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

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

.gitea/workflows/ci-pipeline.yml

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

alembic/versions/073_add_lipsync_tts_fields.py

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

alembic/versions/074_ai_avatar_render_script_id_optional.py

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

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

@@ -17,7 +17,10 @@ from app.dependencies import get_db_session
 from app.schemas.ai_avatar_render import (
     AiAvatarRenderJobResponse,
     CreateAiAvatarRenderRequest,
+    SmartCoverRequest,
+    SmartCoverResponse,
 )
+from app.services.ai_avatar_cover_service import generate_smart_cover
 from app.services.ai_avatar_render_service import (
     AiAvatarRenderError,
     AiAvatarRenderService,
@@ -49,7 +52,7 @@ def create_render_job(
     """
     try:
         job = svc.create_render_job(
-            user_id=current_user.id,
+            user_id=current_user.user.id,
             lipsync_job_id=body.lipsync_job_id,
             script_id=body.script_id,
             b_roll_segments=[s.model_dump() for s in body.b_roll_segments],
@@ -94,7 +97,7 @@ def list_render_jobs(
 ):
     """获取 AI 数字人渲染任务列表."""
     items, total = svc.list_render_jobs(
-        user_id=current_user.id,
+        user_id=current_user.user.id,
         project_id=project_id,
         status=status,
         offset=offset,
@@ -118,7 +121,7 @@ def get_render_job(
     svc: AiAvatarRenderService = Depends(_get_service),
 ):
     """获取渲染任务详情."""
-    job = svc.get_render_job(job_id, current_user.id)
+    job = svc.get_render_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="渲染任务不存在")
     return job
@@ -134,7 +137,7 @@ def cancel_render_job(
     svc: AiAvatarRenderService = Depends(_get_service),
 ):
     """取消渲染任务（仅 pending 状态可取消）."""
-    job = svc.cancel_render_job(job_id, current_user.id)
+    job = svc.cancel_render_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="渲染任务不存在")
     if job.status != "cancelled":
@@ -155,7 +158,7 @@ def retry_render_job(
     svc: AiAvatarRenderService = Depends(_get_service),
 ):
     """重试失败的渲染任务."""
-    job = svc.retry_render_job(job_id, current_user.id)
+    job = svc.retry_render_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="渲染任务不存在")
     if job.status != "pending":
@@ -173,3 +176,34 @@ def retry_render_job(
         logger.warning("Celery 任务提交失败，重试任务已重置但未触发执行: %s", job.id)
 
     return job
+
+
+
+# ── POST /smart-cover — 智能获取封面（MediaKit 抽帧 + 评分选帧）────────
+
+
+@router.post("/smart-cover", response_model=SmartCoverResponse)
+def generate_avatar_smart_cover(
+    body: SmartCoverRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+) -> SmartCoverResponse:
+    """智能获取数字人视频封面.
+
+    复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧逻辑（非 FFmpeg 简单截帧），
+    并将选中帧转存到自家 OSS，返回非临时的封面公网 URL。
+
+    前端「智能获取封面」按钮可直接调用本接口；不依赖渲染任务完成。
+    """
+    video_url = (body.video_url or "").strip()
+    if not video_url.startswith(("http://", "https://")):
+        raise HTTPException(status_code=400, detail="video_url 必须是合法的 HTTP/HTTPS URL")
+
+    cover_url = generate_smart_cover(video_url, max_frames=body.max_frames)
+    if not cover_url:
+        return SmartCoverResponse(
+            cover_url="",
+            status="fallback_failed",
+            message="智能抽帧失败（MediaKit 不可用或抽帧异常），请稍后重试",
+        )
+    logger.info("智能封面生成成功: user=%s", current_user.user.id)
+    return SmartCoverResponse(cover_url=cover_url, status="completed")

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

@@ -13,11 +13,15 @@ from __future__ import annotations
 import logging
 
 from app.auth import AuthenticatedUser, get_current_user
-from app.dependencies import get_cosyvoice_service, get_db_session, get_voice_clone_profile_repository
+from app.dependencies import (
+    get_cosyvoice_service,
+    get_db_session,
+    get_voice_clone_profile_repository,
+)
 from app.schemas.lipsync import CreateLipsyncJobRequest, LipsyncJobResponse
 from app.services.lipsync_service import LipsyncService
 from app.services.mediakit_client import MediaKitError
-from fastapi import APIRouter, Depends, HTTPException, Query
+from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Query
 from sqlalchemy.orm import Session
 
 from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
@@ -29,35 +33,16 @@ router = APIRouter()
 
 def _get_service(
     db: Session = Depends(get_db_session),
+    voice_clone_repo=Depends(get_voice_clone_profile_repository),
     cosyvoice_service: CosyVoiceService = Depends(get_cosyvoice_service),
 ) -> LipsyncService:
-    return LipsyncService(db, cosyvoice_service=cosyvoice_service)
-
-
-def _resolve_voice_id(
-    raw_voice_id: str,
-    user_id: str,
-    voice_clone_repo,
-) -> str:
-    """解析 voice_id：支持预设音色 ID 或克隆音色 profile UUID.
-
-    与 TTS 路由保持一致：命中 profile → 校验归属 → 取 CosyVoice voice_id。
-    """
-    try:
-        profile = voice_clone_repo.get(raw_voice_id)
-    except Exception as exc:
-        logger.error("查询克隆音色失败: voice_id=%s, error=%s", raw_voice_id, exc)
-        raise HTTPException(
-            status_code=400,
-            detail=f"voice_id 无效: {raw_voice_id}",
-        ) from exc
-    if profile is not None:
-        if profile.user_id != user_id:
-            raise HTTPException(status_code=403, detail="无权访问该音色")
-        if not profile.voice_id:
-            raise HTTPException(status_code=400, detail="音色克隆尚未完成，请稍后再试")
-        return profile.voice_id
-    return raw_voice_id
+    # voice_clone_repo 用于克隆音色 profile 解析；cosyvoice_service 用于 TTS 直生
+    # （TTS 合成、音色解析、错误码归一化都在 LipsyncService 内部完成）
+    return LipsyncService(
+        db,
+        cosyvoice_service=cosyvoice_service,
+        voice_clone_repo=voice_clone_repo,
+    )
 
 
 # ── POST /jobs — 提交对口型任务 ───────────────────────────────────────────
@@ -68,22 +53,22 @@ def create_lipsync_job(
     body: CreateLipsyncJobRequest,
     current_user: AuthenticatedUser = Depends(get_current_user),
     svc: LipsyncService = Depends(_get_service),
-    voice_clone_repo=Depends(get_voice_clone_profile_repository),
 ):
     """提交对口型任务.
 
-    #1809: 前端传 {voice_id, script_text, video_url}，
-    后端内部调 TTS 合成音频，再提交 MediaKit。
+    #1809/#1822: 前端传 {video_url, voice_id, script_text, speed?, emotion?}，
+    后端内部解析音色、调 TTS 合成音频、转存 OSS，再提交 MediaKit；
+    也支持直接传 {video_url, audio_url}。
     """
-    # 解析 voice_id（支持克隆音色 profile UUID）
-    actual_voice_id = _resolve_voice_id(body.voice_id, current_user.id, voice_clone_repo)
-
     try:
         job = svc.create_job(
-            user_id=current_user.id,
+            user_id=current_user.user.id,
             video_url=body.video_url,
-            voice_id=actual_voice_id,
+            audio_url=body.audio_url,
+            voice_id=body.voice_id,
             script_text=body.script_text,
+            speed=body.speed,
+            emotion=body.emotion,
             enable_video_loop=body.enable_video_loop,
             project_id=body.project_id,
         )
@@ -97,12 +82,20 @@ def create_lipsync_job(
             detail={"code": "TTSSynthesisFailed", "message": str(exc)},
         ) from exc
     except MediaKitError as exc:
+        # TTS 合成失败 / 音色无权访问 → 400/403；MediaKit 提交失败 → 502
+        status_code = 502
+        if exc.code in ("VoiceForbidden",):
+            status_code = 403
+        elif exc.code in ("InvalidInput", "TTSInvalidParam", "VoiceNotReady"):
+            status_code = 400
+        elif exc.code == "TTSSynthesisFailed":
+            status_code = 502
         raise HTTPException(
-            status_code=502,
+            status_code=status_code,
             detail={
                 "code": exc.code,
                 "message": str(exc),
-                "request_id": exc.request_id,
+                "request_id": getattr(exc, "request_id", ""),
             },
         ) from exc
     except Exception as exc:
@@ -130,7 +123,7 @@ def list_lipsync_jobs(
 ):
     """获取对口型任务列表."""
     items, total = svc.list_jobs(
-        user_id=current_user.id,
+        user_id=current_user.user.id,
         project_id=project_id,
         status=status,
         offset=offset,
@@ -150,13 +143,22 @@ def list_lipsync_jobs(
 @router.get("/jobs/{job_id}", response_model=LipsyncJobResponse)
 def get_lipsync_job(
     job_id: str,
+    background: BackgroundTasks,
     current_user: AuthenticatedUser = Depends(get_current_user),
     svc: LipsyncService = Depends(_get_service),
 ):
-    """获取对口型任务详情."""
-    job = svc.get_job(job_id, current_user.id)
+    """获取对口型任务详情.
+
+    非终态任务：先返回 DB 缓存，挂后台刷新（下次轮询拿到新状态），
+    避免 MediaKit 慢响应阻塞前端轮询。
+    """
+    job = svc.get_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="任务不存在")
+
+    if job.status not in ("completed", "failed"):
+        background.add_task(svc.refresh_job_status, job_id, current_user.user.id)
+
     return job
 
 
@@ -170,7 +172,7 @@ def refresh_lipsync_job(
     svc: LipsyncService = Depends(_get_service),
 ):
     """从 MediaKit 拉取最新状态并更新."""
-    job = svc.refresh_job_status(job_id, current_user.id)
+    job = svc.refresh_job_status(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="任务不存在")
     return job
@@ -186,7 +188,7 @@ def cancel_lipsync_job(
     svc: LipsyncService = Depends(_get_service),
 ):
     """取消对口型任务（仅 pending/submitted 状态可取消）."""
-    job = svc.cancel_job(job_id, current_user.id)
+    job = svc.cancel_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="任务不存在")
     if job.status != "cancelled":

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

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

apps/api/app/schemas/ai_avatar_render.py

@@ -50,7 +50,7 @@ class CreateAiAvatarRenderRequest(BaseModel):
     """创建渲染任务请求."""
 
     lipsync_job_id: str = Field(..., description="对口型任务 ID")
-    script_id: str = Field(..., description="文案 ID")
+    script_id: str = Field("", description="文案 ID（选自文案库时传；手动输入文案直生场景可留空）")
     b_roll_segments: list[BRollSegment] = Field(default_factory=list, description="B-roll 片段列表")
     title_config: dict[str, Any] = Field(default_factory=dict, description="标题配置")
     cover_config: dict[str, Any] = Field(default_factory=dict, description="封面配置")
@@ -67,10 +67,8 @@ class CreateAiAvatarRenderRequest(BaseModel):
     @field_validator("script_id")
     @classmethod
     def validate_script_id(cls, v: str) -> str:
-        v = v.strip()
-        if not v:
-            raise ValueError("script_id 不能为空")
-        return v
+        # script_id 可选：手动输入文案（TTS 直生）场景不关联文案库条目
+        return (v or "").strip()
 
 
 class AiAvatarRenderJobResponse(BaseModel):
@@ -80,7 +78,7 @@ class AiAvatarRenderJobResponse(BaseModel):
     user_id: str
     project_id: str
     lipsync_job_id: str
-    script_id: str
+    script_id: str = ""
     b_roll_segments: list[dict[str, Any]]
     title_config: dict[str, Any]
     cover_config: dict[str, Any]
@@ -109,3 +107,18 @@ class AiAvatarRenderProgressResponse(BaseModel):
     output_cover_url: str
     output_duration: float
     error_message: str
+
+
+class SmartCoverRequest(BaseModel):
+    """智能封面请求 — MediaKit 抽帧 + 质量评分选最佳帧."""
+
+    video_url: str = Field(..., description="数字人视频 URL（对口型/渲染成片）")
+    max_frames: int = Field(5, ge=1, le=10, description="抽帧数量（默认 5）")
+
+
+class SmartCoverResponse(BaseModel):
+    """智能封面响应."""
+
+    cover_url: str = Field("", description="封面图公网 URL（OSS，非临时）；失败为空")
+    status: str = Field("completed", description="completed / fallback_failed")
+    message: str = Field("", description="失败原因（如有）")

apps/api/app/schemas/lipsync.py

@@ -1,11 +1,17 @@
-"""对口型 API Schema 定义 — #1796, #1809 参数调整."""
+"""对口型 API Schema 定义 — #1796 / #1809 / #1822.
+
+支持两种输入模式（二选一）：
+1. TTS 直生模式（推荐）：传 voice_id + script_text（+ speed/emotion），
+   后端内部先调 CosyVoice 合成音频，再提交 MediaKit 对口型。
+2. 直接音频模式：传 video_url + audio_url（音频已由调用方准备好）。
+"""
 
 from __future__ import annotations
 
 from datetime import datetime
 from typing import Optional
 
-from pydantic import BaseModel, Field, field_validator
+from pydantic import BaseModel, Field, model_validator
 
 
 class LipsyncJobResponse(BaseModel):
@@ -17,6 +23,10 @@ class LipsyncJobResponse(BaseModel):
     video_url: str
     audio_url: str
     enable_video_loop: bool
+    voice_id: str = ""
+    script_text: str = ""
+    speed: float = 1.0
+    emotion: str = ""
     mediakit_task_id: str
     status: str
     output_video_url: str
@@ -33,45 +43,58 @@ class LipsyncJobResponse(BaseModel):
 
 
 class CreateLipsyncJobRequest(BaseModel):
-    """创建对口型任务请求 — #1809.
+    """创建对口型任务请求.
 
-    前端传 {voice_id, script_text, video_url}，
-    后端内部调 TTS 生成 audio_url 再提交 MediaKit。
+    两种模式（二选一）：
+    - TTS 直生：voice_id + script_text 必填（+ 可选 speed/emotion）；audio_url 留空。
+    - 直接音频：video_url + audio_url 必填。
     """
 
     video_url: str = Field(..., description="人物视频 URL（MP4，≤30min，单人真人）")
-    voice_id: str = Field(..., description="音色 ID（预设音色或克隆音色 profile ID）")
-    script_text: str = Field(..., description="要合成的脚本文本")
+
+    # 模式 2：直接音频
+    audio_url: str = Field("", description="驱动音频 URL（mp3/aac/wav/m4a/flac）；直生模式留空")
+
+    # 模式 1：TTS 直生
+    voice_id: str = Field("", description="音色 ID（预置音色或克隆音色 profile UUID）")
+    script_text: str = Field("", description="要合成的文案（直生模式必填，最长 5000 字符）")
+    speed: float = Field(1.0, ge=0.5, le=2.0, description="语速（0.5-2.0），默认 1.0")
+    emotion: str = Field("", description="情绪（natural/excited/calm/friendly 或中文 自然/兴奋/沉稳/亲切）")
+
     enable_video_loop: bool = Field(False, description="音频长于视频时是否循环画面")
     project_id: str = Field("", description="项目 ID（可选）")
 
-    @field_validator("video_url")
-    @classmethod
-    def validate_video_url(cls, v: str) -> str:
-        v = v.strip()
-        if not v:
+    @model_validator(mode="after")
+    def _validate_input_mode(self) -> "CreateLipsyncJobRequest":
+        video = (self.video_url or "").strip()
+        if not video:
             raise ValueError("video_url 不能为空")
-        if not v.startswith(("http://", "https://")):
+        if not video.startswith(("http://", "https://")):
             raise ValueError("video_url 必须是 HTTP/HTTPS URL")
-        lower = v.lower().split("?")[0]
+        lower = video.lower().split("?")[0]
         if not lower.endswith(".mp4"):
             raise ValueError("video_url 仅支持 MP4 格式")
-        return v
 
-    @field_validator("voice_id")
-    @classmethod
-    def validate_voice_id(cls, v: str) -> str:
-        v = v.strip()
-        if not v:
-            raise ValueError("voice_id 不能为空")
-        return v
+        has_audio = bool((self.audio_url or "").strip())
+        has_tts = bool((self.voice_id or "").strip()) and bool((self.script_text or "").strip())
 
-    @field_validator("script_text")
-    @classmethod
-    def validate_script_text(cls, v: str) -> str:
-        v = v.strip()
-        if not v:
-            raise ValueError("script_text 不能为空")
-        if len(v) > 5000:
+        if not has_audio and not has_tts:
+            raise ValueError(
+                "必须提供驱动音频：要么传 audio_url（直接音频模式），"
+                "要么同时传 voice_id + script_text（TTS 直生模式）"
+            )
+
+        if has_tts and len(self.script_text) > 5000:
             raise ValueError("script_text 最长 5000 字符")
-        return v
+
+        if has_audio:
+            au = self.audio_url.strip()
+            if not au.startswith(("http://", "https://")):
+                raise ValueError("audio_url 必须是 HTTP/HTTPS URL")
+            au_lower = au.lower().split("?")[0]
+            allowed = (".mp3", ".aac", ".wav", ".m4a", ".flac")
+            if not any(au_lower.endswith(ext) for ext in allowed):
+                raise ValueError(f"audio_url 格式不支持，仅支持: {', '.join(allowed)}")
+            self.audio_url = au
+
+        return self

apps/api/app/schemas/tts.py

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

apps/api/app/services/ai_avatar_cover_service.py

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

apps/api/app/services/ai_avatar_render_service.py

@@ -53,8 +53,8 @@ class AiAvatarRenderService:
         *,
         user_id: str,
         lipsync_job_id: str,
-        script_id: str,
-        b_roll_segments: list[dict[str, Any]],
+        script_id: str = "",
+        b_roll_segments: list[dict[str, Any]] | None = None,
         title_config: dict[str, Any],
         cover_config: dict[str, Any],
         project_id: str = "",
@@ -83,17 +83,19 @@ class AiAvatarRenderService:
         if not lipsync_job.output_video_url:
             raise AiAvatarRenderError("对口型任务输出视频 URL 为空", code="LipsyncJobNoOutput")
 
-        # 2. 验证文案归属
-        script = (
-            self.db.query(ScriptModel)
-            .filter(
-                ScriptModel.id == script_id,
-                ScriptModel.user_id == user_id,
+        # 2. 验证文案归属（仅当选了文案库条目时；手动输入文案直生场景 script_id 可空）
+        script_id = (script_id or "").strip()
+        if script_id:
+            script = (
+                self.db.query(ScriptModel)
+                .filter(
+                    ScriptModel.id == script_id,
+                    ScriptModel.user_id == user_id,
+                )
+                .first()
             )
-            .first()
-        )
-        if script is None:
-            raise AiAvatarRenderError("文案不存在或无权访问", code="ScriptNotFound")
+            if script is None:
+                raise AiAvatarRenderError("文案不存在或无权访问", code="ScriptNotFound")
 
         # 3. 创建渲染任务
         job_id = str(uuid.uuid4())
@@ -103,7 +105,7 @@ class AiAvatarRenderService:
             project_id=project_id,
             lipsync_job_id=lipsync_job_id,
             script_id=script_id,
-            b_roll_segments=[s if isinstance(s, dict) else s.model_dump() for s in b_roll_segments],
+            b_roll_segments=[s if isinstance(s, dict) else s.model_dump() for s in (b_roll_segments or [])],
             title_config=title_config,
             cover_config=cover_config,
             status="pending",
@@ -288,7 +290,22 @@ class AiAvatarRenderService:
                 output_video_url = self._upload_to_oss(output_video_path, f"ai-avatar/{job_id}/output.mp4")
                 job.output_video_url = output_video_url
 
-                if cover_path:
+                # 封面：优先复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧；
+                # MediaKit 不可用时回退到 FFmpeg 已按 cover_config 抽取的 cover_path
+                smart_cover_url = ""
+                if output_video_url:
+                    try:
+                        from app.services.ai_avatar_cover_service import (
+                            generate_smart_cover,
+                        )
+
+                        smart_cover_url = generate_smart_cover(output_video_url, job_id=job_id, max_frames=5)
+                    except Exception:
+                        logger.warning("智能封面(MediaKit)失败，回退 FFmpeg 封面 job_id=%s", job_id, exc_info=True)
+
+                if smart_cover_url:
+                    job.output_cover_url = smart_cover_url
+                elif cover_path:
                     output_cover_url = self._upload_to_oss(cover_path, f"ai-avatar/{job_id}/cover.jpg")
                     job.output_cover_url = output_cover_url
 

apps/api/app/services/lipsync_service.py

@@ -1,15 +1,16 @@
 """对口型 Service — #1796 MediaKit 对口型业务逻辑, #1809 参数调整.
 
 职责：
-- 创建/查询/取消对口型任务
-- 调用 TTS 合成音频（#1809：前端不再传 audio_url）
+- 创建/查询对口型任务
+- 双输入模式：TTS 直生（voice_id + script_text，内部先合成音频转存 OSS）或直接音频（audio_url）
 - 调用 MediaKit 客户端提交异步任务
-- 轮询更新任务状态
+- 轮询更新任务状态（中间状态同步 DB，成片转存自家 OSS）
 - 用户隔离（每个用户只能操作自己的任务）
 """
 
 from __future__ import annotations
 
+import io
 import logging
 import uuid
 from datetime import datetime, timezone
@@ -26,7 +27,9 @@ from app.services.mediakit_client import (
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
-from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
+from packages.application.cosyvoice_service import CosyVoiceError, normalize_emotion
+from packages.shared.storage import get_shared_storage_service
+from packages.shared.url_security import ALLOWED_AUDIO_MIME_TYPES, safe_download_bytes
 
 logger = logging.getLogger(__name__)
 
@@ -38,19 +41,98 @@ class LipsyncService:
         self,
         db: Session,
         client: Optional[MediaKitClient] = None,
-        cosyvoice_service: Optional[CosyVoiceService] = None,
+        cosyvoice_service=None,
+        voice_clone_repo=None,
     ):
         self.db = db
         self.client = client or get_mediakit_client()
-        self._cosyvoice_service = cosyvoice_service
+        self._cosyvoice = cosyvoice_service
+        self._voice_clone_repo = voice_clone_repo
 
-    @property
-    def cosyvoice_service(self) -> CosyVoiceService:
-        if self._cosyvoice_service is None:
+    def _get_cosyvoice(self):
+        """延迟获取 CosyVoiceService（与 tts 路由一致，含 OSS 预签名配置）."""
+        if self._cosyvoice is None:
             from app.dependencies import get_cosyvoice_service
 
-            self._cosyvoice_service = get_cosyvoice_service()
-        return self._cosyvoice_service
+            self._cosyvoice = get_cosyvoice_service()
+        return self._cosyvoice
+
+    def _resolve_voice_id(self, voice_id: str, user_id: str) -> str:
+        """将克隆音色 profile UUID 解析为 CosyVoice voice_id。
+
+        与 /tts/synthesize 保持一致：命中 profile → 校验归属 → 返回其 voice_id；
+        未命中（预置音色 ID 或克隆 CosyVoice voice_id）原样返回。
+        """
+        if not voice_id:
+            return ""
+        if self._voice_clone_repo is None:
+            try:
+                from app.dependencies import get_voice_clone_profile_repository
+
+                self._voice_clone_repo = get_voice_clone_profile_repository(self.db)
+            except Exception:
+                return voice_id
+        try:
+            profile = self._voice_clone_repo.get(voice_id)
+        except Exception:
+            return voice_id
+        if profile is None:
+            return voice_id
+        if getattr(profile, "user_id", "") != user_id:
+            raise MediaKitError("无权访问该音色", code="VoiceForbidden")
+        if not getattr(profile, "voice_id", ""):
+            raise MediaKitError("音色克隆尚未完成，请稍后再试", code="VoiceNotReady")
+        return profile.voice_id
+
+    def _synthesize_and_persist_audio(
+        self,
+        *,
+        user_id: str,
+        job_id: str,
+        voice_id: str,
+        script_text: str,
+        speed: float,
+        emotion: str,
+    ) -> str:
+        """TTS 直生：调 CosyVoice 合成音频并转存 OSS，返回可公网访问的音频 URL.
+
+        Raises:
+            MediaKitError: 合成失败
+        """
+        actual_voice_id = self._resolve_voice_id(voice_id, user_id)
+        cosyvoice = self._get_cosyvoice()
+        try:
+            result = cosyvoice.submit_synthesize_task(
+                text=script_text,
+                voice_id=actual_voice_id,
+                speed=speed,
+                emotion=normalize_emotion(emotion),
+            )
+        except CosyVoiceError as exc:
+            raise MediaKitError(f"TTS 合成失败: {exc}", code="TTSSynthesisFailed") from exc
+        except ValueError as exc:
+            raise MediaKitError(f"TTS 参数错误: {exc}", code="TTSInvalidParam") from exc
+
+        temp_url = result.get("audio_url", "")
+        if not temp_url:
+            raise MediaKitError("TTS 未返回音频 URL", code="TTSNoAudio")
+
+        # 转存到自家 OSS，避免临时 URL 过期导致 MediaKit 拉取失败
+        try:
+            audio_data = safe_download_bytes(
+                temp_url,
+                purpose="lipsync_tts_audio",
+                allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
+                timeout=60.0,
+            )
+            storage = get_shared_storage_service()
+            storage_key = f"lipsync-tts/{user_id}/{job_id}.mp3"
+            permanent_url = storage.upload_file(io.BytesIO(audio_data), storage_key, content_type="audio/mpeg")
+            logger.info("对口型 TTS 音频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
+            return permanent_url
+        except Exception as exc:
+            logger.warning("TTS 音频转存 OSS 失败，回退临时 URL: job_id=%s err=%s", job_id, exc)
+            return temp_url
 
     # ── 创建任务 ──────────────────────────────────────────────────────────
 
@@ -59,47 +141,42 @@ class LipsyncService:
         *,
         user_id: str,
         video_url: str,
-        voice_id: str,
-        script_text: str,
+        audio_url: str = "",
+        voice_id: str = "",
+        script_text: str = "",
+        speed: float = 1.0,
+        emotion: str = "",
         enable_video_loop: bool = False,
         project_id: str = "",
     ) -> LipsyncJobModel:
         """创建对口型任务并提交到 MediaKit.
 
-        #1809: 内部调 TTS 合成音频，不再由前端传 audio_url。
+        两种输入模式：
+        - TTS 直生：voice_id + script_text（audio_url 留空），后端先合成音频
+        - 直接音频：提供 audio_url
 
         Raises:
-            CosyVoiceError: TTS 合成失败
-            MediaKitError: API 调用失败
+            MediaKitError: TTS 合成或 MediaKit 提交失败
         """
-        # 1. 调 TTS 合成音频
-        try:
-            tts_result = self.cosyvoice_service.synthesize_speech(
-                text=script_text,
-                voice_id=voice_id,
-            )
-            audio_url = tts_result.audio_url
-        except CosyVoiceError as exc:
-            logger.error("TTS 合成失败: voice_id=%s, error=%s", voice_id, exc)
-            # 创建失败记录
-            job_id = str(uuid.uuid4())
-            job = LipsyncJobModel(
-                id=job_id,
+        # 0. TTS 直生模式：先合成音频（在创建 DB 记录之前完成，失败直接抛出）
+        if not audio_url:
+            if not (voice_id and script_text):
+                raise MediaKitError(
+                    "必须提供 audio_url 或 voice_id+script_text",
+                    code="InvalidInput",
+                )
+            # 预合成：用临时 job_id 命名 OSS 对象
+            pre_job_id = str(uuid.uuid4())
+            audio_url = self._synthesize_and_persist_audio(
                 user_id=user_id,
-                project_id=project_id,
-                video_url=video_url,
-                audio_url="",
-                enable_video_loop=enable_video_loop,
-                status="failed",
-                error_message=f"TTS 合成失败: {exc}",
-                error_code="TTSSynthesisFailed",
+                job_id=pre_job_id,
+                voice_id=voice_id,
+                script_text=script_text,
+                speed=speed,
+                emotion=emotion,
             )
-            self.db.add(job)
-            self.db.commit()
-            self.db.refresh(job)
-            raise
 
-        # 2. 创建数据库记录
+        # 1. 创建数据库记录
         job_id = str(uuid.uuid4())
         job = LipsyncJobModel(
             id=job_id,
@@ -108,6 +185,10 @@ class LipsyncService:
             video_url=video_url,
             audio_url=audio_url,
             enable_video_loop=enable_video_loop,
+            voice_id=voice_id or "",
+            script_text=script_text or "",
+            speed=speed,
+            emotion=normalize_emotion(emotion),
             status="pending",
         )
         self.db.add(job)
@@ -192,11 +273,14 @@ class LipsyncService:
             return job
 
         mk_status = status_data.get("status", STATUS_RUNNING)
+        logger.info("MediaKit 对口型状态 [%s]: %s", job_id, mk_status)
 
         if mk_status == STATUS_COMPLETED:
             result = status_data.get("result", {})
             job.status = STATUS_COMPLETED
-            job.output_video_url = result.get("video_url", "")
+            output_url = result.get("video_url", "")
+            # MediaKit 输出为临时 URL，转存自家 OSS 防止过期（失败则回退临时 URL）
+            job.output_video_url = self._persist_output_video(output_url, job_id, user_id)
             job.output_duration = result.get("duration", 0.0)
             job.completed_at = datetime.now(timezone.utc)
         elif mk_status == STATUS_FAILED:
@@ -205,12 +289,38 @@ class LipsyncService:
             job.error_message = error.get("message", "任务执行失败")
             job.error_code = error.get("code", "TaskFailed")
             job.completed_at = datetime.now(timezone.utc)
-        # running 状态只更新时间戳
+        else:
+            # 中间状态（running/processing/queued 等）同步到 DB，避免前端永远卡在 submitted
+            if isinstance(mk_status, str) and mk_status:
+                job.status = mk_status
         job.updated_at = datetime.now(timezone.utc)
         self.db.commit()
         self.db.refresh(job)
         return job
 
+    def _persist_output_video(self, temp_url: str, job_id: str, user_id: str) -> str:
+        """将 MediaKit 输出的临时视频 URL 转存到自家 OSS.
+
+        失败时回退返回原始临时 URL，不影响任务完成。
+        """
+        if not temp_url:
+            return ""
+        try:
+            import httpx
+
+            with httpx.Client(timeout=180.0, follow_redirects=True) as client:
+                resp = client.get(temp_url)
+                resp.raise_for_status()
+                data = resp.content
+            storage = get_shared_storage_service()
+            storage_key = f"lipsync-outputs/{user_id}/{job_id}.mp4"
+            permanent_url = storage.upload_file(io.BytesIO(data), storage_key, content_type="video/mp4")
+            logger.info("对口型输出视频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
+            return permanent_url or temp_url
+        except Exception as exc:
+            logger.warning("对口型输出视频转存 OSS 失败，回退临时 URL: job_id=%s err=%s", job_id, exc)
+            return temp_url
+
     # ── 取消任务 ──────────────────────────────────────────────────────────
 
     def cancel_job(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:

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

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

apps/web/src/pages/ai-avatar/AiAvatar.css

@@ -1150,3 +1150,29 @@
   flex-direction: column;
   gap: 4px;
 }
+
+/* ─ 对口型生成弹窗 Spinner ── */
+.aa-lipsync-spinner {
+  width: 48px;
+  height: 48px;
+  border: 4px solid #f0f0f5;
+  border-top-color: #6366f1;
+  border-radius: 50%;
+  animation: aa-spin 0.8s linear infinite;
+}
+
+@keyframes aa-spin {
+  to {
+    transform: rotate(360deg);
+  }
+}
+
+.aa-btn--danger {
+  background: #ff4d4f;
+  color: #fff;
+  border: none;
+}
+
+.aa-btn--danger:hover {
+  background: #ff7875;
+}

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

@@ -19,7 +19,13 @@ import {
   createLipsyncJob,
   getLipsyncJob,
   submitRender,
+  generateSmartCover,
 } from "./api/aiAvatar"
+import {
+  normalizeEmotion,
+  buildTitleConfigPayload,
+  buildCoverConfigPayload,
+} from "./utils/contract"
 
 /** 面板折叠状态 */
 type PanelKey = "video" | "voice" | "script" | "title" | "cover"
@@ -34,6 +40,15 @@ const AiAvatarPage: React.FC = () => {
     cover: false,
   })
 
+  /* ── 对口型生成弹窗 ── */
+  const [showLipsyncModal, setShowLipsyncModal] = useState(false)
+  const [lipsyncStatus, setLipsyncStatus] = useState<"generating" | "completed" | "failed">(
+    "generating",
+  )
+  const [lipsyncErrorMessage, setLipsyncErrorMessage] = useState("")
+  /* ── 智能封面加载态 ── */
+  const [smartCoverLoading, setSmartCoverLoading] = useState(false)
+
   /* ── 对口型轮询 ── */
   const lipsyncTimerRef = useRef<ReturnType<typeof setInterval> | null>(null)
 
@@ -56,46 +71,91 @@ const AiAvatarPage: React.FC = () => {
       return
     }
     try {
+      // 显示生成弹窗
+      setShowLipsyncModal(true)
+      setLipsyncStatus("generating")
+      setLipsyncErrorMessage("")
+
       // ① 先按素材 id 拿 file_url（#1809 补充：对齐后端新参数 video_url）
+      console.log("[对口型] 开始生成:", {
+        videoId: video.id,
+        voiceId: voice.voice_id,
+        voiceType: voice.type,
+        textLen: state.scriptText.length,
+      })
       const asset = await getAssetById(video.id)
+      console.log("[对口型] getAssetById 响应:", {
+        id: asset?.id,
+        file_url: asset?.file_url?.substring(0, 100),
+      })
       const videoUrl = asset?.file_url
       if (!videoUrl) {
+        console.error("[对口型] file_url 为空，asset:", asset)
+        setShowLipsyncModal(false)
         message.error("获取出镜视频播放地址失败，请重新选择素材")
         return
       }
-      // ② voice_id（预设/克隆 UUID 均由后端内部调 TTS）+ script_text + video_url
-      const job = await createLipsyncJob({
+      // ② 模式A TTS直生：video_url + voice_id + script_text，语速/情绪英文枚举透传（#1822）
+      const payload = {
         voice_id: voice.voice_id,
         script_text: state.scriptText,
         video_url: videoUrl,
-      })
+        speed: state.speed, // 语速 0.5~2.0
+        emotion: normalizeEmotion(state.emotion), // natural/excited/calm/friendly
+      }
+      console.log("[对口型] createLipsyncJob 请求:", payload)
+      const job = await createLipsyncJob(payload)
+      console.log("[对口型] createLipsyncJob 响应:", { id: job.id, status: job.status })
       state.setLipsyncJob(job)
-      message.success("对口型任务已提交，生成中…")
       // 开始轮询
       if (lipsyncTimerRef.current) clearInterval(lipsyncTimerRef.current)
       lipsyncTimerRef.current = setInterval(async () => {
         try {
           const updated = await getLipsyncJob(job.id)
           state.setLipsyncJob(updated)
-          if (updated.status === "completed" || updated.status === "failed") {
+          console.log("[对口型] 轮询状态:", {
+            id: updated.id,
+            status: updated.status,
+            error: updated.error_message,
+          })
+          if (updated.status === "completed") {
             if (lipsyncTimerRef.current) clearInterval(lipsyncTimerRef.current)
-            if (updated.status === "completed") {
+            setLipsyncStatus("completed")
+            setTimeout(() => {
+              setShowLipsyncModal(false)
               message.success("对口型视频生成完成")
-            } else {
-              message.error(updated.error_message || "对口型生成失败")
-            }
+            }, 1000)
+          } else if (updated.status === "failed") {
+            if (lipsyncTimerRef.current) clearInterval(lipsyncTimerRef.current)
+            setLipsyncStatus("failed")
+            setLipsyncErrorMessage(updated.error_message || "对口型生成失败")
           }
-        } catch {
-          // 忽略轮询错误（轮询期间不打扰用户）
+        } catch (err) {
+          console.error("[对口型] 轮询错误:", err)
         }
       }, 3000)
     } catch (err) {
-      // ② 接口失败弹错误提示，不只 console
-      console.error("对口型任务创建失败:", err)
+      console.error("[对口型] 创建失败:", {
+        status: (err as { response?: { status?: number } })?.response?.status,
+        data: (err as { response?: { data?: unknown } })?.response?.data,
+        message: err instanceof Error ? err.message : String(err),
+      })
+      setShowLipsyncModal(false)
       message.error(err instanceof Error ? err.message : "对口型任务提交失败，请重试")
     }
     // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [state.selectedVideo, state.selectedVoice, state.scriptText])
+  }, [state.selectedVideo, state.selectedVoice, state.scriptText, state.speed, state.emotion])
+
+  // 取消对口型生成
+  const handleCancelLipsync = useCallback(() => {
+    if (lipsyncTimerRef.current) {
+      clearInterval(lipsyncTimerRef.current)
+      lipsyncTimerRef.current = null
+    }
+    setShowLipsyncModal(false)
+    setLipsyncStatus("generating")
+    setLipsyncErrorMessage("")
+  }, [])
 
   // 清理轮询
   useEffect(() => {
@@ -117,8 +177,10 @@ const AiAvatarPage: React.FC = () => {
         lipsync_job_id: state.lipsyncJob.id,
         script_id: state.script?.id,
         b_roll_segments: state.bRollSegments as never,
-        title_config: state.titleConfig as unknown as Record<string, unknown>,
-        cover_config: state.coverConfig as unknown as Record<string, unknown>,
+        // 字段映射：build_title_drawtext_filter 真实口径 text/font_size/font_color/position/...
+        title_config: buildTitleConfigPayload(state.titleConfig),
+        // cover_config：智能封面 cover_url + 截帧 timestamp
+        cover_config: buildCoverConfigPayload(state.coverConfig, state.coverConfig.smart_cover_url),
         resolution: state.resolution,
       })
       message.success("渲染任务已提交，可在视频管理中查看进度")
@@ -138,6 +200,37 @@ const AiAvatarPage: React.FC = () => {
     state.resolution,
   ])
 
+  /* ── 智能封面：调后端 MediaKit 选帧接口（#1822） ── */
+  const handleSmartCover = useCallback(async () => {
+    // 基于对口型成片抽帧，必须先完成对口型
+    const videoUrl = state.lipsyncJob?.output_video_url
+    if (state.lipsyncJob?.status !== "completed" || !videoUrl) {
+      message.warning("请先生成对口型视频，完成后再智能获取封面")
+      return
+    }
+    setSmartCoverLoading(true)
+    try {
+      const res = await generateSmartCover(videoUrl, 5)
+      if (res.cover_url) {
+        state.setCoverConfig((prev) => ({
+          ...prev,
+          mode: "auto_frame",
+          smart_cover_url: res.cover_url,
+          thumbnail_url: res.cover_url,
+        }))
+        message.success("智能封面已生成")
+      } else {
+        message.error(res.message || "智能封面生成失败，请稍后重试")
+      }
+    } catch (err) {
+      console.error("智能封面生成失败:", err)
+      message.error(err instanceof Error ? err.message : "智能封面生成失败，请重试")
+    } finally {
+      setSmartCoverLoading(false)
+    }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [state.lipsyncJob])
+
   /* ── 配置汇总 ── */
   const summary = {
     videoName: state.selectedVideo?.name || null,
@@ -166,6 +259,7 @@ const AiAvatarPage: React.FC = () => {
               selectedVideo={state.selectedVideo}
               onSelectVideo={() => state.setShowAssetPicker(true)}
               onRemoveVideo={state.removeVideo}
+              titleConfig={state.titleConfig}
             />
           </div>
         </div>
@@ -235,6 +329,9 @@ const AiAvatarPage: React.FC = () => {
               onCoverConfigChange={(partial) =>
                 state.setCoverConfig((prev) => ({ ...prev, ...partial }))
               }
+              onSmartCover={handleSmartCover}
+              smartCoverLoading={smartCoverLoading}
+              canSmartCover={state.lipsyncJob?.status === "completed"}
               resolution={state.resolution}
               onResolutionChange={state.setResolution}
               isGenerating={state.isGenerating}
@@ -276,6 +373,74 @@ const AiAvatarPage: React.FC = () => {
           onRemove={state.removeBRollSegment}
         />
       )}
+
+      {/* 对口型生成弹窗 */}
+      {showLipsyncModal && (
+        <div className="aa-modal-overlay">
+          <div className="aa-modal" onClick={(e) => e.stopPropagation()}>
+            <div className="aa-modal__header">
+              <span className="aa-modal__title">对口型生成</span>
+              <button className="aa-modal__close" onClick={handleCancelLipsync}>
+                ✕
+              </button>
+            </div>
+            <div
+              className="aa-modal__body"
+              style={{
+                display: "flex",
+                flexDirection: "column",
+                alignItems: "center",
+                padding: "40px 20px",
+              }}
+            >
+              {lipsyncStatus === "generating" && (
+                <>
+                  <div className="aa-lipsync-spinner" />
+                  <div style={{ marginTop: 20, fontSize: 15, color: "#1a1a2e" }}>
+                    对口型视频生成中…
+                  </div>
+                  <div style={{ marginTop: 8, fontSize: 13, color: "#8c8ca1" }}>
+                    请勿关闭页面，完成后将自动提示
+                  </div>
+                </>
+              )}
+              {lipsyncStatus === "completed" && (
+                <>
+                  <div style={{ fontSize: 48 }}>✅</div>
+                  <div style={{ marginTop: 16, fontSize: 15, color: "#1a1a2e" }}>
+                    对口型视频生成完成
+                  </div>
+                </>
+              )}
+              {lipsyncStatus === "failed" && (
+                <>
+                  <div style={{ fontSize: 48 }}>❌</div>
+                  <div style={{ marginTop: 16, fontSize: 15, color: "#1a1a2e" }}>
+                    对口型生成失败
+                  </div>
+                  {lipsyncErrorMessage && (
+                    <div style={{ marginTop: 8, fontSize: 13, color: "#ff4d4f" }}>
+                      {lipsyncErrorMessage}
+                    </div>
+                  )}
+                </>
+              )}
+            </div>
+            <div className="aa-modal__footer">
+              {lipsyncStatus === "generating" && (
+                <button className="aa-btn aa-btn--danger" onClick={handleCancelLipsync}>
+                  取消生成
+                </button>
+              )}
+              {lipsyncStatus !== "generating" && (
+                <button className="aa-btn" onClick={handleCancelLipsync}>
+                  关闭
+                </button>
+              )}
+            </div>
+          </div>
+        </div>
+      )}
     </div>
   )
 }

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

@@ -1,5 +1,5 @@
 /**
- * AI数字人 — API 封装
+ * AI数字人 — API 封装（#1822 契约对齐）
  */
 import apiClient from "@/api/client"
 import type { Script, LipsyncJob, RenderJob, BRollSegment } from "../types"
@@ -28,17 +28,26 @@ export const deleteScript = async (id: string): Promise<void> => {
   await apiClient.delete(`/scripts/${id}`)
 }
 
-/* ── 素材单查（用于拿到 file_url 传给对口型等新接口） ── */
+/* ── 素材单查（拿到 file_url 作为对口型的 video_url） ── */
 export const getAssetById = async (id: string): Promise<{ file_url?: string; id: string }> => {
   const response = await apiClient.get<{ file_url?: string; id: string }>(`/assets/${id}`)
   return response.data
 }
 
-/* ── 对口型 ── */
+/* ── 对口型（模式A：TTS 直生，后端内部合成音频；不要先调 TTS 拿 audio_url） ── */
 export const createLipsyncJob = async (data: {
-  voice_id: string
-  script_text: string
+  /** 人物视频 URL（MP4）；由素材 id 经 getAssetById 拿 file_url，禁止传 video_asset_id */
   video_url: string
+  /** 音色 ID（预置音色 或 克隆音色 profile UUID，后端会解析） */
+  voice_id: string
+  /** 要合成的文案（手动输入或文案库内容） */
+  script_text: string
+  /** 语速 0.5~2.0，默认 1.0 */
+  speed?: number
+  /** 情绪英文枚举：natural/excited/calm/friendly */
+  emotion?: string
+  enable_video_loop?: boolean
+  project_id?: string
 }): Promise<LipsyncJob> => {
   const response = await apiClient.post<LipsyncJob>("/lipsync/jobs", data)
   return response.data
@@ -49,6 +58,18 @@ export const getLipsyncJob = async (id: string): Promise<LipsyncJob> => {
   return response.data
 }
 
+/* ── 智能封面（MediaKit 抽帧 + 质量评分选最佳帧，独立于渲染任务） ── */
+export const generateSmartCover = async (
+  video_url: string,
+  max_frames = 5,
+): Promise<{ cover_url: string; status: string; message: string }> => {
+  const response = await apiClient.post<{ cover_url: string; status: string; message: string }>(
+    "/ai-avatar/render/smart-cover",
+    { video_url, max_frames },
+  )
+  return response.data
+}
+
 /* ── 渲染 ── */
 export const submitRender = async (data: {
   lipsync_job_id: string

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

@@ -17,6 +17,10 @@ interface PanelCoverAndGenerateProps {
   onResolutionChange: (r: string) => void
   isGenerating: boolean
   onGenerate: () => void
+  /** 智能获取封面（MediaKit 选帧） */
+  onSmartCover: () => void
+  smartCoverLoading: boolean
+  canSmartCover: boolean
   /** 配置汇总信息 */
   summary: {
     videoName: string | null
@@ -50,6 +54,9 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
   onResolutionChange,
   isGenerating,
   onGenerate,
+  onSmartCover,
+  smartCoverLoading,
+  canSmartCover,
   summary,
 }) => {
   const uploadInputRef = useRef<HTMLInputElement>(null)
@@ -69,9 +76,10 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
     e.target.value = ""
   }
 
-  /** 从视频截取（使用配置的帧时间，默认首帧） */
-  const handleCaptureFromVideo = () => {
+  /** 智能获取封面（调后端 MediaKit 抽帧评分选最佳帧，#1822） */
+  const handleSmartCover = () => {
     onCoverConfigChange({ mode: "auto_frame" })
+    onSmartCover()
   }
 
   const lipsync = summary.lipsyncStatus ? LIPSYNC_STATUS_LABEL[summary.lipsyncStatus] : null
@@ -93,9 +101,11 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
         <button
           type="button"
           className={`aa-btn aa-btn--ghost${coverConfig.mode === "auto_frame" ? " active" : ""}`}
-          onClick={handleCaptureFromVideo}
+          onClick={handleSmartCover}
+          disabled={smartCoverLoading || !canSmartCover}
+          title={canSmartCover ? "基于对口型成片智能选帧" : "请先完成对口型生成"}
         >
-          🎬 从视频截取
+          {smartCoverLoading ? "⏳ 智能选帧中…" : "🎬 智能获取封面"}
         </button>
         <button
           type="button"

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

@@ -7,16 +7,17 @@
  * - AiAvatarTitleConfig ↔ TitleSettings 的双向适配
  * - 自动生成字幕开关
  */
-import React, { useMemo, useState } from "react"
+import React, { useMemo, useState, useEffect } from "react"
+import { Input } from "antd"
 import TitleStylePanel from "@/pages/generate/components/title/TitleStylePanel"
+import TitleLibraryAutoComplete from "@/pages/generate/components/title/TitleLibraryAutoComplete"
+import type { TitleOption } from "@/pages/generate/components/title/TitleLibraryAutoComplete"
 import type { TitleSettings } from "@/pages/generate/types"
-import {
-  POSITION_OPTIONS,
-  FONT_OPTIONS,
-  TITLE_PRESETS,
-  getFontFamily,
-} from "@/pages/generate/constants"
+import { POSITION_OPTIONS, FONT_OPTIONS, TITLE_PRESETS } from "@/pages/generate/constants"
 import type { AiAvatarTitleConfig } from "../types"
+import { getTitles } from "@/api/titles"
+
+const { TextArea } = Input
 
 interface PanelTitleConfigProps {
   titleConfig: AiAvatarTitleConfig
@@ -27,6 +28,14 @@ const PanelTitleConfig: React.FC<PanelTitleConfigProps> = ({ titleConfig, onUpda
   /** TitleStylePanel 内部高亮的预设 key（面板本地状态） */
   const [activePreset, setActivePreset] = useState<string | null>(null)
 
+  /** 标题库选项（复用智能剪辑的标题库） */
+  const [titleOptions, setTitleOptions] = useState<TitleOption[]>([])
+  useEffect(() => {
+    getTitles()
+      .then((items) => setTitleOptions(items.map((t) => ({ label: t.content, value: t.content }))))
+      .catch(() => setTitleOptions([]))
+  }, [])
+
   /** AiAvatarTitleConfig → TitleSettings（补齐 aiAutoSelect / 自由坐标字段） */
   const titleSettings: TitleSettings = useMemo(
     () => ({
@@ -62,37 +71,32 @@ const PanelTitleConfig: React.FC<PanelTitleConfigProps> = ({ titleConfig, onUpda
 
   return (
     <div className="aa-title-config">
-      {/* 主标题输入 */}
+      {/* 主标题输入 — TextArea 多行 + 标题库选择 */}
       <div className="aa-form-field">
         <label className="aa-label">主标题</label>
-        <input
-          className="aa-input aa-title-input"
-          type="text"
-          placeholder="输入视频标题（留空则不显示标题）"
+        <TextArea
+          className="aa-title-input"
+          placeholder="输入视频标题（支持 / 分行）"
           value={titleConfig.title}
-          maxLength={30}
+          autoSize={{ minRows: 2, maxRows: 4 }}
+          maxLength={200}
           onChange={(e) => onUpdate({ title: e.target.value })}
+          style={{ fontSize: 15 }}
         />
-        {titleConfig.title && (
-          <div
-            style={{
-              fontSize: 13,
-              padding: "6px 8px",
-              background: "#f8f8fc",
-              borderRadius: 6,
-              fontFamily: getFontFamily(titleConfig.font),
-              fontWeight: titleConfig.bold ? 700 : 400,
-              fontStyle: titleConfig.italic ? "italic" : "normal",
-              color: titleConfig.color,
-              textShadow: titleConfig.shadow ? "1px 1px 3px rgba(0,0,0,0.6)" : undefined,
-              overflow: "hidden",
-              textOverflow: "ellipsis",
-              whiteSpace: "nowrap",
+        <div style={{ marginTop: 8, display: "flex", alignItems: "center", gap: 8 }}>
+          <span style={{ fontSize: 12, color: "#8c8ca1", whiteSpace: "nowrap" }}>📚 标题库</span>
+          <TitleLibraryAutoComplete
+            key={titleConfig.title}
+            placeholder="选择标题填入上方"
+            value=""
+            onChange={(val) => {
+              if (val) onUpdate({ title: val })
             }}
-          >
-            {titleConfig.title}
-          </div>
-        )}
+            options={titleOptions}
+            maxLength={200}
+            style={{ flex: 1 }}
+          />
+        </div>
       </div>
 
       {/* 标题样式：直接复用智能剪辑 TitleStylePanel（位置/字体/字号/样式/预设） */}

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

@@ -4,12 +4,15 @@
  * - 已选视频：竖屏 9:16 预览播放器 + 视频信息卡片 + 移除按钮
  */
 import type { AssetItem } from "@/api/assets"
+import type { AiAvatarTitleConfig } from "../types"
+import { getFontFamily } from "@/pages/generate/constants"
 
 export interface PanelVideoSelectorProps {
   selectedVideo: AssetItem | null
   /** 触发打开素材库弹窗 */
   onSelectVideo: () => void
   onRemoveVideo: () => void
+  titleConfig?: AiAvatarTitleConfig
 }
 
 /** 格式化时长（秒 → mm:ss） */
@@ -24,6 +27,7 @@ export function PanelVideoSelector({
   selectedVideo,
   onSelectVideo,
   onRemoveVideo,
+  titleConfig,
 }: PanelVideoSelectorProps) {
   /* 未选视频：虚线上传区，点击打开素材库弹窗 */
   if (!selectedVideo) {
@@ -53,13 +57,42 @@ export function PanelVideoSelector({
 
   return (
     <div>
-      {/* 竖屏 9:16 视频预览播放器 */}
-      <div className="aa-video-preview">
+      {/* 竖屏 9:16 视频预览播放器 + 标题实时预览 */}
+      <div className="aa-video-preview" style={{ position: "relative" }}>
         {fileUrl ? (
           <video src={fileUrl} poster={selectedVideo.thumbnail_url} controls playsInline />
         ) : (
           <div className="aa-video-preview__placeholder">视频暂不可预览</div>
         )}
+        {titleConfig?.title && (
+          <div
+            style={{
+              position: "absolute",
+              left: "50%",
+              transform: "translateX(-50%)",
+              ...(titleConfig.position === "top"
+                ? { top: "10%" }
+                : titleConfig.position === "bottom"
+                  ? { bottom: "10%" }
+                  : { top: "50%", transform: "translate(-50%, -50%)" }),
+              fontSize: Math.max(titleConfig.size, 32),
+              fontFamily: getFontFamily(titleConfig.font),
+              color: titleConfig.color,
+              fontWeight: titleConfig.bold ? 700 : 400,
+              fontStyle: titleConfig.italic ? "italic" : "normal",
+              textShadow: "0 2px 4px rgba(0,0,0,0.5)",
+              WebkitTextStroke: "2px #000",
+              pointerEvents: "none",
+              zIndex: 10,
+              maxWidth: "90%",
+              textAlign: "center",
+              whiteSpace: "pre-wrap",
+              lineHeight: 1.3,
+            }}
+          >
+            {titleConfig.title}
+          </div>
+        )}
       </div>
 
       {/* 视频信息卡片：文件名 / 时长 / 分辨率 */}

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

@@ -6,6 +6,7 @@ import { useEffect, useRef, useState } from "react"
 import { message } from "antd"
 import { fetchVoices } from "@/api/voices/voices"
 import { previewTts } from "@/api/tts"
+import { normalizeEmotion } from "../utils/contract"
 import type { UnifiedVoiceItem } from "@/api/voices/types"
 import {
   type VoiceSource,
@@ -92,11 +93,14 @@ export function PanelVoiceSelector({
 
   /** 用指定 URL 真实播放（抽取公共） */
   const playAudioUrl = (voiceId: string, url: string) => {
+    // 临时兼容：后端 /tts/preview 返回 HTTP URL，staging 是 HTTPS，Mixed Content 会阻止加载
+    // OSS 同时支持 HTTP/HTTPS，直接替换协议即可
+    const safeUrl = url.startsWith("http://") ? url.replace("http://", "https://") : url
     if (audioRef.current) {
       audioRef.current.pause()
       audioRef.current = null
     }
-    const audio = new Audio(url)
+    const audio = new Audio(safeUrl)
     audioRef.current = audio
     setPreviewingId(voiceId)
     audio.onended = () => {
@@ -134,12 +138,25 @@ export function PanelVoiceSelector({
         return
       }
       const targetId = voice.voice_clone_profile_id || voice.id
+      // DEBUG: 打印请求参数，帮助定位 /tts/preview 失败原因
+      console.log("[AI数字人-克隆试听] previewTts 请求:", {
+        voice_id: targetId,
+        voice_name: voice.name,
+        voice_type: voice.type,
+        voice_clone_profile_id: voice.voice_clone_profile_id,
+        voice_id_field: voice.voice_id,
+      })
       setPreviewingId(voice.id)
       try {
         const res = await previewTts({
           text: VOICE_PREVIEW_TEXT,
           voice_id: targetId,
-          speed: 1.0,
+          speed: speed, // 透传用户选择的语速（#1822）
+          emotion: normalizeEmotion(emotion), // 情绪中文→英文枚举
+        })
+        console.log("[AI数字人-克隆试听] previewTts 响应:", {
+          audio_url: res.audio_url?.substring(0, 80),
+          duration: res.duration,
         })
         if (!res.audio_url) {
           setPreviewingId(null)
@@ -148,8 +165,14 @@ export function PanelVoiceSelector({
         }
         previewCacheRef.current.set(targetId, res.audio_url)
         playAudioUrl(voice.id, res.audio_url)
-      } catch {
+      } catch (err) {
         setPreviewingId(null)
+        // DEBUG: 打印详细错误信息
+        console.error("[AI数字人-克隆试听] previewTts 失败:", {
+          status: (err as { response?: { status?: number } })?.response?.status,
+          data: (err as { response?: { data?: unknown } })?.response?.data,
+          message: err instanceof Error ? err.message : String(err),
+        })
         // apiClient 拦截器已统一 toast
       }
       return
@@ -229,7 +252,7 @@ export function PanelVoiceSelector({
                   type="button"
                   className="aa-voice-card__preview"
                   title={previewingId === voice.id ? "停止试听" : "试听"}
-                  disabled={!previewUrl}
+                  disabled={voice.type === "preset" && !previewUrl}
                   onClick={(e) => {
                     e.stopPropagation()
                     handlePreview(voice)

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

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

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

@@ -77,6 +77,9 @@ export interface AiAvatarTitleConfig {
   shadow: boolean
   color: string
   auto_subtitle: boolean
+  /** 自定义位置坐标（position=custom 时生效，像素） */
+  pos_x?: number
+  pos_y?: number
 }
 
 /* ── 封面配置 ── */
@@ -86,6 +89,8 @@ export interface AiAvatarCoverConfig {
   frame_time: number
   upload_url: string | null
   thumbnail_url: string | null
+  /** 智能封面（MediaKit 选帧）返回的 OSS 非临时 URL（#1822） */
+  smart_cover_url: string | null
 }
 
 /* ── 渲染任务 ── */
@@ -103,7 +108,7 @@ export interface RenderJob {
 /* ── 默认值 ── */
 export const DEFAULT_TITLE_CONFIG: AiAvatarTitleConfig = {
   title: "",
-  position: "top",
+  position: "bottom",
   font: "思源黑体",
   size: 28,
   bold: true,
@@ -112,6 +117,8 @@ export const DEFAULT_TITLE_CONFIG: AiAvatarTitleConfig = {
   shadow: false,
   color: "#ffffff",
   auto_subtitle: true,
+  pos_x: undefined,
+  pos_y: undefined,
 }
 
 export const DEFAULT_COVER_CONFIG: AiAvatarCoverConfig = {
@@ -120,4 +127,5 @@ export const DEFAULT_COVER_CONFIG: AiAvatarCoverConfig = {
   frame_time: 0,
   upload_url: null,
   thumbnail_url: null,
+  smart_cover_url: null,
 }

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

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

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

@@ -548,10 +548,21 @@ const GeneratePage: React.FC = () => {
             <div
               style={{
                 display: "flex",
-                justifyContent: "center",
+                flexDirection: "column",
+                alignItems: "center",
                 marginTop: 16,
               }}
             >
+              <h2
+                style={{
+                  textAlign: "center",
+                  marginBottom: 12,
+                  fontSize: "1.5rem",
+                  fontWeight: 600,
+                }}
+              >
+                🎬 确认生成
+              </h2>
               <div
                 style={{
                   display: "flex",

docs/ai-avatar-api-contract-1822.md

@@ -0,0 +1,169 @@
+# AI 数字人前后端接口契约（#1797 / #1822）
+
+> 分支：`fix/ai-avatar-v3-1797`
+> 范围：TTS→对口型链路打通、语速/情绪透传、封面智能选帧、标题字段对齐
+> 本文档为前后端联调的唯一字段口径。
+
+---
+
+## 1. 对口型创建接口 `POST /api/v1/lipsync/jobs`
+
+支持两种输入模式，**二选一**：
+
+### 模式 A（推荐）：TTS 直生 —— 传音色 + 文案，后端内部合成音频
+
+前端无需先调 TTS。后端收到请求后：先调 CosyVoice 合成音频 → 转存 OSS → 再提交 MediaKit 对口型。
+
+```jsonc
+{
+  "video_url": "https://oss.../person.mp4",   // 必填，人物视频（MP4）
+  "voice_id": "cosyvoice-v3-flash-99-xxxx",    // 必填，音色 ID（预置音色 或 克隆 profile UUID）
+  "script_text": "省是浙江省，市是永康市……",     // 必填，要合成的文案
+  "speed": 1.0,                                 // 可选，语速 0.5~2.0，默认 1.0
+  "emotion": "excited",                         // 可选，情绪，见 §3
+  "enable_video_loop": false,                   // 可选，音频长于视频时是否循环画面
+  "project_id": ""                              // 可选
+}
+```
+
+### 模式 B：直接音频 —— 前端已准备好音频
+
+```jsonc
+{
+  "video_url": "https://oss.../person.mp4",     // 必填
+  "audio_url": "https://oss.../voice.mp3",      // 必填，mp3/aac/wav/m4a/flac
+  "enable_video_loop": false
+}
+```
+
+### 校验与错误码
+
+| 场景 | HTTP | detail.code |
+|------|------|-------------|
+| 既无 audio_url 又无 voice_id+script_text | 422 | （schema 校验） |
+| video_url 非 MP4 / audio_url 格式不支持 | 422 | （schema 校验） |
+| 克隆音色不属于当前用户 | 403 | `VoiceForbidden` |
+| 克隆音色尚未合成完成 | 400 | `VoiceNotReady` |
+| TTS 合成失败（如 CosyVoice 欠费） | 502 | `TTSSynthesisFailed` |
+| MediaKit 提交失败 | 502 | `*`（透传 MediaKit code） |
+
+### 轮询
+
+- `GET /api/v1/lipsync/jobs/{id}`：非终态任务先返回 DB 缓存，**后台异步刷新 MediaKit**（不会阻塞轮询）。
+- `status` 流转：`pending` → `submitted` → `running`/`processing`（MediaKit 中间态同步）→ `completed` / `failed`。
+- `completed` 时 `output_video_url` 为**已转存自家 OSS 的非临时 URL**（不会过期）。
+- 前端每 3s 轮询，命中 `completed`/`failed` 即停。
+
+---
+
+## 2. TTS 合成接口语速/情绪透传
+
+- `POST /api/v1/tts/synthesize`（异步任务）与 `POST /api/v1/tts/preview`（即时试听）均新增：
+  - `speed`：float，0.5~2.0，默认 1.0 → 透传 CosyVoice payload 的 `rate`
+  - `emotion`：string，见 §3 映射 → 透传 `emotion`
+- 透传链路：`route → CreateTTSJobUseCase(metadata) → TTSJobWorkflow.start_synthesis / 分段合成 → CosyVoiceService.submit_synthesize_task(rate/emotion)`。
+- 分段合成（长文案）与失败重合成路径同样透传 speed/emotion。
+
+---
+
+## 3. 情绪枚举（前后端统一）
+
+前端把中文选项映射成英文后传后端；后端同时接受中文/英文，非法值忽略（走默认自然）。
+
+| 前端选项 | 传参值 | CosyVoice 枚举 |
+|---------|--------|---------------|
+| 自然 | `natural` | natural |
+| 兴奋 | `excited` | excited |
+| 沉稳 | `calm` | calm |
+| 亲切 | `friendly` | friendly |
+
+后端 `normalize_emotion()` 也接受中文（自然/兴奋/沉稳/亲切）做兜底映射。
+
+---
+
+## 4. 智能封面接口 `POST /api/v1/ai-avatar/render/smart-cover`
+
+独立接口，**不依赖渲染任务**，前端「智能获取封面」按钮直接调用。
+
+**请求**
+```jsonc
+{
+  "video_url": "https://oss.../avatar_output.mp4",  // 必填，数字人视频
+  "max_frames": 5                                    // 可选，抽帧数量 1~10，默认 5
+}
+```
+
+**响应**
+```jsonc
+{
+  "cover_url": "https://oss.../ai-avatar/covers/xxx/cover_yy.jpg", // OSS 非临时 URL
+  "status": "completed",       // completed / fallback_failed
+  "message": ""                // 失败原因
+}
+```
+
+**实现**：复用智能剪辑同款能力 —— MediaKit `extract_frames(SpecifiedFrames)` 抽 5 帧 → `cover_frame_scorer.score_frames`（清晰度+亮度+色彩）评分选最佳 → 转存 OSS。
+**不再使用 FFmpeg 简单首帧**。渲染管线最终封面也优先走该智能选帧，MediaKit 不可用时才回退 FFmpeg。
+
+---
+
+## 5. 渲染接口 `POST /api/v1/ai-avatar/render`
+
+```jsonc
+{
+  "lipsync_job_id": "7c29a3b2-...",     // 必填，已 completed 的对口型任务
+  "script_id": "",                       // 可选！见下方说明
+  "b_roll_segments": [],                 // 可选，B-roll 片段
+  "title_config": { ... },               // 可选，单个标题配置 dict（见 §6）
+  "cover_config": { ... },               // 可选，封面配置（建议改用 smart-cover）
+  "project_id": ""
+}
+```
+
+**`script_id` 是否必填：可选。**
+- 从文案库选了文案时传对应文案 ID（后端做归属校验）。
+- **手动输入文案、走 TTS 直生模式时不传（留空）即可**——渲染管线不依赖文案内容，`script_id` 仅用于归属校验。留空不会卡手动文案用户。
+
+---
+
+## 6. 标题配置 `title_config` 字段清单（以 build_title_drawtext_filter 为准）
+
+渲染请求收的是**单个 `title_config` dict**（不是 `titles[]` 数组），字段与 `packages/domain/video_filter_builder.py` 的 `build_title_drawtext_filter()` 完全对齐：
+
+| 字段 | 别名 | 类型 | 必填 | 默认 | 说明 |
+|------|------|------|------|------|------|
+| `text` | `content` | string | ✅ | — | 标题文字；为空或 `enabled=false` 时不渲染标题 |
+| `enabled` | — | bool | ❌ | `true` | 是否启用标题；false 跳过 |
+| `font` | `font_preset` | string | ❌ | 思源黑体 | 字体名（后端按名字解析字体文件） |
+| `font_size` | `size` | int | ❌ | 36 | 字号（像素） |
+| `font_color` | `color` | string | ❌ | `#ffffff` | 文字颜色，`#RRGGBB`；后端自动去掉 `#`，也可传 `RRGGBB` 或颜色名 |
+| `position` | — | string | ❌ | `top` | 预设位置：`top`(y=50) / `center`(垂直居中) / `bottom`(底部上移50px) / `custom` |
+| `pos_x` | — | int/float | ❌ | — | 自定义 X 坐标（像素），仅 `position=custom` 生效 |
+| `pos_y` | — | int/float | ❌ | — | 自定义 Y 坐标（像素），仅 `position=custom` 生效 |
+| `bold` | — | bool | ❌ | `true` | 粗体（Bold 字体变体，回退 borderw 模拟） |
+| `stroke` | — | bool/object | ❌ | — | 描边。`true`=黑描边宽2；object 见下 |
+| `stroke.enabled` | — | bool | ❌ | true | 是否描边 |
+| `stroke.width` | — | int | ❌ | 2 | 描边宽度 |
+| `stroke.color` | — | string | ❌ | `#000000` | 描边颜色 |
+| `shadow` | — | bool/object | ❌ | — | 阴影。`true`=黑色阴影偏移2px；object 见下 |
+| `shadow.enabled` | — | bool | ❌ | true | 是否阴影 |
+| `shadow.color` | — | string | ❌ | `#000000` | 阴影颜色 |
+| `shadow.offset_x` | — | int | ❌ | 2 | 阴影 X 偏移 |
+| `shadow.offset_y` | — | int | ❌ | 2 | 阴影 Y 偏移 |
+
+**前端注意事项**
+- 标题是**整条成片一个标题**（单个 dict），不是按时间段的标题数组；没有 `start`/`end`/`frame`/`fontSize` 这些字段。
+- 位置用 `position` 四档枚举；自由摆放用 `position="custom"` + `pos_x`/`pos_y`（像素坐标，非比例）。
+- 颜色统一传 `#RRGGBB` 即可，后端会处理 `#`；三档预设位置下标题始终水平居中。
+- `stroke`/`shadow` 传 `true` 用默认样式，或传 object 精细控制颜色/宽度/偏移。
+
+---
+
+## 7. 前端对接清单
+
+1. 对口型：改用**模式 A**（voice_id + script_text + speed + emotion），不要再先调 TTS 拿 audio_url。
+2. 音色 ID：`voice_id` 可直接传克隆音色的 profile UUID，后端会解析为 CosyVoice voice_id（与 /tts 一致）。
+3. 情绪下拉：自然/兴奋/沉稳/亲切 → natural/excited/calm/friendly。
+4. 封面：点「智能获取封面」→ POST `/ai-avatar/render/smart-cover`，用返回的 `cover_url`。
+5. 渲染：手动文案直生场景 `script_id` 留空；标题传**单个** `title_config` dict（字段见 §6）。
+6. 轮询：识别 `running` 等中间态，不要只认 `submitted`。

infra/docker/compose.yml

@@ -128,9 +128,9 @@ services:
       - xiaoxia-net
 
     # 健康检查配置
-    # 注：celery inspect ping 依赖 broker 连接，在容器内不可靠，改用进程检查
+    # 注：容器内无 pgrep/ps，扫描 /proc 所有进程的 cmdline 查找 celery 进程
     healthcheck:
-      test: ["CMD-SHELL", "for pid in /proc/[0-9]*/cmdline; do if grep -ql celery \"$pid\" 2>/dev/null; then exit 0; fi; done; exit 1"]
+      test: ["CMD-SHELL", "grep -lq celery /proc/[0-9]*/cmdline 2>/dev/null || exit 1"]
       interval: 30s
       timeout: 10s
       retries: 3

infra/docker/deploy-production-registry.sh

@@ -155,7 +155,7 @@ docker run -d \
   --restart unless-stopped \
   --cpus 2 \
   --memory 2g \
-  --health-cmd "sh -c \"for pid in /proc/[0-9]*/cmdline; do if grep -ql celery \"$pid\" 2>/dev/null; then exit 0; fi; done; exit 1\"" \
+  --health-cmd "sh -c \"grep -lq celery /proc/[0-9]*/cmdline 2>/dev/null || exit 1\"" \
   --health-interval 30s \
   --health-timeout 10s \
   --health-retries 3 \

infra/docker/deploy-staging-registry.sh

@@ -116,7 +116,7 @@ docker run -d \
   -v "$GENERATED_DIR:/app/generated" \
   --restart unless-stopped \
   --label com.centurylinklabs.watchtower.enable=true \
-  --health-cmd "sh -c \"for pid in /proc/[0-9]*/cmdline; do if grep -ql celery \"$pid\" 2>/dev/null; then exit 0; fi; done; exit 1\"" \
+  --health-cmd "sh -c \"grep -lq celery /proc/[0-9]*/cmdline 2>/dev/null || exit 1\"" \
   --health-interval 30s \
   --health-timeout 10s \
   --health-retries 3 \

infra/docker/worker.Dockerfile

@@ -38,6 +38,10 @@ RUN chmod +x /usr/local/bin/entrypoint-worker.sh
 # 业务代码（变化最频繁，放最后）
 COPY apps/worker/ /app/apps/worker/
 
+# 健康检查：扫描所有进程的 cmdline 查找 celery 进程
+HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \
+    CMD grep -lq celery /proc/[0-9]*/cmdline 2>/dev/null || exit 1
+
 USER celery
 WORKDIR /app/apps/worker
 CMD ["/usr/local/bin/entrypoint-worker.sh"]

packages/adapters/sqlalchemy_impl/models.py

@@ -684,9 +684,15 @@ class LipsyncJobModel(Base):
 
     # 输入参数
     video_url = Column(Text, nullable=False)
-    audio_url = Column(Text, nullable=False)
+    audio_url = Column(Text, nullable=True)  # 直生模式（voice_id+script_text）下 TTS 合成后回填
     enable_video_loop = Column(Boolean, nullable=False, default=False)
 
+    # TTS 直生字段：传音色 + 文案，由后端先合成音频再对口型
+    voice_id = Column(String(200), nullable=False, default="")
+    script_text = Column(Text, nullable=False, default="")
+    speed = Column(Float, nullable=False, default=1.0)
+    emotion = Column(String(20), nullable=False, default="")
+
     # MediaKit 任务状态
     mediakit_task_id = Column(String(200), nullable=False, default="", index=True)
     status = Column(
@@ -715,7 +721,8 @@ class AiAvatarRenderJob(Base):
 
     # 输入参数
     lipsync_job_id = Column(String(36), nullable=False)
-    script_id = Column(String(36), nullable=False)
+    # 文案 ID 可选：手动输入文案（TTS 直生）场景不关联文案库条目
+    script_id = Column(String(36), nullable=False, default="")
     b_roll_segments = Column(JSON, nullable=False, default=list)
     # b_roll_segments 格式: [{"script_segment_index": 0, "asset_url": "...", "mode": "fullscreen|pip", "start_time": 5.0, "end_time": 10.0}, ...]
     title_config = Column(JSON, nullable=False, default=dict)

packages/application/cosyvoice_service.py

@@ -25,6 +25,35 @@ from packages.shared.config import get_shared_settings
 logger = logging.getLogger(__name__)
 
 
+# CosyVoice 支持的情绪：中文标签 → API 英文值
+EMOTION_MAP = {
+    "自然": "natural",
+    "兴奋": "excited",
+    "沉稳": "calm",
+    "亲切": "friendly",
+    "natural": "natural",
+    "excited": "excited",
+    "calm": "calm",
+    "friendly": "friendly",
+}
+VALID_EMOTIONS = {"natural", "excited", "calm", "friendly"}
+
+
+def normalize_emotion(emotion: str) -> str:
+    """将前端情绪值归一化为 CosyVoice 英文枚举。
+
+    支持中文（自然/兴奋/沉稳/亲切）和英文；非法值返回空串（不传，走默认）。
+    """
+    if not emotion:
+        return ""
+    key = emotion.strip().lower()
+    mapped = EMOTION_MAP.get(emotion.strip()) or EMOTION_MAP.get(key)
+    if mapped and mapped in VALID_EMOTIONS:
+        return mapped
+    logger.warning("未知的 emotion 值，忽略: %r", emotion)
+    return ""
+
+
 class CosyVoiceError(Exception):
     """CosyVoice API 调用异常。"""
 
@@ -431,6 +460,7 @@ class CosyVoiceService:
         format: str = "",
         speed: float = 1.0,
         volume: int = 50,
+        emotion: str = "",
     ) -> dict:
         """提交语音合成任务（同步非流式，直接返回结果）.
 
@@ -444,6 +474,7 @@ class CosyVoiceService:
             format: 输出格式（mp3/wav/pcm），空表示使用配置默认值
             speed: 语速（0.5-2.0），1.0 为正常速度
             volume: 音量（0-100），默认 50
+            emotion: 情绪（natural/excited/calm/friendly），空串不传
 
         Returns:
             dict: {"audio_url": str, "request_id": str,
@@ -463,17 +494,20 @@ class CosyVoiceService:
 
         settings = get_shared_settings()
 
-        payload = {
-            "model": self._model,
-            "input": {
-                "text": text,
-                "voice": voice_id,
-                "format": format or settings.cosyvoice_format,
-                "sample_rate": sample_rate or settings.cosyvoice_sample_rate,
-                "rate": speed,
-                "volume": volume,
-            },
+        input_payload: dict[str, Any] = {
+            "text": text,
+            "voice": voice_id,
+            "format": format or settings.cosyvoice_format,
+            "sample_rate": sample_rate or settings.cosyvoice_sample_rate,
+            "rate": speed,
+            "volume": volume,
         }
+        # 情绪：归一化（中文→英文）后透传；空/非法则不传，走 CosyVoice 默认
+        norm_emotion = normalize_emotion(emotion)
+        if norm_emotion:
+            input_payload["emotion"] = norm_emotion
+
+        payload = {"model": self._model, "input": input_payload}
 
         response = self._call_api(
             method="POST",
@@ -490,6 +524,10 @@ class CosyVoiceService:
         if not audio_url:
             raise CosyVoiceError(f"CosyVoice API 未返回 audio_url: {response}")
 
+        # DashScope 返回 http://，统一升级为 https://
+        if audio_url.startswith("http://"):
+            audio_url = audio_url.replace("http://", "https://", 1)
+
         return {
             "task_id": "",  # 同步接口无 task_id，兼容旧接口
             "audio_url": audio_url,
@@ -517,6 +555,7 @@ class CosyVoiceService:
         format: str = "",
         speed: float = 1.0,
         volume: int = 50,
+        emotion: str = "",
         timeout: float = 120.0,
     ) -> SynthesizeResult:
         """语音合成（同步非流式）.
@@ -548,6 +587,7 @@ class CosyVoiceService:
             format=format,
             speed=speed,
             volume=volume,
+            emotion=emotion,
         )
 
         return SynthesizeResult(

packages/application/tts_job/workflow.py

@@ -143,11 +143,16 @@ class TTSWorkflowService:
             return self._start_segment_synthesis(job)
 
         try:
+            _meta = dict(job.metadata)
+            _speed = float(_meta.get("speed", 1.0) or 1.0)
+            _emotion = str(_meta.get("emotion", "") or "")
             submit_result = self.cosyvoice_service.submit_synthesize_task(
                 text=job.input_text,
                 voice_id=job.voice_id,
                 sample_rate=job.sample_rate,
                 format=job.format,
+                speed=_speed,
+                emotion=_emotion,
             )
 
             # 保存 task_id / request_id 到 metadata
@@ -283,6 +288,7 @@ class TTSWorkflowService:
             job_metadata = job.metadata or {}
             speed = float(job_metadata.get("speed", 1.0))
             volume = int(job_metadata.get("volume", 50))
+            emotion = str(job_metadata.get("emotion", "") or "")
 
             result = self.cosyvoice_service.submit_synthesize_task(
                 text=job.input_text,
@@ -291,6 +297,7 @@ class TTSWorkflowService:
                 format=job.format,
                 speed=speed,
                 volume=volume,
+                emotion=emotion,
             )
             audio_url = result.get("audio_url", "")
             if not audio_url:
@@ -401,6 +408,9 @@ class TTSWorkflowService:
         """
         max_workers = min(len(segments), _MAX_SEGMENT_WORKERS)
         results: list[dict | None] = [None] * len(segments)
+        _seg_meta = job.metadata or {}
+        _seg_speed = float(_seg_meta.get("speed", 1.0) or 1.0)
+        _seg_emotion = str(_seg_meta.get("emotion", "") or "")
 
         with ThreadPoolExecutor(max_workers=max_workers) as executor:
             future_to_idx = {}
@@ -411,6 +421,8 @@ class TTSWorkflowService:
                     voice_id=job.voice_id,
                     sample_rate=job.sample_rate,
                     format=job.format,
+                    speed=_seg_speed,
+                    emotion=_seg_emotion,
                 )
                 future_to_idx[future] = idx
 
@@ -500,6 +512,7 @@ class TTSWorkflowService:
         job_metadata = job.metadata or {}
         speed = float(job_metadata.get("speed", 1.0))
         volume = int(job_metadata.get("volume", 50))
+        emotion = str(job_metadata.get("emotion", "") or "")
 
         # 分段文本（用于缺失段重新合成）
         segments = split_text(job.input_text, max_chars=_SEGMENT_THRESHOLD)
@@ -534,6 +547,7 @@ class TTSWorkflowService:
                         format=job.format,
                         speed=speed,
                         volume=volume,
+                        emotion=emotion,
                     )
                     future_to_idx[future] = idx
 

tests/unit/test_ai_avatar_emotion_tts_lipsync.py

@@ -0,0 +1,297 @@
+"""#1822 情绪/语速透传 + 对口型 TTS 直生 + 智能封面 单元测试.
+
+CI 增量映射:
+  cosyvoice_service.normalize_emotion / payload emotion
+  lipsync_service TTS 直生分支（voice_id+script_text）
+  ai_avatar_cover_service 智能选帧
+"""
+
+import os
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+os.environ.setdefault("JWT_SECRET_KEY", "dev-secret-key-for-testing")
+
+
+# ── 情绪归一化 ──────────────────────────────────────────────────────────
+
+
+def test_normalize_emotion_english_values():
+    from packages.application.cosyvoice_service import normalize_emotion
+
+    assert normalize_emotion("natural") == "natural"
+    assert normalize_emotion("excited") == "excited"
+    assert normalize_emotion("calm") == "calm"
+    assert normalize_emotion("friendly") == "friendly"
+    # 大小写 / 空白容错
+    assert normalize_emotion("  Excited  ") == "excited"
+
+
+def test_normalize_emotion_chinese_values():
+    from packages.application.cosyvoice_service import normalize_emotion
+
+    assert normalize_emotion("自然") == "natural"
+    assert normalize_emotion("兴奋") == "excited"
+    assert normalize_emotion("沉稳") == "calm"
+    assert normalize_emotion("亲切") == "friendly"
+
+
+def test_normalize_emotion_invalid_returns_empty():
+    from packages.application.cosyvoice_service import normalize_emotion
+
+    assert normalize_emotion("") == ""
+    assert normalize_emotion("angry") == ""
+    assert normalize_emotion("喜怒哀乐") == ""
+
+
+# ── CosyVoice payload 携带 emotion + rate ──────────────────────────────
+
+
+def _make_service_with_captured_client(captured: dict):
+    """构造 CosyVoiceService，拦截 post 请求体到 captured['json']."""
+    import httpx as _httpx
+
+    from packages.application import cosyvoice_service as mod
+
+    mock_client = MagicMock(spec=_httpx.Client)
+    resp = MagicMock()
+    resp.status_code = 200
+    resp.json.return_value = {
+        "request_id": "req-1",
+        "output": {"audio": {"url": "https://tts/a.mp3", "duration": 1.0}},
+    }
+    resp.raise_for_status = MagicMock()
+
+    def fake_request(method, url, headers, json, timeout):
+        captured["json"] = json
+        return resp
+
+    mock_client.request.side_effect = fake_request
+
+    with patch.object(mod, "get_shared_settings") as settings_patch:
+        s = MagicMock()
+        s.cosyvoice_api_key = "sk-test"
+        s.cosyvoice_base_url = "https://x/api/v1"
+        s.cosyvoice_model = "cosyvoice-v3-flash"
+        s.cosyvoice_clone_model = "voice-enrollment"
+        s.cosyvoice_format = "mp3"
+        s.cosyvoice_sample_rate = 22050
+        s.cosyvoice_voice = "longxiaochun"
+        settings_patch.return_value = s
+        svc = mod.CosyVoiceService(http_client=mock_client)
+    return svc
+
+
+def test_submit_synthesize_payload_includes_emotion_and_rate():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="v-1", speed=1.5, emotion="兴奋")
+
+    inp = captured["json"]["input"]
+    assert inp["emotion"] == "excited"
+    assert inp["rate"] == 1.5
+
+
+def test_submit_synthesize_payload_omits_emotion_when_empty():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="v-1")
+
+    assert "emotion" not in captured["json"]["input"]
+
+
+# ── 对口型 TTS 直生分支 ─────────────────────────────────────────────────
+
+
+def _lipsync_service_with_mocks():
+    from app.services.lipsync_service import LipsyncService
+
+    db = MagicMock()
+    client = MagicMock()
+    client.is_available = True
+    client.submit_lipsync.return_value = {
+        "success": True,
+        "task_id": "mk-1",
+        "request_id": "req-1",
+    }
+    cosy = MagicMock()
+    cosy.submit_synthesize_task.return_value = {
+        "audio_url": "https://tts/raw.mp3",
+        "request_id": "tts-req",
+        "audio_duration": 3.0,
+    }
+    svc = LipsyncService(db, client=client, cosyvoice_service=cosy, voice_clone_repo=MagicMock())
+    # _resolve_voice_id 默认原样返回（repo.get 返回 None）
+    svc._voice_clone_repo.get.return_value = None
+    return svc, client, cosy
+
+
+def test_create_job_tts_direct_mode_synthesizes_audio():
+    svc, client, cosy = _lipsync_service_with_mocks()
+
+    with (
+        patch("app.services.lipsync_service.get_shared_storage_service") as storage_patch,
+        patch("app.services.lipsync_service.safe_download_bytes") as dl_patch,
+    ):
+        storage = MagicMock()
+        storage.upload_file.return_value = "https://oss/tts.mp3"
+        storage_patch.return_value = storage
+        dl_patch.return_value = b"FAKEAUDIO"
+
+        job = svc.create_job(
+            user_id="user-1",
+            video_url="https://oss/person.mp4",
+            voice_id="cosy-v1",
+            script_text="你好世界",
+            speed=1.2,
+            emotion="兴奋",
+        )
+
+    # 调了 TTS 合成，带 speed/emotion
+    cosy.submit_synthesize_task.assert_called_once()
+    _, kwargs = cosy.submit_synthesize_task.call_args
+    assert kwargs["speed"] == 1.2
+    assert kwargs["emotion"] == "excited"
+    assert kwargs["voice_id"] == "cosy-v1"
+    # MediaKit 用合成后的 OSS 音频 URL 提交
+    _, submit_kwargs = client.submit_lipsync.call_args
+    assert submit_kwargs["audio_url"] == "https://oss/tts.mp3"
+    assert submit_kwargs["video_url"] == "https://oss/person.mp4"
+    # DB 记录了 TTS 字段
+    assert job.emotion == "excited"
+    assert job.speed == 1.2
+
+
+def test_create_job_direct_audio_mode_skips_tts():
+    svc, client, cosy = _lipsync_service_with_mocks()
+
+    job = svc.create_job(
+        user_id="user-1",
+        video_url="https://oss/person.mp4",
+        audio_url="https://oss/ready.mp3",
+    )
+
+    cosy.submit_synthesize_task.assert_not_called()
+    _, submit_kwargs = client.submit_lipsync.call_args
+    assert submit_kwargs["audio_url"] == "https://oss/ready.mp3"
+
+
+def test_create_job_tts_failure_raises():
+    from app.services.mediakit_client import MediaKitError
+
+    from packages.application.cosyvoice_service import CosyVoiceError
+
+    svc, client, cosy = _lipsync_service_with_mocks()
+    cosy.submit_synthesize_task.side_effect = CosyVoiceError("Arrearage")
+
+    with pytest.raises(MediaKitError) as exc:
+        svc.create_job(
+            user_id="user-1",
+            video_url="https://oss/person.mp4",
+            voice_id="v-1",
+            script_text="文本",
+        )
+    assert exc.value.code == "TTSSynthesisFailed"
+    # TTS 失败不应提交 MediaKit
+    client.submit_lipsync.assert_not_called()
+
+
+# ── refresh 同步中间状态 ────────────────────────────────────────────────
+
+
+def test_refresh_syncs_running_status():
+    from app.services.lipsync_service import LipsyncService
+
+    db = MagicMock()
+    client = MagicMock()
+    client.get_task_status.return_value = {"success": True, "status": "running"}
+    svc = LipsyncService(db, client=client)
+
+    job = MagicMock()
+    job.status = "submitted"
+    job.mediakit_task_id = "mk-1"
+    job.id = "j-1"
+    svc.get_job = MagicMock(return_value=job)
+
+    result = svc.refresh_job_status("j-1", "user-1")
+    assert result.status == "running"
+
+
+# ── 智能封面 ────────────────────────────────────────────────────────────
+
+
+def test_smart_cover_selects_best_frame_and_persists():
+    from app.services import ai_avatar_cover_service as cov
+
+    snapshots = [
+        {"image_url": "https://mk/f0.jpg"},
+        {"image_url": "https://mk/f1.jpg"},
+    ]
+    with (
+        patch("packages.shared.mediakit_client.get_mediakit_client") as mk_patch,
+        patch("packages.shared.cover_frame_scorer.score_frames") as score_patch,
+        patch("httpx.get") as http_get,
+        patch("packages.shared.storage.get_shared_storage_service") as storage_patch,
+    ):
+        mk = MagicMock()
+        mk.is_available = True
+        mk.extract_frames.return_value = snapshots
+        mk_patch.return_value = mk
+        # score_frames 把 f1 选为最佳
+        score_patch.side_effect = lambda cands: [
+            {"url": "https://mk/f1.jpg", "score": 90.0, "image_path": cands[1]["image_path"]},
+            {"url": "https://mk/f0.jpg", "score": 60.0, "image_path": cands[0]["image_path"]},
+        ]
+        resp = MagicMock()
+        resp.content = b"IMGDATA"
+        resp.raise_for_status = MagicMock()
+        http_get.return_value = resp
+        storage = MagicMock()
+        storage.upload_file.return_value = "https://oss/cover.jpg"
+        storage_patch.return_value = storage
+
+        url = cov.generate_smart_cover("https://oss/avatar.mp4", job_id="job-1")
+
+    assert url == "https://oss/cover.jpg"
+    mk.extract_frames.assert_called_once()
+    score_patch.assert_called_once()
+
+
+def test_smart_cover_returns_empty_when_mediakit_unavailable():
+    from app.services import ai_avatar_cover_service as cov
+
+    with patch("packages.shared.mediakit_client.get_mediakit_client") as mk_patch:
+        mk = MagicMock()
+        mk.is_available = False
+        mk_patch.return_value = mk
+        url = cov.generate_smart_cover("https://oss/avatar.mp4")
+    assert url == ""
+
+
+# ── 渲染 script_id 可选（手动文案直生场景）──────────────────────────────
+
+
+def test_render_request_script_id_optional():
+    import os
+    import sys
+
+    sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "apps", "api"))
+    from app.schemas.ai_avatar_render import CreateAiAvatarRenderRequest
+
+    # 手动文案直生：不传 script_id 也合法
+    req = CreateAiAvatarRenderRequest(lipsync_job_id="j-1")
+    assert req.script_id == ""
+
+    # 空白被 strip
+    req2 = CreateAiAvatarRenderRequest(lipsync_job_id="j-1", script_id="   ")
+    assert req2.script_id == ""
+
+    # title_config 是单个 dict
+    req3 = CreateAiAvatarRenderRequest(
+        lipsync_job_id="j-1",
+        title_config={"text": "标题", "position": "top", "font_size": 40},
+    )
+    assert req3.title_config["position"] == "top"

tests/unit/test_ai_avatar_render_service.py

@@ -176,13 +176,23 @@ class TestSchemaValidation:
                 script_id="script-1",
             )
 
-    def test_create_request_empty_script_id(self):
+    def test_create_request_empty_script_id_normalized(self):
+        """script_id 改为可选（TTS 直生场景）：空白值应规范化为空串而非抛错。"""
         from app.schemas.ai_avatar_render import CreateAiAvatarRenderRequest
 
-        with pytest.raises(ValueError, match="script_id 不能为空"):
+        req = CreateAiAvatarRenderRequest(
+            lipsync_job_id="lipsync-1",
+            script_id="  ",
+        )
+        assert req.script_id == ""
+
+    def test_create_request_empty_lipsync_job_id_raises(self):
+        from app.schemas.ai_avatar_render import CreateAiAvatarRenderRequest
+
+        with pytest.raises(ValueError, match="lipsync_job_id 不能为空"):
             CreateAiAvatarRenderRequest(
-                lipsync_job_id="lipsync-1",
-                script_id="  ",
+                lipsync_job_id="  ",
+                script_id="script-1",
             )
 
 

tests/unit/test_lipsync_routes.py

@@ -35,14 +35,14 @@ def mock_mediakit():
 
 @pytest.fixture
 def mock_cosyvoice():
-    """Mock CosyVoice 服务."""
+    """Mock CosyVoice 服务（v3: service 内部走 submit_synthesize_task，返回 dict）."""
     service = MagicMock()
-    service.synthesize_speech.return_value = MagicMock(
-        audio_url="https://oss.example.com/tts-output.mp3",
-        duration=15.0,
-        file_size=12345,
-        request_id="tts-req-789",
-    )
+    service.submit_synthesize_task.return_value = {
+        "audio_url": "https://oss.example.com/tts-output.mp3",
+        "request_id": "tts-req-789",
+    }
+    # synthesize_speech 保留给直接同步调用场景
+    service.synthesize_speech.return_value = MagicMock(audio_url="https://oss.example.com/tts-output.mp3")
     return service
 
 
@@ -172,86 +172,121 @@ class TestSchemaValidation:
         )
         assert "?token=" in req.video_url
 
-    def test_no_audio_url_in_request(self):
-        """#1809: 请求体不应包含 audio_url 字段."""
+    def test_dual_mode_fields_present(self):
+        """v3 契约: 双模式——支持直接音频 audio_url，也支持 TTS 直生 voice_id+script_text."""
+        from app.schemas.lipsync import CreateLipsyncJobRequest
+
+        fields = CreateLipsyncJobRequest.model_fields.keys()
+        # 直接音频模式
+        assert "audio_url" in fields
+        # TTS 直生模式
+        assert "voice_id" in fields
+        assert "script_text" in fields
+        # 语速/情绪透传
+        assert "speed" in fields
+        assert "emotion" in fields
+
+    def test_direct_audio_mode_accepted(self):
+        """v3 契约: 只传 audio_url（直接音频模式）也合法，无需 voice_id/script_text."""
         from app.schemas.lipsync import CreateLipsyncJobRequest
 
         req = CreateLipsyncJobRequest(
             video_url="https://example.com/video.mp4",
-            voice_id="longxiaochun_v3",
-            script_text="测试文本",
+            audio_url="https://example.com/audio.mp3",
         )
-        assert not hasattr(req, "audio_url")
-        fields = req.model_fields.keys()
-        assert "audio_url" not in fields
-        assert "voice_id" in fields
-        assert "script_text" in fields
+        assert req.audio_url == "https://example.com/audio.mp3"
+
+    def test_neither_mode_rejected(self):
+        """v3 契约: audio_url 与 voice_id+script_text 都缺时应报错."""
+        from app.schemas.lipsync import CreateLipsyncJobRequest
+
+        with pytest.raises(ValueError):
+            CreateLipsyncJobRequest(video_url="https://example.com/video.mp4")
 
 
 class TestLipsyncServiceUnit:
     """Service 层单元测试（纯 mock，不依赖数据库）— #1809 更新."""
 
     def test_create_job_success(self, mock_mediakit, mock_cosyvoice):
+        """v3: TTS 直生——service 内部 submit_synthesize_task 合成后转存 OSS，再提交 MediaKit."""
         from app.services.lipsync_service import LipsyncService
 
         mock_db = MagicMock()
-        mock_db.add = MagicMock()
-        mock_db.flush = MagicMock()
-        mock_db.commit = MagicMock()
-        mock_db.refresh = MagicMock()
+        mock_repo = MagicMock()
+        mock_repo.get.return_value = None  # 预置音色，原样返回 voice_id
 
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        with (
+            patch("app.services.lipsync_service.get_shared_storage_service") as storage_patch,
+            patch("app.services.lipsync_service.safe_download_bytes") as dl_patch,
+        ):
+            storage_patch.return_value.upload_file.return_value = "https://my-oss/tts.mp3"
+            dl_patch.return_value = b"audio-bytes"
 
-        job = svc.create_job(
-            user_id="user-1",
-            video_url="https://example.com/video.mp4",
-            voice_id="longxiaochun_v3",
-            script_text="大家好，欢迎来到直播间",
-        )
+            svc = LipsyncService(
+                mock_db,
+                client=mock_mediakit,
+                cosyvoice_service=mock_cosyvoice,
+                voice_clone_repo=mock_repo,
+            )
+
+            job = svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="longxiaochun_v3",
+                script_text="大家好，欢迎来到直播间",
+                speed=1.2,
+                emotion="兴奋",
+            )
 
         assert job.status == "submitted"
         assert job.mediakit_task_id == "mk-task-123"
-        # TTS 应该被调用
-        mock_cosyvoice.synthesize_speech.assert_called_once_with(
-            text="大家好，欢迎来到直播间",
-            voice_id="longxiaochun_v3",
-        )
-        # MediaKit 应该用 TTS 生成的 audio_url
-        mock_mediakit.submit_lipsync.assert_called_once()
+        # TTS 直生走 submit_synthesize_task，带语速/情绪
+        mock_cosyvoice.submit_synthesize_task.assert_called_once()
+        _, kwargs = mock_cosyvoice.submit_synthesize_task.call_args
+        assert kwargs["text"] == "大家好，欢迎来到直播间"
+        assert kwargs["voice_id"] == "longxiaochun_v3"
+        assert kwargs["speed"] == 1.2
+        assert kwargs["emotion"] == "excited"  # 兴奋→excited
+        # job 记录透传字段
+        assert job.speed == 1.2
+        assert job.emotion == "excited"
+        # MediaKit 用转存后的 OSS audio_url
         call_kwargs = mock_mediakit.submit_lipsync.call_args
-        assert call_kwargs.kwargs["audio_url"] == "https://oss.example.com/tts-output.mp3"
+        assert call_kwargs.kwargs["audio_url"] == "https://my-oss/tts.mp3"
 
     def test_create_job_tts_failure(self, mock_mediakit):
-        """TTS 合成失败时，应创建 failed 记录并抛出 CosyVoiceError."""
+        """v3: TTS 合成失败时，CosyVoiceError 被包装为 MediaKitError(TTSSynthesisFailed)，
+        在建 DB 记录之前抛出，不提交 MediaKit。"""
         from app.services.lipsync_service import LipsyncService
+        from app.services.mediakit_client import MediaKitError
 
         from packages.application.cosyvoice_service import CosyVoiceError
 
         mock_cosyvoice = MagicMock()
-        mock_cosyvoice.synthesize_speech.side_effect = CosyVoiceError("TTS 服务不可用")
+        mock_cosyvoice.submit_synthesize_task.side_effect = CosyVoiceError("Arrearage 欠费")
 
         mock_db = MagicMock()
-        mock_db.add = MagicMock()
-        mock_db.flush = MagicMock()
-        mock_db.commit = MagicMock()
-        mock_db.refresh = MagicMock()
+        mock_repo = MagicMock()
+        mock_repo.get.return_value = None
 
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=mock_repo,
+        )
 
-        with pytest.raises(CosyVoiceError, match="TTS 服务不可用"):
+        with pytest.raises(MediaKitError) as exc_info:
             svc.create_job(
                 user_id="user-1",
                 video_url="https://example.com/video.mp4",
                 voice_id="longxiaochun_v3",
                 script_text="测试文本",
             )
+        assert exc_info.value.code == "TTSSynthesisFailed"
 
         # 不应提交到 MediaKit
         mock_mediakit.submit_lipsync.assert_not_called()
-        # 应该记录了失败状态
-        added_job = mock_db.add.call_args[0][0]
-        assert added_job.status == "failed"
-        assert "TTS" in added_job.error_message
 
     def test_create_job_api_failure(self, mock_mediakit, mock_cosyvoice):
         """MediaKit 提交失败."""
@@ -399,78 +434,165 @@ class TestLipsyncServiceUnit:
         assert result.status == "completed"
 
     def test_create_job_stores_tts_audio_url(self, mock_mediakit, mock_cosyvoice):
-        """#1809: 验证 job 的 audio_url 来自 TTS 合成结果."""
+        """v3: TTS 直生模式下 job.audio_url 为转存到自家 OSS 的永久地址."""
         from app.services.lipsync_service import LipsyncService
 
         mock_db = MagicMock()
-        mock_db.add = MagicMock()
-        mock_db.flush = MagicMock()
-        mock_db.commit = MagicMock()
-        mock_db.refresh = MagicMock()
+        mock_repo = MagicMock()
+        mock_repo.get.return_value = None
 
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        with (
+            patch("app.services.lipsync_service.get_shared_storage_service") as storage_patch,
+            patch("app.services.lipsync_service.safe_download_bytes") as dl_patch,
+        ):
+            storage_patch.return_value.upload_file.return_value = "https://my-oss/permanent.mp3"
+            dl_patch.return_value = b"audio-bytes"
 
+            svc = LipsyncService(
+                mock_db,
+                client=mock_mediakit,
+                cosyvoice_service=mock_cosyvoice,
+                voice_clone_repo=mock_repo,
+            )
+            job = svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="my-clone-voice",
+                script_text="这是一段测试文本",
+            )
+
+        # job.audio_url 是转存 OSS 后的永久地址
+        assert job.audio_url == "https://my-oss/permanent.mp3"
+
+    def test_create_job_direct_audio_skips_tts(self, mock_mediakit, mock_cosyvoice):
+        """v3: 直接音频模式（传 audio_url）不触发 TTS，原样把 audio_url 提交 MediaKit."""
+        from app.services.lipsync_service import LipsyncService
+
+        mock_db = MagicMock()
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=MagicMock(),
+        )
         job = svc.create_job(
             user_id="user-1",
             video_url="https://example.com/video.mp4",
-            voice_id="my-clone-voice",
-            script_text="这是一段测试文本",
+            audio_url="https://example.com/direct-audio.mp3",
         )
 
-        # job.audio_url 应该是 TTS 返回的 URL
-        assert job.audio_url == "https://oss.example.com/tts-output.mp3"
+        mock_cosyvoice.submit_synthesize_task.assert_not_called()
+        call_kwargs = mock_mediakit.submit_lipsync.call_args
+        assert call_kwargs.kwargs["audio_url"] == "https://example.com/direct-audio.mp3"
+        assert job.audio_url == "https://example.com/direct-audio.mp3"
 
 
 class TestErrorHandling:
-    """#1809 补充：错误返回 400 而非 500."""
+    """v3: 音色解析与错误码在 service 层处理，路由层做 HTTP 状态码映射."""
 
-    def test_voice_id_resolve_failure_returns_400(self, mock_mediakit, mock_cosyvoice):
-        """voice_clone_repo 查询异常时返回 400 而非 500."""
-        from app.api.routes.lipsync import _resolve_voice_id
-        from fastapi import HTTPException
+    def test_voice_id_resolve_forbidden(self, mock_mediakit, mock_cosyvoice):
+        """v3: 克隆音色属于他人时 service._resolve_voice_id 抛 VoiceForbidden（路由映射 403）."""
+        from app.services.lipsync_service import LipsyncService
+        from app.services.mediakit_client import MediaKitError
 
+        mock_db = MagicMock()
         mock_repo = MagicMock()
-        mock_repo.get.side_effect = Exception("DB connection error")
+        other_profile = MagicMock()
+        other_profile.user_id = "user-other"
+        other_profile.voice_id = "cv-voice-1"
+        mock_repo.get.return_value = other_profile
 
-        with pytest.raises(HTTPException) as exc_info:
-            _resolve_voice_id("bad-voice-id", "user-1", mock_repo)
-        assert exc_info.value.status_code == 400
-        assert "voice_id" in str(exc_info.value.detail)
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=mock_repo,
+        )
 
-    def test_create_job_value_error_returns_400(self, mock_mediakit):
-        """ValueError（参数无效）返回 400 而非 500."""
-        from app.services.lipsync_service import LipsyncService
-
-        mock_cosyvoice = MagicMock()
-        mock_cosyvoice.synthesize_speech.side_effect = ValueError("voice_id 为空")
-
-        mock_db = MagicMock()
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
-
-        # Service 层会 catch CosyVoiceError 但 ValueError 会穿透
-        # 路由层 catch ValueError → 400
-        with pytest.raises(ValueError):
+        with pytest.raises(MediaKitError) as exc_info:
             svc.create_job(
                 user_id="user-1",
                 video_url="https://example.com/video.mp4",
-                voice_id="",
-                script_text="test",
+                voice_id="clone-profile-id",
+                script_text="测试",
             )
+        assert exc_info.value.code == "VoiceForbidden"
+        mock_mediakit.submit_lipsync.assert_not_called()
 
-    def test_create_job_unexpected_exception_returns_400(self, mock_mediakit):
-        """未预期的异常应被路由层捕获返回 400 而非 500."""
+    def test_voice_id_resolve_not_ready(self, mock_mediakit, mock_cosyvoice):
+        """v3: 克隆音色尚未生成 voice_id 时抛 VoiceNotReady（路由映射 400）."""
         from app.services.lipsync_service import LipsyncService
-
-        mock_cosyvoice = MagicMock()
-        mock_cosyvoice.synthesize_speech.side_effect = RuntimeError("unexpected")
+        from app.services.mediakit_client import MediaKitError
 
         mock_db = MagicMock()
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        mock_repo = MagicMock()
+        profile = MagicMock()
+        profile.user_id = "user-1"
+        profile.voice_id = ""  # 克隆未完成
+        mock_repo.get.return_value = profile
 
-        with pytest.raises(RuntimeError):
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=mock_repo,
+        )
+
+        with pytest.raises(MediaKitError) as exc_info:
             svc.create_job(
                 user_id="user-1",
                 video_url="https://example.com/video.mp4",
-                voice_id="test-voice",
+                voice_id="clone-profile-id",
+                script_text="测试",
+            )
+        assert exc_info.value.code == "VoiceNotReady"
+
+    def test_tts_value_error_mapped_to_invalid_param(self, mock_mediakit):
+        """v3: CosyVoice 抛 ValueError（参数无效）被包装为 TTSInvalidParam（路由映射 400）."""
+        from app.services.lipsync_service import LipsyncService
+        from app.services.mediakit_client import MediaKitError
+
+        mock_cosyvoice = MagicMock()
+        mock_cosyvoice.submit_synthesize_task.side_effect = ValueError("voice_id 为空")
+
+        mock_db = MagicMock()
+        mock_repo = MagicMock()
+        mock_repo.get.return_value = None
+
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=mock_repo,
+        )
+
+        with pytest.raises(MediaKitError) as exc_info:
+            svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="some-voice",
                 script_text="test",
             )
+        assert exc_info.value.code == "TTSInvalidParam"
+
+    def test_missing_both_inputs_raises_invalid_input(self, mock_mediakit, mock_cosyvoice):
+        """v3: 既无 audio_url 又无 voice_id+script_text 时抛 InvalidInput（路由映射 400）."""
+        from app.services.lipsync_service import LipsyncService
+        from app.services.mediakit_client import MediaKitError
+
+        mock_db = MagicMock()
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=MagicMock(),
+        )
+
+        with pytest.raises(MediaKitError) as exc_info:
+            svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+            )
+        assert exc_info.value.code == "InvalidInput"
+        mock_cosyvoice.submit_synthesize_task.assert_not_called()
+        mock_mediakit.submit_lipsync.assert_not_called()

tests/unit/test_tts_preview.py

@@ -101,6 +101,7 @@ class TestTTSPreviewEndpoint:
             text="你好世界",
             voice_id="longxiaochun",
             speed=1.0,
+            emotion="",
         )
 
     def test_preview_with_speed(self):
@@ -146,6 +147,7 @@ class TestTTSPreviewEndpoint:
             text="测试",
             voice_id="v1",
             speed=1.5,
+            emotion="",
         )
 
     def test_preview_cosyvoice_error_returns_502(self):
@@ -330,6 +332,7 @@ class TestTTSPreviewEndpoint:
             text="克隆音色测试",
             voice_id="cosyvoice_actual_voice_123",
             speed=1.0,
+            emotion="",
         )
         # Verify repo was queried with the UUID
         mock_clone_repo.get.assert_called_once_with("abc123-uuid-of-profile")
@@ -406,6 +409,7 @@ class TestTTSPreviewEndpoint:
             text="预设音色测试",
             voice_id="longxiaoxia_v3",
             speed=1.0,
+            emotion="",
         )
 
     def test_preview_clone_voice_wrong_user_returns_403(self):