ss */
+const fmtDuration = (seconds?: number): string => {
+  if (!seconds && seconds !== 0) return "--:--"
+  const m = Math.floor(seconds / 60)
+  const s = Math.floor(seconds % 60)
+  return `${String(m).padStart(2, "0")}:${String(s).padStart(2, "0")}`
+}
+
+/** 单个素材卡片 */
+const MaterialCard: React.FC<{
+  asset: AssetItem
+  checked: boolean
+  onToggle: () => void
+}> = ({ asset, checked, onToggle }) => {
+  const videoRef = useRef<HTMLVideoElement>(null)
+  const [isPlaying, setIsPlaying] = useState(false)
+  const isVideo = asset.mime_type?.startsWith("video/") ?? false
+  const thumbSrc = asset.thumbnail_url || undefined
+
+  const handlePlayToggle = useCallback(
+    (e: React.MouseEvent) => {
+      e.stopPropagation()
+      const video = videoRef.current
+      if (!video || !isVideo) return
+
+      if (isPlaying) {
+        video.pause()
+        setIsPlaying(false)
+      } else {
+        video.play().catch(() => {})
+        setIsPlaying(true)
+      }
+    },
+    [isPlaying, isVideo],
+  )
+
+  const handleVideoEnded = useCallback(() => {
+    setIsPlaying(false)
+  }, [])
+
+  const handleCardClick = useCallback(() => {
+    // 如果视频正在播放，点击卡片空白区域暂停视频
+    if (isPlaying) {
+      const video = videoRef.current
+      if (video) {
+        video.pause()
+        setIsPlaying(false)
+      }
+      return
+    }
+    onToggle()
+  }, [isPlaying, onToggle])
+
+  return (
+    <div
+      data-testid="material-card"
+      onClick={handleCardClick}
+      style={{
+        position: "relative",
+        aspectRatio: "9 / 16",
+        borderRadius: 10,
+        overflow: "hidden",
+        cursor: "pointer",
+        border: checked ? "2px solid var(--primary-color, #4f46e5)" : "2px solid transparent",
+        boxShadow: checked ? "0 0 0 2px rgba(79, 70, 229, 0.2)" : "0 1px 3px rgba(0, 0, 0, 0.1)",
+        background: "#1e293b",
+        transition: "all 0.15s ease",
+      }}
+    >
+      {/* 视频元素 */}
+      {isVideo && asset.file_url ? (
+        <video
+          ref={videoRef}
+          src={asset.file_url}
+          poster={thumbSrc}
+          muted
+          loop
+          playsInline
+          preload="metadata"
+          onEnded={handleVideoEnded}
+          style={{
+            width: "100%",
+            height: "100%",
+            objectFit: "cover",
+            display: "block",
+          }}
+        />
+      ) : thumbSrc ? (
+        <img
+          src={thumbSrc}
+          alt={asset.name}
+          style={{
+            width: "100%",
+            height: "100%",
+            objectFit: "cover",
+            display: "block",
+          }}
+          onError={(e) => {
+            const target = e.target as HTMLImageElement
+            target.style.display = "none"
+          }}
+        />
+      ) : (
+        <div
+          style={{
+            width: "100%",
+            height: "100%",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            background: "linear-gradient(135deg, #334155, #1e293b)",
+            color: "rgba(255,255,255,0.5)",
+            fontSize: 28,
+          }}
+        >
+          {isVideo ? "🎬" : "🎵"}
+        </div>
+      )}
+
+      {/* 底部渐变遮罩 */}
+      <div
+        style={{
+          position: "absolute",
+          bottom: 0,
+          left: 0,
+          right: 0,
+          height: "50%",
+          background: "linear-gradient(0deg, rgba(0,0,0,0.6) 0%, transparent 100%)",
+          pointerEvents: "none",
+        }}
+      />
+
+      {/* 播放按钮 — 播放中隐藏 */}
+      {!isPlaying && (
+        <div
+          onClick={handlePlayToggle}
+          style={{
+            position: "absolute",
+            top: "50%",
+            left: "50%",
+            transform: "translate(-50%, -50%)",
+            width: 36,
+            height: 36,
+            borderRadius: "50%",
+            background: "rgba(99, 102, 241, 0.85)",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            zIndex: 3,
+            transition: "opacity 0.2s ease",
+          }}
+        >
+          <svg width="16" height="16" viewBox="0 0 24 24" fill="white">
+            <path d="M8 5v14l11-7z" />
+          </svg>
+        </div>
+      )}
+
+      {/* 文件名（左下角） */}
+      <div
+        style={{
+          position: "absolute",
+          bottom: 6,
+          left: 6,
+          right: 50,
+          color: "white",
+          fontSize: 11,
+          fontWeight: 500,
+          whiteSpace: "nowrap",
+          overflow: "hidden",
+          textOverflow: "ellipsis",
+          textShadow: "0 1px 2px rgba(0,0,0,0.5)",
+          pointerEvents: "none",
+          zIndex: 1,
+        }}
+      >
+        {asset.name}
+      </div>
+
+      {/* 时长（右下角） */}
+      <div
+        style={{
+          position: "absolute",
+          bottom: 6,
+          right: 6,
+          background: "rgba(0, 0, 0, 0.7)",
+          color: "white",
+          padding: "1px 5px",
+          borderRadius: 3,
+          fontSize: 10,
+          fontWeight: 600,
+          fontVariantNumeric: "tabular-nums",
+          pointerEvents: "none",
+          zIndex: 1,
+        }}
+      >
+        {fmtDuration(asset.duration)}
+      </div>
+
+      {/* 选中勾选标记（左上角） */}
+      {checked && (
+        <div
+          data-testid="material-card-check"
+          aria-label="已选中"
+          style={{
+            position: "absolute",
+            top: 6,
+            left: 6,
+            width: 20,
+            height: 20,
+            borderRadius: "50%",
+            background: "var(--primary-color, #4f46e5)",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            color: "white",
+            fontSize: 12,
+            fontWeight: 700,
+            zIndex: 2,
+            pointerEvents: "none",
+          }}
+        >
+          ✓
+        </div>
+      )}
+    </div>
+  )
+}
+
 const ManualMaterialList: React.FC<ManualMaterialListProps> = ({
   materials,
   materialsLoading,
+  allExhausted,
   selectedMaterials,
   onToggle,
 }) => {
-  // 追踪当前正在播放的视频元素，确保同时只有一个视频播放
-  const activeVideoRef = useRef<HTMLVideoElement | null>(null)
-
-  const handleVideoMouseEnter = useCallback((e: React.MouseEvent<HTMLVideoElement>) => {
-    const video = e.currentTarget
-    // 暂停之前正在播放的视频（检查是否仍在 DOM 中）
-    if (
-      activeVideoRef.current &&
-      activeVideoRef.current !== video &&
-      document.body.contains(activeVideoRef.current)
-    ) {
-      activeVideoRef.current.pause()
-      activeVideoRef.current.currentTime = 0
-    }
-    activeVideoRef.current = video
-    video.play().catch(() => {})
-  }, [])
-
-  const handleVideoMouseLeave = useCallback((e: React.MouseEvent<HTMLVideoElement>) => {
-    const video = e.currentTarget
-    video.pause()
-    video.currentTime = 0
-    if (activeVideoRef.current === video) {
-      activeVideoRef.current = null
-    }
-  }, [])
-
   return (
     <div style={{ marginTop: 14 }}>
       {materialsLoading ? (
         <Text style={{ color: "var(--text-secondary)", padding: "16px 0" }}>加载素材中…</Text>
       ) : materials.items.length === 0 ? (
         <Text style={{ color: "var(--text-secondary)", padding: "16px 0" }}>
-          暂无素材，请先在视频库中上传
+          {allExhausted
+            ? "暂无可选素材（素材可能已用尽，请先上传新素材）"
+            : "暂无素材，请先在视频库中上传"}
         </Text>
       ) : (
-        <div style={{ display: "flex", flexDirection: "column", gap: 6 }}>
-          {materials.items.map((m) => {
-            const checked = selectedMaterials.includes(m.id)
-            const isVideo = m.mime_type?.startsWith("video/") ?? false
-            const thumbSrc = m.thumbnail_url || undefined
-            return (
-              <label
-                key={m.id}
-                style={{
-                  display: "flex",
-                  alignItems: "center",
-                  gap: 10,
-                  padding: "8px 12px",
-                  background: checked ? "var(--primary-soft, #eef2ff)" : "#f8fafc",
-                  borderRadius: 10,
-                  cursor: "pointer",
-                  border: checked
-                    ? "1px solid var(--primary-color, #4f46e5)"
-                    : "1px solid transparent",
-                  transition: "all 0.15s ease",
-                }}
-              >
-                <input
-                  type="checkbox"
-                  checked={checked}
-                  onChange={() => onToggle(m.id)}
-                  style={{
-                    accentColor: "var(--primary-color, #4f46e5)",
-                    flexShrink: 0,
-                  }}
-                />
-                {/* 缩略图预览 48×48 */}
-                <div
-                  style={{
-                    width: 48,
-                    height: 48,
-                    borderRadius: 6,
-                    overflow: "hidden",
-                    background: "#e2e8f0",
-                    flexShrink: 0,
-                    display: "flex",
-                    alignItems: "center",
-                    justifyContent: "center",
-                  }}
-                >
-                  {isVideo && m.file_url ? (
-                    <video
-                      src={m.file_url}
-                      poster={m.thumbnail_url || undefined}
-                      muted
-                      loop
-                      playsInline
-                      preload="none"
-                      onMouseEnter={handleVideoMouseEnter}
-                      onMouseLeave={handleVideoMouseLeave}
-                      style={{
-                        width: "100%",
-                        height: "100%",
-                        objectFit: "cover",
-                        cursor: "pointer",
-                      }}
-                    />
-                  ) : thumbSrc ? (
-                    <img
-                      src={thumbSrc}
-                      alt={m.name}
-                      style={{
-                        width: "100%",
-                        height: "100%",
-                        objectFit: "cover",
-                      }}
-                      onError={(e) => {
-                        const target = e.target as HTMLImageElement
-                        target.style.display = "none"
-                        const fallback = target.nextElementSibling as HTMLElement | null
-                        if (fallback) fallback.style.display = "flex"
-                      }}
-                    />
-                  ) : null}
-                  {!thumbSrc && !isVideo && (
-                    <span
-                      style={{
-                        fontSize: 20,
-                        opacity: 0.5,
-                        display: "flex",
-                      }}
-                    >
-                      🎵
-                    </span>
-                  )}
-                  {!thumbSrc && isVideo && !m.file_url && (
-                    <span
-                      style={{
-                        fontSize: 20,
-                        opacity: 0.5,
-                        display: "flex",
-                      }}
-                    >
-                      🎬
-                    </span>
-                  )}
-                  {/* img onError 时显示的 fallback（初始隐藏） */}
-                  {thumbSrc && !(isVideo && m.file_url) && (
-                    <span
-                      style={{
-                        fontSize: 20,
-                        opacity: 0.5,
-                        display: "none",
-                      }}
-                    >
-                      {isVideo ? "🎬" : "🎵"}
-                    </span>
-                  )}
-                </div>
-                <span
-                  style={{
-                    fontSize: 13,
-                    color: "var(--text-primary)",
-                    flex: 1,
-                    minWidth: 0,
-                    overflow: "hidden",
-                    textOverflow: "ellipsis",
-                    whiteSpace: "nowrap",
-                  }}
-                >
-                  {m.name}
-                </span>
-                <span
-                  style={{
-                    fontSize: 11,
-                    color: "var(--text-tertiary, #94a3b8)",
-                    flexShrink: 0,
-                  }}
-                >
-                  {m.mime_type?.split("/")?.[1]?.toUpperCase() ?? "FILE"}
-                </span>
-              </label>
-            )
-          })}
+        <div
+          style={{
+            display: "grid",
+            gridTemplateColumns: "repeat(auto-fill, minmax(110px, 1fr))",
+            gap: 10,
+          }}
+        >
+          {materials.items.map((asset) => (
+            <MaterialCard
+              key={asset.id}
+              asset={asset}
+              checked={selectedMaterials.includes(asset.id)}
+              onToggle={() => onToggle(asset.id)}
+            />
+          ))}
         </div>
       )}
     </div>

apps/web/src/pages/generate/components/material/MaterialModeTabs.tsx

@@ -23,7 +23,7 @@ const MaterialModeTabs: React.FC<MaterialModeTabsProps> = ({ mode, onModeChange
         onClick={() => onModeChange("auto")}
         type="button"
       >
-        选择视频库自动匹配
+        AI智能匹配
       </button>
     </div>
   )