debug: verify staging post-deploy

.gitea/workflows/verify-staging.yml

@@ -0,0 +1,29 @@
+name: "Verify staging post-deploy"
+on:
+  push:
+    branches: [debug/verify-staging]
+  workflow_dispatch:
+jobs:
+  verify:
+    runs-on: runtime-builder
+    timeout-minutes: 10
+    steps:
+      - name: Setup SSH
+        shell: bash
+        env:
+          STAGING_SSH_KEY: ${{ secrets.PREVIEW_SSH_KEY }}
+        run: |
+          set -eux
+          which ssh || (apt-get update -qq && apt-get install -y -qq openssh-client)
+          mkdir -p ~/.ssh && chmod 700 ~/.ssh
+          printf "%s" "$STAGING_SSH_KEY" > ~/.ssh/id_rsa
+          chmod 600 ~/.ssh/id_rsa
+          H=47.98.113.167; P=22222
+          ssh-keyscan -p $P -H $H >> ~/.ssh/known_hosts 2>/dev/null
+      - name: Run verify
+        shell: bash
+        run: |
+          set -x
+          echo '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' | base64 -d > /tmp/verify.sh
+          chmod +x /tmp/verify.sh
+          ssh -p 22222 -i ~/.ssh/id_rsa -o StrictHostKeyChecking=no root@47.98.113.167 'bash -s' < /tmp/verify.sh

alembic/versions/075_add_sentence_timings.py

@@ -1,27 +0,0 @@
-"""add sentence_timings to lipsync_jobs
-
-Revision ID: 075_add_sentence_timings
-Revises: 074_ai_avatar_render_script_id_optional
-Create Date: 2026-09-12
-"""
-
-import sqlalchemy as sa
-
-from alembic import op
-
-revision = "075_add_sentence_timings"
-down_revision = "074_render_script_id_optional"
-branch_labels = None
-depends_on = None
-
-
-def upgrade() -> None:
-    with op.batch_alter_table("lipsync_jobs") as batch:
-        batch.add_column(
-            sa.Column("sentence_timings", sa.JSON(), nullable=True),
-        )
-
-
-def downgrade() -> None:
-    with op.batch_alter_table("lipsync_jobs") as batch:
-        batch.drop_column("sentence_timings")

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

@@ -11,7 +11,6 @@
 from __future__ import annotations
 
 import logging
-from datetime import datetime, timezone
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.dependencies import get_db_session
@@ -78,16 +77,10 @@ def create_render_job(
         from app.tasks.ai_avatar_render import execute_ai_avatar_render
 
         execute_ai_avatar_render.delay(job.id)
-    except Exception as exc:
-        logger.exception("Celery 任务投递失败（创建）: job_id=%s err=%s", job.id, exc)
-        job.status = "failed"
-        job.error_message = f"任务提交失败：{exc}"
-        job.updated_at = datetime.now(timezone.utc)
-        svc.db.commit()
-        svc.db.refresh(job)
-        return AiAvatarRenderJobResponse.model_validate(job)
+    except Exception:
+        logger.warning("Celery 任务提交失败，渲染任务已创建但未触发执行: %s", job.id)
 
-    return AiAvatarRenderJobResponse.model_validate(job)
+    return job
 
 
 # ── GET /jobs — 任务列表 ─────────────────────────────────────────────────
@@ -179,16 +172,10 @@ def retry_render_job(
         from app.tasks.ai_avatar_render import execute_ai_avatar_render
 
         execute_ai_avatar_render.delay(job.id)
-    except Exception as exc:
-        logger.exception("Celery 任务投递失败（重试）: job_id=%s err=%s", job.id, exc)
-        job.status = "failed"
-        job.error_message = f"任务提交失败：{exc}"
-        job.updated_at = datetime.now(timezone.utc)
-        svc.db.commit()
-        svc.db.refresh(job)
-        return AiAvatarRenderJobResponse.model_validate(job)
+    except Exception:
+        logger.warning("Celery 任务提交失败，重试任务已重置但未触发执行: %s", job.id)
 
-    return AiAvatarRenderJobResponse.model_validate(job)
+    return job
 
 
 
@@ -212,11 +199,7 @@ def generate_avatar_smart_cover(
         raise HTTPException(status_code=400, detail="video_url 必须是合法的 HTTP/HTTPS URL")
 
     try:
-        cover_url = generate_smart_cover(
-            video_url,
-            max_frames=body.max_frames,
-            title_config=getattr(body, "title_config", None),
-        )
+        cover_url = generate_smart_cover(video_url, max_frames=body.max_frames)
     except Exception as exc:
         logger.error(
             "智能封面生成异常: user=%s video_url=%s err=%s",

apps/api/app/schemas/ai_avatar_render.py

@@ -110,13 +110,10 @@ class AiAvatarRenderProgressResponse(BaseModel):
 
 
 class SmartCoverRequest(BaseModel):
-    """智能封面请求 — MediaKit 抽帧 + 质量评分选最佳帧 + 可选标题 drawtext 叠加."""
+    """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）")
-    title_config: Optional[dict[str, Any]] = Field(
-        None, description="标题配置；传入时在封面上用 drawtext 叠加标题（竖屏 720x1280）"
-    )
 
 
 class SmartCoverResponse(BaseModel):

apps/api/app/schemas/lipsync.py

@@ -33,7 +33,6 @@ class LipsyncJobResponse(BaseModel):
     output_duration: float
     error_message: str
     error_code: str
-    sentence_timings: Optional[list] = None
     submitted_at: Optional[datetime] = None
     completed_at: Optional[datetime] = None
     created_at: datetime

apps/api/app/services/ai_avatar_cover_service.py

@@ -11,8 +11,6 @@
 from __future__ import annotations
 
 import logging
-import os
-import subprocess
 import tempfile
 import uuid
 from pathlib import Path
@@ -21,9 +19,9 @@ from urllib.parse import urlparse
 
 logger = logging.getLogger(__name__)
 
-# MediaKit 抽帧轮询参数：poll_interval=1s × max_poll=15 → 最长 15s，配合前端 120s 超时足够
-COVER_POLL_INTERVAL = 1.0
-COVER_MAX_POLL_ATTEMPTS = 15
+# MediaKit 抽帧轮询参数（与 MediaKit API timeout=60s 对齐）
+COVER_POLL_INTERVAL = 3.0
+COVER_MAX_POLL_ATTEMPTS = 20  # 最多等 60 秒
 
 # 帧图片下载超时（秒）
 FRAME_DOWNLOAD_TIMEOUT = 20
@@ -158,79 +156,13 @@ def select_best_cover_frame(video_url: str, *, max_frames: int = 5) -> str:
         return ""
 
 
-def apply_title_to_cover(local_frame: str, *, title_config: dict | None) -> str:
-    """用 ffmpeg drawtext 在封面图上叠加标题，返回叠加后图片的本地路径.
-
-    ffmpeg 失败时回退返回原始 local_frame。竖屏封面按 720x1280 计算位置。
-    """
-    if not title_config or not isinstance(title_config, dict):
-        return local_frame
-    text = (title_config.get("text") or title_config.get("content") or "").strip()
-    if not text:
-        return local_frame
-    enabled = title_config.get("enabled", True)
-    if not enabled:
-        return local_frame
-
-    try:
-        from packages.domain.video_filter_builder import build_title_drawtext_filter
-
-        drawtext_filter = build_title_drawtext_filter(
-            title_config,
-            output_width=720,
-            output_height=1280,
-        )
-        if not drawtext_filter:
-            return local_frame
-
-        base, ext = os.path.splitext(local_frame)
-        titled_path = f"{base}_titled{ext or '.jpg'}"
-        cmd = [
-            "ffmpeg",
-            "-i",
-            local_frame,
-            "-vf",
-            drawtext_filter,
-            "-y",
-            titled_path,
-        ]
-        logger.info("[数字人封面] 叠加标题: text=%s", text[:30])
-        result = subprocess.run(
-            cmd,
-            capture_output=True,
-            text=True,
-            timeout=30,
-        )
-        if result.returncode != 0:
-            logger.warning(
-                "[数字人封面] drawtext 失败，回退无标题: exit=%s stderr=%s",
-                result.returncode,
-                (result.stderr or "")[-300:],
-            )
-            return local_frame
-        if not os.path.exists(titled_path) or os.path.getsize(titled_path) == 0:
-            logger.warning("[数字人封面] drawtext 输出为空，回退无标题")
-            return local_frame
-        return titled_path
-    except Exception as exc:
-        logger.warning("[数字人封面] 标题叠加异常，回退无标题: %s", exc, exc_info=True)
-        return local_frame
-
-
-def persist_cover_to_oss(
-    frame_url: str,
-    *,
-    job_id: str = "",
-    prefix: str = "ai-avatar/covers",
-    title_config: dict | None = None,
-) -> str:
+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
-        title_config: 可选标题配置；传入时用 drawtext 叠加标题（竖屏 720x1280）
 
     Returns:
         OSS 公网 URL；失败回退原始 frame_url
@@ -238,7 +170,6 @@ def persist_cover_to_oss(
     if not frame_url:
         return ""
     tmp_path: Optional[str] = None
-    titled_path: Optional[str] = None
     try:
         import httpx
 
@@ -258,21 +189,12 @@ def persist_cover_to_oss(
         storage = get_shared_storage_service()
         token = job_id or uuid.uuid4().hex[:12]
         cover_key = f"{prefix}/{token}/cover_{uuid.uuid4().hex[:8]}.jpg"
-
-        upload_path = apply_title_to_cover(tmp_path, title_config=title_config)
-        if upload_path != tmp_path:
-            titled_path = upload_path
-
         public_url = storage.upload_file(
-            file_or_path=upload_path,
+            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 titled=%s",
-            cover_key,
-            bool(titled_path),
-        )
+        logger.info("[数字人封面] 封面已转存 OSS: key=%s", cover_key)
         # 私有桶：返回预签名 URL（前端才能加载）
         if public_url:
             signed = storage.get_download_url(cover_key, expires_seconds=86400)
@@ -282,32 +204,19 @@ def persist_cover_to_oss(
         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:
-        for p in (tmp_path, titled_path):
-            if p:
-                try:
-                    Path(p).unlink(missing_ok=True)
-                except Exception:
-                    pass
+        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,
-    title_config: dict | None = None,
-) -> str:
-    """一站式：MediaKit 智能抽帧选最佳 → （可选）drawtext 叠加标题 → 转存 OSS.
+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。
-
-    Args:
-        video_url: 可公网访问的视频 URL
-        job_id: 关联任务 ID
-        max_frames: 抽帧数量
-        title_config: 可选标题配置；传入时在封面上叠加 drawtext 标题（竖屏 720x1280）
     """
     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, title_config=title_config)
+    return persist_cover_to_oss(best_frame, job_id=job_id)

apps/api/app/services/ai_avatar_render_service.py

@@ -11,7 +11,6 @@ from __future__ import annotations
 
 import logging
 import os
-import subprocess
 import tempfile
 import uuid
 from datetime import datetime, timezone
@@ -25,7 +24,7 @@ from packages.adapters.sqlalchemy_impl.models import (
     ScriptModel,
 )
 from packages.domain.video_filter_builder import (
-    build_broll_overlay_filter,
+    build_cover_extract_command,
     build_title_drawtext_filter,
 )
 from packages.shared.storage import get_shared_storage_service
@@ -229,49 +228,27 @@ class AiAvatarRenderService:
             self.db.commit()
 
             # 2. 构建 FFmpeg 滤镜链 (40%)
-            # 用 ffprobe 探测输入视频分辨率，确保 B-roll 缩放与标题位置与实际输出一致。
-            # AI 数字人对口型输出为 9:16 竖屏，默认兜底 720x1280；探测失败时使用默认值不阻断渲染。
-            output_width, output_height = self._probe_video_resolution(input_video_path)
-            if output_width <= 0 or output_height <= 0:
-                output_width, output_height = 720, 1280
-                logger.info(
-                    "[数字人渲染] ffprobe 探测分辨率失败或无效，使用默认竖屏尺寸 %sx%s",
-                    output_width,
-                    output_height,
-                )
-            else:
-                logger.info("[数字人渲染] 探测输入视频分辨率: %sx%s", output_width, output_height)
+            from packages.domain.video_filter_builder import build_broll_overlay_filter
 
-            broll_filter, broll_label = build_broll_overlay_filter(
+            filter_complex = build_broll_overlay_filter(
                 b_roll_segments=job.b_roll_segments,
                 video_duration=lipsync_job.output_duration,
-                output_width=output_width,
-                output_height=output_height,
             )
 
-            # 标题叠加（传入实际输出尺寸，保证位置计算正确）
-            title_filter = build_title_drawtext_filter(
-                job.title_config,
-                output_width=output_width,
-                output_height=output_height,
-            )
+            # 标题叠加
+            title_filter = build_title_drawtext_filter(job.title_config)
+            if title_filter:
+                if filter_complex:
+                    filter_complex += f"[vout]{title_filter}[vout_titled];"
+                else:
+                    filter_complex = f"[0:v]{title_filter}[vout_titled];"
 
-            filter_complex = ""
-            final_label = None
-            if broll_filter and title_filter:
-                # B-roll → 标题叠在 B-roll 输出上
-                filter_complex = broll_filter + f";[{broll_label}]{title_filter}[vout_titled]"
-                final_label = "vout_titled"
-            elif broll_filter:
-                filter_complex = broll_filter
-                final_label = broll_label
-            elif title_filter:
-                filter_complex = f"[0:v]{title_filter}[vout_titled]"
-                final_label = "vout_titled"
-            else:
-                # 无滤镜：直接拷贝视频流
-                filter_complex = ""
-                final_label = None
+            # 清理末尾分号
+            if filter_complex.endswith(";"):
+                filter_complex = filter_complex[:-1]
+
+            # 最终输出标签
+            final_label = "vout_titled" if title_filter else ("vout" if filter_complex else None)
 
             job.progress = 40
             self.db.commit()
@@ -280,7 +257,7 @@ class AiAvatarRenderService:
             with tempfile.TemporaryDirectory() as tmpdir:
                 output_video_path = os.path.join(tmpdir, "output.mp4")
 
-                cmd_list = self._build_ffmpeg_command(
+                cmd = self._build_ffmpeg_command(
                     input_video=input_video_path,
                     b_roll_segments=job.b_roll_segments,
                     filter_complex=filter_complex,
@@ -288,25 +265,9 @@ class AiAvatarRenderService:
                     output_path=output_video_path,
                 )
 
-                try:
-                    render_result = subprocess.run(
-                        cmd_list,
-                        capture_output=True,
-                        text=True,
-                        timeout=600,
-                    )
-                except subprocess.TimeoutExpired as exc:
-                    raise AiAvatarRenderError(
-                        "FFmpeg 渲染超时（600s）",
-                        code="FFmpegTimeout",
-                    ) from exc
-
-                if render_result.returncode != 0:
-                    stderr_tail = (render_result.stderr or "").strip()[-800:]
-                    raise AiAvatarRenderError(
-                        f"FFmpeg 渲染失败，退出码: {render_result.returncode}, stderr: {stderr_tail}",
-                        code="FFmpegFailed",
-                    )
+                exit_code = os.system(cmd)
+                if exit_code != 0:
+                    raise AiAvatarRenderError(f"FFmpeg 渲染失败，退出码: {exit_code}", code="FFmpegFailed")
 
                 job.progress = 80
                 self.db.commit()
@@ -315,27 +276,11 @@ class AiAvatarRenderService:
                 cover_path = ""
                 if job.cover_config:
                     cover_path = os.path.join(tmpdir, "cover.jpg")
-                    cover_cmd = self._build_cover_extract_cmd(
-                        cover_config=job.cover_config,
-                        input_video=output_video_path,
-                        output_path=cover_path,
-                    )
-                    try:
-                        cover_result = subprocess.run(
-                            cover_cmd,
-                            capture_output=True,
-                            text=True,
-                            timeout=60,
-                        )
-                        if cover_result.returncode != 0:
-                            logger.warning(
-                                "封面提取失败（非致命），跳过: exit=%s stderr=%s",
-                                cover_result.returncode,
-                                (cover_result.stderr or "")[-300:],
-                            )
-                            cover_path = ""
-                    except Exception as cover_err:
-                        logger.warning("封面提取异常（非致命），跳过: %s", cover_err)
+                    cover_cmd = build_cover_extract_command(job.cover_config, cover_path)
+                    cover_cmd = cover_cmd.replace("INPUT_VIDEO", output_video_path)
+                    cover_exit = os.system(cover_cmd)
+                    if cover_exit != 0:
+                        logger.warning("封面提取失败，跳过: %s", cover_cmd)
                         cover_path = ""
 
                 job.progress = 90
@@ -345,7 +290,7 @@ 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
 
-                # 封面：优先复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧（支持 drawtext 标题叠加）；
+                # 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:
@@ -354,13 +299,7 @@ class AiAvatarRenderService:
                             generate_smart_cover,
                         )
 
-                        smart_cover_url = generate_smart_cover(
-                            output_video_url,
-                            job_id=job_id,
-                            max_frames=5,
-                            # 注意：不传 title_config —— 最终输出视频已经通过 drawtext 叠加了标题，
-                            # 再传会导致封面标题双重叠加
-                        )
+                        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)
 
@@ -392,13 +331,9 @@ class AiAvatarRenderService:
                     from packages.domain.generated_video import GeneratedVideo
 
                     clip_name = f"AI数字人_{job_id[:8]}"
-                    # AI数字人入口是独立页面，前端可能不传 project_id（无项目概念），
-                    # 兜底为 "ai_avatar" 避免 DB 非空约束/查询问题；generation_task_id 同样兜底用 render_job_id
-                    clip_project_id = (job.project_id or "").strip() or "ai_avatar"
-                    clip_generation_task_id = (job.lipsync_job_id or "").strip() or job_id
                     clip = GeneratedVideo.create(
-                        project_id=clip_project_id,
-                        generation_task_id=clip_generation_task_id,
+                        project_id=job.project_id,
+                        generation_task_id=job.lipsync_job_id,
                         name=clip_name,
                         file_url=job.output_video_url,
                         user_id=job.user_id,
@@ -412,11 +347,11 @@ class AiAvatarRenderService:
                     video_repo = SQLAlchemyGeneratedVideoRepository(self.db)
                     video_repo.create(clip)
                     logger.info("成片记录已保存到成片库: clip_id=%s, render_job=%s", clip.id, job_id)
-                except Exception:
-                    logger.error(
-                        "自动保存成片记录失败（不影响渲染任务状态）: render_job=%s",
+                except Exception as clip_err:
+                    logger.warning(
+                        "自动保存成片记录失败（不影响渲染任务状态）: render_job=%s, error=%s",
                         job_id,
-                        exc_info=True,
+                        clip_err,
                     )
 
         except AiAvatarRenderError as exc:
@@ -425,14 +360,12 @@ class AiAvatarRenderService:
             job.updated_at = datetime.now(timezone.utc)
             self.db.commit()
             logger.error("渲染任务失败 [%s]: %s", job_id, exc)
-            raise
         except Exception as exc:
             job.status = "failed"
             job.error_message = f"渲染异常: {str(exc)}"
             job.updated_at = datetime.now(timezone.utc)
             self.db.commit()
             logger.exception("渲染任务异常 [%s]", job_id)
-            raise
 
     def _download_video(self, url: str) -> str:
         """下载视频到临时文件."""
@@ -450,37 +383,6 @@ class AiAvatarRenderService:
                 os.unlink(tmp.name)
             raise
 
-    @staticmethod
-    def _probe_video_resolution(video_path: str) -> tuple[int, int]:
-        """用 ffprobe 探测视频分辨率，返回 (width, height)；失败返回 (0, 0)。"""
-        try:
-            result = subprocess.run(
-                [
-                    "ffprobe",
-                    "-v",
-                    "error",
-                    "-select_streams",
-                    "v:0",
-                    "-show_entries",
-                    "stream=width,height",
-                    "-of",
-                    "csv=p=0:s=x",
-                    video_path,
-                ],
-                capture_output=True,
-                text=True,
-                timeout=15,
-            )
-            if result.returncode == 0 and result.stdout.strip():
-                parts = result.stdout.strip().split("x")
-                if len(parts) == 2:
-                    w, h = int(parts[0]), int(parts[1])
-                    if w > 0 and h > 0:
-                        return w, h
-        except Exception as exc:
-            logger.warning("[数字人渲染] ffprobe 探测分辨率失败: %s", exc)
-        return 0, 0
-
     def _build_ffmpeg_command(
         self,
         *,
@@ -489,86 +391,24 @@ class AiAvatarRenderService:
         filter_complex: str,
         final_label: Optional[str],
         output_path: str,
-    ) -> list[str]:
-        """构建 FFmpeg 命令（list 形式，shell=False）.
-
-        根因修复 #1798 P0：OSS 预签名 URL 含 `&Expires=...&Signature=...` 特殊字符，
-        os.system(shell=True) 会把 `&` 解释为后台命令分隔符，导致 -filter_complex 被
-        当成独立命令报 sh: -filter_complex: not found（exit 127 → Python 32512）。
-        list + shell=False 彻底规避 shell 转义问题。
-        """
-        cmd: list[str] = ["ffmpeg", "-i", input_video]
+    ) -> str:
+        """构建 FFmpeg 命令."""
+        # 输入文件
+        inputs = f"-i {input_video}"
         for seg in b_roll_segments:
             asset_url = seg.get("asset_url", "")
             if asset_url:
-                cmd.extend(["-i", asset_url])
+                inputs += f" -i {asset_url}"
 
+        # 滤镜
         if filter_complex and final_label:
-            cmd.extend(
-                [
-                    "-filter_complex",
-                    filter_complex,
-                    "-map",
-                    f"[{final_label}]",
-                    "-map",
-                    "0:a?",
-                ]
-            )
+            filter_arg = f'-filter_complex "{filter_complex}" -map "[{final_label}]"'
         elif filter_complex:
-            cmd.extend(["-filter_complex", filter_complex])
-
-        cmd.extend(
-            [
-                "-c:v",
-                "libx264",
-                "-preset",
-                "veryfast",
-                "-crf",
-                "23",
-                "-c:a",
-                "aac",
-                "-b:a",
-                "128k",
-                "-y",
-                output_path,
-            ]
-        )
-        return cmd
-
-    def _build_cover_extract_cmd(
-        self,
-        *,
-        cover_config: dict[str, Any],
-        input_video: str,
-        output_path: str,
-    ) -> list[str]:
-        """构建封面截帧 FFmpeg 命令（list 形式，shell=False）."""
-        if not cover_config or not isinstance(cover_config, dict):
-            timestamp = 0.0
-            width = 0
-            height = 0
+            filter_arg = f'-filter_complex "{filter_complex}"'
         else:
-            timestamp = cover_config.get("timestamp", 0.0)
-            width = cover_config.get("width", 0)
-            height = cover_config.get("height", 0)
+            filter_arg = ""
 
-        cmd: list[str] = [
-            "ffmpeg",
-            "-ss",
-            str(timestamp),
-            "-i",
-            input_video,
-            "-frames:v",
-            "1",
-        ]
-        if width > 0 and height > 0:
-            vf = (
-                f"scale={width}:{height}:force_original_aspect_ratio=decrease,"
-                f"pad={width}:{height}:(ow-iw)/2:(oh-ih)/2"
-            )
-            cmd.extend(["-vf", vf])
-        cmd.extend(["-y", output_path])
-        return cmd
+        return f"ffmpeg {inputs} {filter_arg} -c:v libx264 -preset veryfast -crf 23 -y {output_path}"
 
     def _upload_to_oss(self, local_path: str, oss_key: str) -> str:
         """上传文件到 OSS，返回 URL.

apps/api/app/tasks/lipsync_tts.py

@@ -54,160 +54,6 @@ def _sign_media_url(url: str) -> str:
         return url
 
 
-def _split_script_into_sentences(script_text: str) -> list[str]:
-    """按句号/问号/感叹号/分号/换行分句（与前端 splitScriptIntoSentences 一致）."""
-    import re
-
-    text = (script_text or "").strip()
-    if not text:
-        return []
-    parts = re.split(r"[。！？!?；;\n\r]+", text)
-    return [p.strip() for p in parts if p.strip()]
-
-
-def _compute_sentence_timings(audio_data: bytes, script_text: str, total_duration: float) -> list[dict]:
-    """基于 TTS 音频的静音检测，精确计算每句文案的起止时间.
-
-    使用 ffmpeg silencedetect 检测静音段，将静音点与句子边界对齐。
-    比字数比例估算准确得多。
-
-    Args:
-        audio_data: TTS 音频二进制数据（MP3）
-        script_text: 文案全文
-        total_duration: 音频总时长（秒）
-
-    Returns:
-        list[{"index": int, "text": str, "start_time": float, "end_time": float}]
-    """
-    import re
-    import subprocess
-    import tempfile
-
-    sentences = _split_script_into_sentences(script_text)
-    if not sentences:
-        return []
-
-    # 写入临时音频文件
-    with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as tmp:
-        tmp.write(audio_data)
-        tmp_path = tmp.name
-
-    try:
-        # 用 ffmpeg silencedetect 检测静音段
-        result = subprocess.run(
-            [
-                "ffmpeg",
-                "-i",
-                tmp_path,
-                "-af",
-                "silencedetect=noise=-25dB:d=0.3",
-                "-f",
-                "null",
-                "-",
-            ],
-            capture_output=True,
-            text=True,
-            timeout=30,
-        )
-        stderr = result.stderr or ""
-
-        # 解析静音结束时间点（silence_end: X.XXX）
-        silence_ends = []
-        for match in re.finditer(r"silence_end:\s*([\d.]+)", stderr):
-            t = float(match.group(1))
-            if 0 < t < total_duration:
-                silence_ends.append(t)
-
-        # 如果没有检测到足够的静音点，降级为字数比例估算
-        if len(silence_ends) < len(sentences) - 1:
-            logger.warning(
-                "[sentence_timings] 静音点不足(%d < %d)，降级为字数比例估算",
-                len(silence_ends),
-                len(sentences) - 1,
-            )
-            return _estimate_sentence_timings_by_chars(sentences, total_duration)
-
-        # 贪心匹配：N-1 个句子边界对应 N-1 个静音点
-        # 按时间均匀分布期望值，选择最近的静音点
-        n_boundaries = len(sentences) - 1
-        boundaries = []
-        used_indices = set()
-
-        for i in range(n_boundaries):
-            # 期望的边界位置（按句子数量均匀分布）
-            expected_pos = (i + 1) / len(sentences) * total_duration
-            # 找最近的未使用静音点
-            best_idx = None
-            best_dist = float("inf")
-            for j, t in enumerate(silence_ends):
-                if j in used_indices:
-                    continue
-                dist = abs(t - expected_pos)
-                if dist < best_dist:
-                    best_dist = dist
-                    best_idx = j
-            if best_idx is not None:
-                used_indices.add(best_idx)
-                boundaries.append(silence_ends[best_idx])
-
-        boundaries.sort()
-
-        # 构建 sentence_timings
-        timings = []
-        prev_end = 0.0
-        for i, sent in enumerate(sentences):
-            start = prev_end
-            end = boundaries[i] if i < len(boundaries) else total_duration
-            timings.append(
-                {
-                    "index": i,
-                    "text": sent,
-                    "start_time": round(start, 2),
-                    "end_time": round(end, 2),
-                }
-            )
-            prev_end = end
-
-        return timings
-
-    except Exception as exc:
-        logger.warning("[sentence_timings] 静音检测异常，降级为字数比例估算: %s", exc)
-        return _estimate_sentence_timings_by_chars(sentences, total_duration)
-    finally:
-        import os
-
-        try:
-            os.unlink(tmp_path)
-        except Exception:
-            pass
-
-
-def _estimate_sentence_timings_by_chars(sentences: list[str], total_duration: float) -> list[dict]:
-    """降级方案：按字数比例估算句子时间（与原前端逻辑一致）."""
-    if not sentences or total_duration <= 0:
-        return []
-    total_chars = sum(len(s.replace(r"\s", "")) for s in sentences)
-    if total_chars == 0:
-        return []
-
-    timings = []
-    acc = 0
-    for i, sent in enumerate(sentences):
-        chars = len(sent.replace(r"\s", ""))
-        start = (acc / total_chars) * total_duration
-        end = ((acc + chars) / total_chars) * total_duration
-        timings.append(
-            {
-                "index": i,
-                "text": sent,
-                "start_time": round(start, 2),
-                "end_time": round(end, 2),
-            }
-        )
-        acc += chars
-    return timings
-
-
 @shared_task(
     bind=True,
     name="lipsync_tts.synthesize_and_submit",
@@ -306,14 +152,13 @@ def tts_synthesize_and_submit(
             audio_data = safe_download_bytes(
                 temp_url,
                 purpose="lipsync_tts_audio",
-                allowed_mime_types={
+                allowed_mime_types=(
                     "audio/mpeg",
                     "audio/mp3",
                     "audio/wav",
-                    "audio/x-wav",  # CosyVoice 部分接口返回 audio/x-wav，与 audio/wav 等价（RIFF/WAVE）
                     "audio/mp4",
                     "audio/x-m4a",
-                },
+                ),
                 timeout=60.0,
             )
             from packages.shared.storage import get_shared_storage_service
@@ -333,64 +178,6 @@ def tts_synthesize_and_submit(
 
         db.commit()
 
-        # 2.5 计算精确句子时间戳（基于 TTS 音频静音检测）
-        import os as _os
-
-        _st_tmp_path = None
-        try:
-            import subprocess as _sp
-            import tempfile as _tmpf
-
-            # 下载音频用于探测时长和静音检测
-            if isinstance(job.audio_url, str) and job.audio_url:
-                from packages.shared.url_security import safe_download_bytes as _sdl
-
-                _audio_bytes = _sdl(job.audio_url, purpose="sentence_timings", timeout=30.0)
-            else:
-                _audio_bytes = audio_data
-
-            # ffprobe 获取音频时长
-            with _tmpf.NamedTemporaryFile(suffix=".mp3", delete=False) as _atmp:
-                _atmp.write(_audio_bytes)
-                _st_tmp_path = _atmp.name
-
-            _probe_result = _sp.run(
-                [
-                    "ffprobe",
-                    "-v",
-                    "error",
-                    "-show_entries",
-                    "format=duration",
-                    "-of",
-                    "default=noprint_wrappers=1:nokey=1",
-                    _st_tmp_path,
-                ],
-                capture_output=True,
-                text=True,
-                timeout=10,
-            )
-            _audio_duration = float(_probe_result.stdout.strip()) if _probe_result.stdout.strip() else 0.0
-
-            if _audio_duration > 0:
-                _timings = _compute_sentence_timings(_audio_bytes, script_text, _audio_duration)
-                if _timings:
-                    job.sentence_timings = _timings
-                    logger.info(
-                        "[lipsync_tts] 句子时间戳已计算: job_id=%s sentences=%d duration=%.1f",
-                        job_id,
-                        len(_timings),
-                        _audio_duration,
-                    )
-            db.commit()
-        except Exception as _st_err:
-            logger.warning("[lipsync_tts] 句子时间戳计算失败（不影响主流程）: job_id=%s err=%s", job_id, _st_err)
-        finally:
-            if _st_tmp_path:
-                try:
-                    _os.unlink(_st_tmp_path)
-                except Exception:
-                    pass
-
         # 3. 签名 URL 并提交到 MediaKit（复用模块内 _sign_media_url，避免对 LipsyncService 的耦合）
         audio_url = _sign_media_url(job.audio_url)
         video_url = _sign_media_url(job.video_url)

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

@@ -1,6 +1,6 @@
 /**
  * 成品 / 视频相关 API 函数
- * 后端实际接口：/videos（分页：page/page_size，返回 {items, total, page, page_size}）
+ * 后端实际接口：/videos
  */
 import apiClient from "../client"
 import type {
@@ -12,39 +12,16 @@ import type {
 } from "./types"
 import { mapVideoToProductItem } from "./utils"
 
-/** 分页列表响应（前端消费用） */
-export interface ProductListResult {
-  items: ProductItem[]
-  total: number
-  page: number
-  page_size: number
-}
-
-/**
- * 获取成品列表（分页）
- * @param params 分页与筛选参数：page 默认 1，page_size 默认 20
- */
-export const getProducts = async (params?: ProductListParams): Promise<ProductListResult> => {
-  const response = await apiClient.get("/videos", {
-    params: {
-      page: 1,
-      page_size: 20,
-      ...params,
-    },
-  })
-  const data = response.data as {
-    items?: VideoItem[]
-    total?: number
-    page?: number
-    page_size?: number
-  }
-  const items: VideoItem[] = Array.isArray(data?.items) ? data.items : []
-  return {
-    items: items.map(mapVideoToProductItem),
-    total: data.total ?? items.length,
-    page: data.page ?? params?.page ?? 1,
-    page_size: data.page_size ?? params?.page_size ?? 20,
-  }
+/** 获取成品列表（支持分页和筛选） */
+export const getProducts = async (params?: ProductListParams): Promise<ProductItem[]> => {
+  const response = await apiClient.get("/videos", { params })
+  const data = response.data
+  const videos: VideoItem[] = Array.isArray(data?.items)
+    ? data.items
+    : Array.isArray(data)
+      ? data
+      : []
+  return videos.map(mapVideoToProductItem)
 }
 
 /** 获取单个成品详情 */

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

@@ -552,7 +552,7 @@
   max-width: 240px;
   aspect-ratio: 9/16;
   background: #f0f0f5;
-  border-radius: 12px;
+  border-radius: 8px;
   overflow: hidden;
   display: flex;
   align-items: center;
@@ -564,10 +564,8 @@
 .aa-cover-preview img {
   width: 100%;
   height: 100%;
-  aspect-ratio: 9/16;
   object-fit: cover;
   display: block;
-  border-radius: 12px;
 }
 
 .aa-cover-preview__placeholder {
@@ -575,19 +573,6 @@
   color: #8c8ca1;
 }
 
-.aa-cover-preview__loading {
-  position: absolute;
-  inset: 0;
-  display: flex;
-  align-items: center;
-  justify-content: center;
-  background: rgba(0, 0, 0, 0.45);
-  color: #fff;
-  font-size: 13px;
-  backdrop-filter: blur(4px);
-  -webkit-backdrop-filter: blur(4px);
-}
-
 .aa-cover-actions {
   display: flex;
   gap: 8px;

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

@@ -25,7 +25,6 @@ import {
   getRenderJob,
   generateSmartCover,
 } from "./api/aiAvatar"
-import { getOrCreateDefaultProject } from "@/api/projects"
 import {
   normalizeEmotion,
   buildTitleConfigPayload,
@@ -207,12 +206,9 @@ const AiAvatarPage: React.FC = () => {
     }
     state.setIsGenerating(true)
     try {
-      // 确保有 project_id（AI数字人入口独立，不在项目内，自动取默认项目；#1860 P0 bugfix）
-      const defaultProject = await getOrCreateDefaultProject()
       const job = await submitRender({
         lipsync_job_id: state.lipsyncJob.id,
         script_id: state.script?.id,
-        project_id: defaultProject.id,
         b_roll_segments: state.bRollSegments.map((seg) => ({
           script_segment_index: seg.script_segment_index,
           asset_url: seg.asset.file_url || "",
@@ -241,15 +237,6 @@ const AiAvatarPage: React.FC = () => {
             if (renderTimerRef.current) clearInterval(renderTimerRef.current)
             renderTimerRef.current = null
             setRenderStatus("completed")
-            // 渲染完成后，用最终视频的封面更新前端封面配置
-            if (updated.output_cover_url) {
-              state.setCoverConfig((prev) => ({
-                ...prev,
-                mode: "auto_frame",
-                smart_cover_url: updated.output_cover_url,
-                thumbnail_url: updated.output_cover_url,
-              }))
-            }
             message.success("视频已生成并保存到成片库")
           } else if (updated.status === "failed") {
             if (renderTimerRef.current) clearInterval(renderTimerRef.current)
@@ -292,7 +279,7 @@ const AiAvatarPage: React.FC = () => {
     }
     setSmartCoverLoading(true)
     try {
-      const res = await generateSmartCover(videoUrl, buildTitleConfigPayload(state.titleConfig), 5)
+      const res = await generateSmartCover(videoUrl, 5)
       if (res.cover_url) {
         state.setCoverConfig((prev) => ({
           ...prev,
@@ -355,6 +342,7 @@ const AiAvatarPage: React.FC = () => {
                   selectedVideo={state.selectedVideo}
                   onSelectVideo={() => state.setShowAssetPicker(true)}
                   onRemoveVideo={state.removeVideo}
+                  titleConfig={state.titleConfig}
                 />
               </div>
             </div>
@@ -456,7 +444,6 @@ const AiAvatarPage: React.FC = () => {
                   onCoverConfigChange={(partial) =>
                     state.setCoverConfig((prev) => ({ ...prev, ...partial }))
                   }
-                  titleConfig={state.titleConfig}
                   onSmartCover={handleSmartCover}
                   smartCoverLoading={smartCoverLoading}
                   canSmartCover={state.lipsyncJob?.status === "completed"}
@@ -497,9 +484,8 @@ const AiAvatarPage: React.FC = () => {
           open={state.showBRollModal}
           onClose={() => state.setShowBRollModal(false)}
           existingSegments={state.bRollSegments}
-          scriptText={state.lipsyncJob?.script_text || state.scriptText}
+          scriptText={state.scriptText}
           outputDuration={state.lipsyncJob?.output_duration ?? 0}
-          sentenceTimings={state.lipsyncJob?.sentence_timings}
           onConfirm={state.addBRollSegment}
           onRemove={state.removeBRollSegment}
         />

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

@@ -58,17 +58,15 @@ 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 + 可选 drawtext 标题叠加） ── */
+/* ── 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,
-  title_config?: Record<string, unknown> | null,
   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, title_config: title_config ?? null },
-    // smart-cover 链路：下载视频+抽帧+drawtext 加标题+上传 OSS，需要较长时间，120s 超时
-    { timeout: 120000 },
+    { video_url, max_frames },
+    { timeout: 60000 },
   )
   return response.data
 }
@@ -87,7 +85,7 @@ export const submitRender = async (data: {
 }
 
 export const getRenderJob = async (jobId: string): Promise<RenderJob> => {
-  const response = await apiClient.get<RenderJob>(`/ai-avatar/render/${jobId}`, { timeout: 60000 })
+  const response = await apiClient.get<RenderJob>(`/ai-avatar/render/${jobId}`)
   return response.data
 }
 

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

@@ -5,12 +5,12 @@
  * - 左侧：先选素材库（video 库）→ 再选该库视频素材（已被其他 segment 使用的素材
  *   标灰 + "已选择" 遮罩，pointer-events:none 防重复选择）
  * - 右侧：文案句子列表（点选对应段落，替代原数字索引框）/ 全屏 or 画中画 / 四角位置+大小
- *   （开始/结束时间来自后端精确句子时间戳，基于 TTS 音频静音检测）
+ *   （开始/结束时间已删除，按句子字数占比 × 口播总时长自动估算）
  * - 底部：已配置的画面插入列表（可删除）
  */
 import React, { useEffect, useMemo, useState } from "react"
 import { getAssets, getAssetLibraries, type AssetItem, type AssetLibraryItem } from "@/api/assets"
-import type { BRollSegment, BRollInsertMode, PipPosition, SentenceTiming } from "../types"
+import type { BRollSegment, BRollInsertMode, PipPosition } from "../types"
 import { splitScriptIntoSentences, type ScriptSentence } from "../utils/sentences"
 
 interface ModalBRollEditorProps {
@@ -18,12 +18,10 @@ interface ModalBRollEditorProps {
   onClose: () => void
   /** 当前已有的 B-roll segments（用于标灰已选素材） */
   existingSegments: BRollSegment[]
-  /** 文案全文（优先使用对口型时锁定的 scriptText） */
+  /** 当前文案全文（用于分句） */
   scriptText: string
-  /** 对口型成片总时长（秒） */
+  /** 对口型成片总时长（秒），用于时间自动估算 */
   outputDuration: number
-  /** 后端精确句子时间戳（来自 lipsyncJob.sentence_timings） */
-  sentenceTimings?: SentenceTiming[] | null
   onConfirm: (segment: BRollSegment) => void
   onRemove: (id: string) => void
 }
@@ -45,8 +43,7 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
   onClose,
   existingSegments,
   scriptText,
-  outputDuration: _outputDuration,
-  sentenceTimings,
+  outputDuration,
   onConfirm,
   onRemove,
 }) => {
@@ -65,10 +62,10 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
   const [pipPosition, setPipPosition] = useState<PipPosition>("top-right")
   const [pipScale, setPipScale] = useState(0.3)
 
-  /** 文案分句（使用后端精确时间戳） */
+  /** 文案分句（⑤） */
   const sentences = useMemo(
-    () => splitScriptIntoSentences(scriptText, sentenceTimings),
-    [scriptText, sentenceTimings],
+    () => splitScriptIntoSentences(scriptText, outputDuration),
+    [scriptText, outputDuration],
   )
 
   /** 已被现有 segments 占用的素材 id 集合（标灰、禁止重复选择） */
@@ -267,7 +264,7 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
                           >
                             <span className="aa-sentence-item__idx">{sent.index + 1}</span>
                             <span className="aa-sentence-item__text">{sent.text}</span>
-                            {sent.endTime > 0 && (
+                            {outputDuration > 0 && (
                               <span className="aa-sentence-item__time">
                                 {sent.startTime.toFixed(1)}-{sent.endTime.toFixed(1)}s
                               </span>
@@ -352,7 +349,7 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
                           selectedSentence.endTime,
                           selectedSentence.startTime + 0.5,
                         ).toFixed(1)}
-                        s
+                        s （按字数自动估算）
                       </div>
                     </>
                   ) : (

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

@@ -1,18 +1,17 @@
 /**
  * AI数字人 — 面板5：封面 & 生成
- * - 竖屏 9:16 封面预览（从视频截取 / 自定义上传）+ 标题文字实时叠加预览
+ * - 竖屏 9:16 封面预览（从视频截取 / 自定义上传）
  * - 分辨率选择（720p / 1080p / 4K）
  * - 配置汇总卡片（出镜视频/音色/文案/对口型/B-roll/标题/封面）
  * - 渐变紫色生成按钮
  *
  * 注意：v3 已删除"画面插入模式"，本面板不包含该选项。
  */
-import React, { useMemo, useRef } from "react"
-import type { AiAvatarCoverConfig, AiAvatarTitleConfig } from "../types"
+import React, { useRef } from "react"
+import type { AiAvatarCoverConfig } from "../types"
 
 interface PanelCoverAndGenerateProps {
   coverConfig: AiAvatarCoverConfig
-  titleConfig: AiAvatarTitleConfig
   onCoverConfigChange: (partial: Partial<AiAvatarCoverConfig>) => void
   resolution: string
   onResolutionChange: (r: string) => void
@@ -48,19 +47,8 @@ const LIPSYNC_STATUS_LABEL: Record<string, { text: string; cls: string }> = {
   failed: { text: "失败", cls: "aa-status-badge--failed" },
 }
 
-/** 字体名 → CSS font-family 映射（与后端 drawtext 对齐） */
-const FONT_FAMILY_MAP: Record<string, string> = {
-  思源黑体: "'Noto Sans SC', 'Source Han Sans SC', 'PingFang SC', 'Microsoft YaHei', sans-serif",
-  思源宋体: "'Noto Serif SC', 'Source Han Serif SC', 'SimSun', serif",
-  楷体: "KaiTi, 'STKaiti', serif",
-  黑体: "'Heiti SC', 'SimHei', 'Microsoft YaHei', sans-serif",
-}
-
-const getFontFamily = (font: string): string => FONT_FAMILY_MAP[font] || FONT_FAMILY_MAP["思源黑体"]
-
 const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
   coverConfig,
-  titleConfig,
   onCoverConfigChange,
   resolution,
   onResolutionChange,
@@ -98,85 +86,15 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
 
   const canGenerate = summary.lipsyncStatus === "completed" && !isGenerating
 
-  /** 封面图实际展示的 url：智能封面 > 自定义上传 > 空 */
-  const coverUrl =
-    coverConfig.smart_cover_url || coverConfig.thumbnail_url || coverConfig.upload_url
-  const hasCoverImage = Boolean(coverUrl)
-
-  /** 是否显示标题叠加层：有图、有文字、非加载中 */
-  const showTitleOverlay =
-    hasCoverImage && !smartCoverLoading && titleConfig.title.trim().length > 0
-
-  /** 计算标题叠加层的 inline 样式 */
-  const titleOverlayStyle = useMemo<React.CSSProperties>(() => {
-    const style: React.CSSProperties = {
-      position: "absolute",
-      left: "50%",
-      width: "90%",
-      transform: "translateX(-50%)",
-      textAlign: "center",
-      boxSizing: "border-box",
-      padding: "0 4px",
-      wordBreak: "break-word",
-      whiteSpace: "pre-wrap",
-      color: titleConfig.color || "#ffffff",
-      fontSize: `${(titleConfig.size || 28) * 0.55}px`, // 预览容器缩放，与 PanelLipsyncPreview 对齐
-      fontFamily: getFontFamily(titleConfig.font),
-      fontWeight: titleConfig.bold ? "bold" : "normal",
-      fontStyle: titleConfig.italic ? "italic" : "normal",
-      lineHeight: 1.3,
-      pointerEvents: "none",
-    }
-
-    // 位置
-    const pos = titleConfig.position || "bottom"
-    if (pos === "top") {
-      style.top = "40px"
-    } else if (pos === "center") {
-      style.top = "50%"
-      style.transform = "translate(-50%, -50%)"
-    } else if (pos === "custom" && titleConfig.pos_x != null && titleConfig.pos_y != null) {
-      // pos_x/pos_y 是相对预览容器的百分比坐标
-      style.left = `${titleConfig.pos_x}%`
-      style.top = `${titleConfig.pos_y}%`
-      style.transform = "translate(-50%, -50%)"
-    } else {
-      style.bottom = "40px"
-    }
-
-    // 描边优先于阴影（二者互斥，与 drawtext 对齐）
-    if (titleConfig.stroke) {
-      // 描边宽度按字号估算，保证视觉一致
-      const strokeWidth = Math.max(1, Math.round(titleConfig.size / 18))
-      ;(style as React.CSSProperties)["WebkitTextStroke"] = `${strokeWidth}px rgba(0,0,0,0.75)`
-      style.textShadow = "none"
-    } else if (titleConfig.shadow) {
-      style.textShadow = "0 2px 8px rgba(0,0,0,0.7), 0 0 2px rgba(0,0,0,0.5)"
-    } else {
-      // 无描边无阴影时，不加额外效果（与后端 drawtext 对齐：无 stroke/shadow 则不加）
-      style.textShadow = "none"
-    }
-
-    return style
-  }, [titleConfig])
-
   return (
     <div className="aa-cover-generate">
       {/* 封面预览（竖屏 9:16） */}
       <div className="aa-cover-preview">
-        {hasCoverImage ? (
-          <img src={coverUrl!} alt="封面预览" draggable={false} />
+        {coverConfig.thumbnail_url ? (
+          <img src={coverConfig.thumbnail_url} alt="封面预览" />
         ) : (
           <span className="aa-cover-preview__placeholder">暂无封面</span>
         )}
-        {/* 智能封面加载遮罩 */}
-        {smartCoverLoading && <div className="aa-cover-preview__loading">⏳ 智能选帧中…</div>}
-        {/* 标题文字叠加层（实时预览，仅前端视觉参考，最终由后端 ffmpeg drawtext 叠加） */}
-        {showTitleOverlay && (
-          <div style={titleOverlayStyle} aria-hidden="true">
-            {titleConfig.title}
-          </div>
-        )}
       </div>
 
       <div className="aa-cover-actions">

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

@@ -14,8 +14,8 @@ interface PanelLipsyncPreviewProps {
   onRemoveBRoll: (id: string) => void
   /** 标题配置（实时叠加预览用） */
   titleConfig?: AiAvatarTitleConfig
-  /** 标题位置变更回调（拖拽结束时调用，发送百分比坐标 + position:"custom"） */
-  onTitlePositionChange?: (pos: { pos_x: number; pos_y: number; position: string }) => void
+  /** 标题位置变更回调（拖拽结束时调用） */
+  onTitlePositionChange?: (pos: { pos_x: number; pos_y: number }) => void
 }
 
 const BROLL_MODE_LABEL: Record<BRollSegment["mode"], string> = {
@@ -56,9 +56,11 @@ export function PanelLipsyncPreview({
   const titleOverlayStyle: React.CSSProperties | null = titleConfig?.title
     ? {
         position: "absolute",
+        left: "50%",
+        transform: "translateX(-50%)",
         color: titleConfig.color || "#ffffff",
         fontFamily: titleConfig.font || "思源黑体",
-        fontSize: `${(titleConfig.size || 28) * 0.55}px`,
+        fontSize: `${(titleConfig.size || 36) * 0.55}px`, // 预览等比缩
         fontWeight: titleConfig.bold ? 700 : 400,
         fontStyle: titleConfig.italic ? "italic" : "normal",
         textAlign: "center",
@@ -66,19 +68,11 @@ export function PanelLipsyncPreview({
         padding: "4px 8px",
         textShadow: titleConfig.shadow ? "0 2px 4px rgba(0,0,0,0.8)" : undefined,
         WebkitTextStroke: titleConfig.stroke ? "1.5px #000" : undefined,
-        ...(titleConfig.position === "custom" &&
-        titleConfig.pos_x != null &&
-        titleConfig.pos_y != null
-          ? {
-              left: `${titleConfig.pos_x}%`,
-              top: `${titleConfig.pos_y}%`,
-              transform: "translateX(-50%) translateY(-50%)",
-            }
-          : titleConfig.position === "top"
-            ? { left: "50%", top: 8, transform: "translateX(-50%)" }
-            : titleConfig.position === "bottom"
-              ? { left: "50%", bottom: 8, transform: "translateX(-50%)" }
-              : { left: "50%", top: "50%", transform: "translateX(-50%) translateY(-50%)" }),
+        ...(titleConfig.position === "top"
+          ? { top: 8 }
+          : titleConfig.position === "bottom"
+            ? { bottom: 8 }
+            : { top: "50%", transform: "translateX(-50%) translateY(-50%)" }),
       }
     : null
 
@@ -111,10 +105,7 @@ export function PanelLipsyncPreview({
       const rect = previewContainerRef.current.getBoundingClientRect()
       const relX = Math.max(0, Math.min(rect.width, e.clientX - rect.left))
       const relY = Math.max(0, Math.min(rect.height, e.clientY - rect.top))
-      // 发送百分比坐标（0-100），与后端 drawtext 百分比表达式对齐
-      const xpct = Math.round((relX / rect.width) * 1000) / 10
-      const ypct = Math.round((relY / rect.height) * 1000) / 10
-      onTitlePositionChange({ pos_x: xpct, pos_y: ypct, position: "custom" })
+      onTitlePositionChange({ pos_x: relX, pos_y: relY })
     }
     ;(e.currentTarget as HTMLDivElement).style.cursor = "grab"
   }

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

@@ -2,16 +2,17 @@
  * AI数字人 — 出镜视频选择面板
  * - 未选视频：虚线上传区，点击打开素材库弹窗
  * - 已选视频：竖屏 9:16 预览播放器 + 视频信息卡片 + 移除按钮
- *
- * 注意：本面板只展示原始素材视频，不叠加标题（标题在对口型预览和最终成片上展示）
  */
 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） */
@@ -26,6 +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) {
@@ -55,13 +57,42 @@ export function PanelVideoSelector({
 
   return (
     <div>
-      {/* 竖屏 9:16 视频预览播放器（纯素材预览，不叠加标题） */}
-      <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/types.ts

@@ -44,23 +44,12 @@ export interface LipsyncJob {
   status: LipsyncStatus
   progress: number
   output_video_url: string | null
-  /** 对口型成片总时长（秒），后端返回 */
-  script_text: string
+  /** 对口型成片总时长（秒），后端返回；用于 B-roll 时间自动估算（#1809 ⑥） */
   output_duration?: number
-  /** 精确句子时间戳（后端基于 TTS 音频静音检测计算） */
-  sentence_timings?: SentenceTiming[] | null
   error_message: string | null
   created_at: string
 }
 
-/* ── 句子时间戳（后端精确计算） ── */
-export interface SentenceTiming {
-  index: number
-  text: string
-  start_time: number
-  end_time: number
-}
-
 /* ── B-roll 画面插入 ── */
 export type BRollInsertMode = "fullscreen" | "pip"
 export type PipPosition = "top-left" | "top-right" | "bottom-left" | "bottom-right"
@@ -88,7 +77,7 @@ export interface AiAvatarTitleConfig {
   shadow: boolean
   color: string
   auto_subtitle: boolean
-  /** 自定义位置坐标（position=custom 时生效，百分比 0-100） */
+  /** 自定义位置坐标（position=custom 时生效，像素） */
   pos_x?: number
   pos_y?: number
 }
@@ -112,7 +101,6 @@ export interface RenderJob {
   status: RenderStatus
   progress: number
   output_video_url: string | null
-  output_cover_url: string | null
   error_message: string | null
   created_at: string
 }

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

@@ -39,7 +39,7 @@ export function buildTitleConfigPayload(cfg: AiAvatarTitleConfig): Record<string
     text,
     enabled: true,
     font: cfg.font || "思源黑体",
-    font_size: Math.round(cfg.size) || 28,
+    font_size: Math.round(cfg.size) || 36,
     font_color: cfg.color || "#ffffff",
     position,
     bold: !!cfg.bold,

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

@@ -1,10 +1,6 @@
 /**
- * AI数字人 — 文案分句工具
- *
- * 分句规则与后端 _split_script_into_sentences 保持一致。
- * 时间戳由后端基于 TTS 音频静音检测精确计算，前端不再做字数比例估算。
+ * AI数字人 — 文案分句 & B-roll 时间自动估算（#1809 ⑤⑥）
  */
-import type { SentenceTiming } from "../types"
 
 export interface ScriptSentence {
   /** 句子序号（从 0 开始，对应提交给后端的 script_segment_index） */
@@ -13,21 +9,21 @@ export interface ScriptSentence {
   text: string
   /** 句子字数（按中文/字符计，去除空白） */
   charCount: number
-  /** 累计起始字数 */
+  /** 累计起始字数（用于时间估算） */
   startChar: number
-  /** 精确起始时间（秒），来自后端 sentence_timings；无数据时为 0 */
+  /** 估算的对口型视频内起始时间（秒） */
   startTime: number
-  /** 精确结束时间（秒），来自后端 sentence_timings；无数据时为 0 */
+  /** 估算的对口型视频内结束时间（秒） */
   endTime: number
 }
 
 /**
  * 按句号/问号/感叹号/分号/换行分句（兼容中英文标点）。
- * 时间戳从后端 sentence_timings 获取（精确）；若无则返回 0（由调用方降级处理）。
+ * 空文案返回空数组。时间按「该句字数 ÷ 全文总字数 × 口播总时长」线性估算。
  */
 export function splitScriptIntoSentences(
   scriptText: string,
-  sentenceTimings?: SentenceTiming[] | null,
+  outputDuration: number,
 ): ScriptSentence[] {
   const text = (scriptText || "").trim()
   if (!text) return []
@@ -37,25 +33,30 @@ export function splitScriptIntoSentences(
     .map((part) => part.trim())
     .filter((part) => part.length > 0)
 
+  const totalChars = rawParts.reduce((sum, part) => sum + part.replace(/\s/g, "").length, 0)
+  const duration = outputDuration > 0 ? outputDuration : 0
+
   const sentences: ScriptSentence[] = []
   let accChar = 0
   rawParts.forEach((part, i) => {
     const charCount = part.replace(/\s/g, "").length
-    // 从后端精确时间戳获取；无数据时返回 0
-    const timing = sentenceTimings?.[i]
-    const startTime = timing?.start_time ?? 0
-    const endTime = timing?.end_time ?? 0
-
+    const startTime = duration > 0 && totalChars > 0 ? (accChar / totalChars) * duration : 0
+    const endTime =
+      duration > 0 && totalChars > 0 ? ((accChar + charCount) / totalChars) * duration : 0
     sentences.push({
       index: i,
       text: part,
       charCount,
       startChar: accChar,
-      startTime,
-      endTime,
+      startTime: round1(startTime),
+      endTime: round1(endTime),
     })
     accChar += charCount
   })
 
   return sentences
 }
+
+function round1(n: number): number {
+  return Math.round(n * 10) / 10
+}

apps/web/src/pages/products/ProductLibrary.tsx

@@ -1,22 +1,17 @@
 /**
  * 成片库页面 — V21 设计系统
- * 卡片网格布局，支持视频内联播放/下载/分享、批量操作、筛选、无限滚动分页
+ * 卡片网格布局，支持视频内联播放/下载/分享、批量操作、筛选
  *
  * 主组件仅保留 Hook 组装与整体布局
- * 列表查询 → hooks/useProductList（useInfiniteQuery 分页）
+ * 列表查询 → hooks/useProductList
  * 操作逻辑 → hooks/useProductActions
  * 筛选栏 → components/ProductFilterBar
  * 批量操作栏 → components/ProductBatchBar
  * 空状态 → components/ProductEmptyState
  * 产品卡片 → components/ProductCard（内联视频播放）
  */
-import React, { useEffect, useRef } from "react"
-import {
-  VideoCameraOutlined,
-  DownloadOutlined,
-  ReloadOutlined,
-  LoadingOutlined,
-} from "@ant-design/icons"
+import React from "react"
+import { VideoCameraOutlined, DownloadOutlined, ReloadOutlined } from "@ant-design/icons"
 import { Button } from "@/components/ui"
 import { ProductCard } from "./components/ProductCard"
 import { ProductFilterBar } from "./components/ProductFilterBar"
@@ -29,13 +24,11 @@ import "./products.css"
 
 const ProductLibrary: React.FC = () => {
   const {
+    products,
     filteredProducts,
     isLoading,
-    isFetchingNextPage,
     isError,
     error,
-    hasNextPage,
-    fetchNextPage,
     refetch,
     searchText,
     setSearchText,
@@ -71,40 +64,19 @@ const ProductLibrary: React.FC = () => {
   } = useProductActions({
     selectedIds,
     clearSelection,
-    products: filteredProducts,
+    products,
     setPlayingProduct: () => {}, // 不再使用弹窗播放
   })
 
   const { recomputeDedup, isRecomputing } = useRecomputeDedup()
 
-  /* ── 无限滚动：IntersectionObserver 监听底部哨兵元素 ── */
-  const sentinelRef = useRef<HTMLDivElement>(null)
-
-  useEffect(() => {
-    const el = sentinelRef.current
-    if (!el) return
-    // 已有数据但正在加载中/没有更多页时不触发
-    if (isFetchingNextPage || !hasNextPage) return
-
-    const observer = new IntersectionObserver(
-      (entries) => {
-        if (entries[0]?.isIntersecting) {
-          void fetchNextPage()
-        }
-      },
-      { rootMargin: "200px" },
-    )
-    observer.observe(el)
-    return () => observer.disconnect()
-  }, [fetchNextPage, hasNextPage, isFetchingNextPage])
-
-  // ── Loading 状态（仅首次加载）──
-  if (isLoading && filteredProducts.length === 0) {
+  // ── Loading 状态 ──
+  if (isLoading) {
     return <ProductEmptyState type="loading" />
   }
 
   // ── Error 状态 ──
-  if (isError && filteredProducts.length === 0) {
+  if (isError) {
     console.error("[ProductLibrary] 加载失败:", error)
     const errorMsg = error?.message || "加载失败"
     const is404 = errorMsg.includes("404") || errorMsg.includes("Not Found")
@@ -171,46 +143,22 @@ const ProductLibrary: React.FC = () => {
 
       {/* 卡片网格 */}
       {filteredProducts.length > 0 ? (
-        <>
-          <div className="xx-products-grid">
-            {filteredProducts.map((product) => (
-              <ProductCard
-                key={product.id}
-                product={product}
-                isSelected={selectedIds.has(product.id)}
-                batchMode={batchMode}
-                onToggleSelect={handleToggleSelect}
-                onDownload={handleDownload}
-                onShare={handleShare}
-                onDelete={handleDelete}
-                onPublish={handlePublish}
-                onReviewStatusChange={handleReviewStatusChange}
-              />
-            ))}
-          </div>
-
-          {/* 底部哨兵 + 状态提示 */}
-          <div
-            ref={sentinelRef}
-            style={{
-              gridColumn: "1 / -1",
-              textAlign: "center",
-              padding: "24px 0",
-              fontSize: 13,
-              color: "#8c8ca1",
-            }}
-          >
-            {isFetchingNextPage ? (
-              <>
-                <LoadingOutlined /> 加载中…
-              </>
-            ) : hasNextPage ? (
-              <span style={{ opacity: 0 }}>加载更多</span>
-            ) : (
-              <span>—— 已加载全部 ——</span>
-            )}
-          </div>
-        </>
+        <div className="xx-products-grid">
+          {filteredProducts.map((product) => (
+            <ProductCard
+              key={product.id}
+              product={product}
+              isSelected={selectedIds.has(product.id)}
+              batchMode={batchMode}
+              onToggleSelect={handleToggleSelect}
+              onDownload={handleDownload}
+              onShare={handleShare}
+              onDelete={handleDelete}
+              onPublish={handlePublish}
+              onReviewStatusChange={handleReviewStatusChange}
+            />
+          ))}
+        </div>
       ) : (
         <ProductEmptyState type="empty" />
       )}

apps/web/src/pages/products/hooks/useProductList/index.ts

@@ -1,53 +1,28 @@
 import { useMemo } from "react"
-import { useInfiniteQuery } from "@tanstack/react-query"
+import { useQuery } from "@tanstack/react-query"
 import { getProducts, type ProductItem as ApiProductItem } from "@/api/products"
 import { mapApiProduct } from "../../utils"
-import type { ProductItem } from "../../types"
 import { useProductFiltering } from "./useProductFiltering"
 import { useBatchSelection } from "./useBatchSelection"
 
 export type { Filters } from "./useProductFiltering"
 
-const PAGE_SIZE = 20
-
 export const useProductList = () => {
-  /* ── 无限滚动获取成品列表（每页 20 条） ── */
+  /* ── 获取成品列表 ── */
   const {
-    data,
+    data: apiProducts = [],
     isLoading,
-    isFetchingNextPage,
     isError,
     error,
-    hasNextPage,
-    fetchNextPage,
     refetch,
-  } = useInfiniteQuery<
-    {
-      items: ApiProductItem[]
-      total: number
-      page: number
-      page_size: number
-    },
-    Error
-  >({
+  } = useQuery<ApiProductItem[], Error>({
     queryKey: ["products"],
-    queryFn: async ({ pageParam = 1 }) =>
-      getProducts({ page: pageParam as number, page_size: PAGE_SIZE }),
-    initialPageParam: 1,
-    getNextPageParam: (lastPage) => {
-      const loadedCount = lastPage.page * lastPage.page_size
-      return loadedCount < lastPage.total ? lastPage.page + 1 : undefined
-    },
+    queryFn: () => getProducts(),
     staleTime: 30_000,
   })
 
-  // 将所有页拼接为一维数组，再做前端映射+排序
-  const apiProducts = useMemo<ApiProductItem[]>(() => {
-    if (!data?.pages) return []
-    return data.pages.flatMap((p) => p.items)
-  }, [data])
-
-  const products = useMemo<ProductItem[]>(
+  // 映射为前端类型，按创建时间倒序排列，防御非数组返回
+  const products = useMemo(
     () =>
       (Array.isArray(apiProducts) ? apiProducts : []).map(mapApiProduct).sort((a, b) => {
         if (!a.date || a.date === "—") return 1
@@ -90,11 +65,8 @@ export const useProductList = () => {
     products,
     filteredProducts,
     isLoading,
-    isFetchingNextPage,
     isError,
     error,
-    hasNextPage,
-    fetchNextPage,
     refetch,
     // 筛选
     searchText,

packages/adapters/sqlalchemy_impl/models.py

@@ -703,9 +703,6 @@ class LipsyncJobModel(Base):
     error_message = Column(Text, nullable=False, default="")
     error_code = Column(String(100), nullable=False, default="")
 
-    # 精确句子时间戳（TTS 合成后由 silencedetect 计算，用于 B-roll 精确定位）
-    sentence_timings = Column(JSON, nullable=True)  # list[{index,text,start_time,end_time}]
-
     # 时间戳
     submitted_at = Column(DateTime, nullable=True)
     completed_at = Column(DateTime, nullable=True)

packages/domain/generated_video.py

@@ -49,17 +49,19 @@ class GeneratedVideo:
         thumbnail_url: str | None = None,
         generation_params: dict[str, Any] | None = None,
     ) -> "GeneratedVideo":
-        # project_id / generation_task_id 允许为空：AI数字人等无项目场景下，前端可能不传 project_id；
-        # lipsync 路径下 generation_task_id 也可能暂时为空。空串会被下面统一兜底为 "" 入库。
-        if not name or not name.strip():
+        if not project_id.strip():
+            raise ValueError("project_id cannot be empty")
+        if not generation_task_id.strip():
+            raise ValueError("generation_task_id cannot be empty")
+        if not name.strip():
             raise ValueError("name cannot be empty")
-        if not file_url or not file_url.strip():
+        if not file_url.strip():
             raise ValueError("file_url cannot be empty")
         return cls(
             id=uuid4().hex,
-            project_id=(project_id or "").strip(),
-            user_id=(user_id or "").strip(),
-            generation_task_id=(generation_task_id or "").strip(),
+            project_id=project_id.strip(),
+            user_id=user_id.strip(),
+            generation_task_id=generation_task_id.strip(),
             name=name.strip(),
             file_url=file_url.strip(),
             file_size=file_size,

packages/domain/video_filter_builder.py

@@ -377,30 +377,26 @@ def _append_audio_concat(parts: list[str], clip_chains: list[ClipFilterChain]) -
 
 # ── 标题 drawtext 滤镜构建（#1789）─────────────────────────────────────────────
 
-# drawtext 字体搜索路径：按优先级从高到低排列
-# 服务器使用 Noto Sans SC（思源黑体）作为默认字体。
-# - NotoSansSC-VF.ttf 是 worker-base.Dockerfile 中 COPY 的 VF 字体（含所有字重，无 Mono 变体），优先级最高
-# - .ttc 系列为 fonts-noto-cjk 包预装字体（Dockerfile 已删除含 Mono 变体的旧 .ttc，存在时作为 fallback）
-# - DejaVuSans 仅含拉丁字符不支持中文，已移除
+# drawtext 字体搜索路径：按优先级列出常见安装位置
+# 服务器使用 Noto Sans SC（思源黑体）作为默认字体
 DRAWTEXT_FONT_SEARCH_PATHS: list[str] = [
-    "/usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf",
     "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
-    "/usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc",
     "/usr/share/fonts/noto-cjk/NotoSansCJK-Regular.ttc",
     "/usr/share/fonts/google-noto-cjk/NotoSansCJK-Regular.ttc",
     "/usr/share/fonts/truetype/noto/NotoSansSC-Regular.ttf",
     "/usr/share/fonts/noto/NotoSansSC-Regular.ttf",
+    "/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf",
 ]
 
-# 前端字体名 → drawtext 字体搜索关键字（匹配 DRAWTEXT_FONT_SEARCH_PATHS 中的文件名关键字）
+# 前端字体名 → drawtext 字体搜索关键字
 DRAWTEXT_FONT_MAP: dict[str, str] = {
-    "思源黑体": "NotoSansSC",
+    "思源黑体": "NotoSansCJK",
     "思源宋体": "NotoSerifCJK",
-    "苹方": "NotoSansSC",
-    "PingFang": "NotoSansSC",
-    "微软雅黑": "NotoSansSC",
+    "苹方": "NotoSansCJK",
+    "PingFang": "NotoSansCJK",
+    "微软雅黑": "NotoSansCJK",
     "楷体": "NotoSerifCJK",
-    "华康俪金黑": "NotoSansSC",
+    "华康俪金黑": "NotoSansCJK",
 }
 
 
@@ -481,13 +477,13 @@ def build_title_drawtext_filter(
 
     # ── 样式参数 ──
     font_name = title_config.get("font") or title_config.get("font_preset") or "思源黑体"
-    font_size = int(title_config.get("font_size") or title_config.get("size") or 28)
+    font_size = int(title_config.get("font_size") or title_config.get("size") or 36)
     font_color = title_config.get("font_color") or title_config.get("color") or "#ffffff"
     # 去掉 # 前缀（drawtext 用纯 hex 或颜色名）
     if font_color.startswith("#"):
         font_color = font_color[1:]
 
-    position = title_config.get("position") or "bottom"
+    position = title_config.get("position", "top")
     bold = bool(title_config.get("bold", True))
     stroke = title_config.get("stroke")
     shadow = title_config.get("shadow")
@@ -508,30 +504,26 @@ def build_title_drawtext_filter(
     params.append(f"fontsize={font_size}")
     params.append(f"fontcolor={font_color}")
 
-    # 粗体：drawtext 没有独立的 bold 参数，通过加大 borderw 模拟视觉粗体效果。
-    # 注意：不能使用 `font=bold`——FFmpeg drawtext 的 font 参数需要 fontconfig 能解析的
-    # 字体族名，而 "bold" 不是合法族名，会导致整个 filter_complex 解析失败（exit code 234）。
-    # 当用户未显式配置描边宽度时，bold 模式自动将 borderw 提升到 3 以模拟粗体。
+    # 粗体：bold 在 drawtext 中通过 font 的 Bold 变体实现
+    # 若字体有 Bold 变体可用 fontfont=bold；否则通过 borderw 模拟
+    if bold:
+        # 使用 font 参数尝试加载 Bold 变体（Noto Sans SC 有 Bold 变体文件）
+        params.append("font=bold")
 
     # 描边（borderw 需要 libfreetype 支持）
-    # 粗体无显式描边时，自动用 borderw=3 + 近色描边模拟粗体；显式 stroke 按用户配置走
-    border_width = 0
-    border_color = "000000"
     if stroke:
         if isinstance(stroke, bool):
             border_width = 2
-            border_color = "000000"
+            border_color = "black"
         elif isinstance(stroke, dict):
-            if stroke.get("enabled", True):
-                border_width = int(stroke.get("width", 2))
-                border_color = (stroke.get("color") or "#000000").lstrip("#")
-    elif bold:
-        # 粗体模式且未配描边：加大描边宽度模拟粗体效果
-        border_width = 3
-        border_color = font_color  # 用字体同色描边，视觉上加粗字形而非黑边
-    if border_width > 0:
-        params.append(f"borderw={border_width}")
-        params.append(f"bordercolor={border_color}")
+            border_width = int(stroke.get("width", 2)) if stroke.get("enabled", True) else 0
+            border_color = (stroke.get("color") or "#000000").lstrip("#")
+        else:
+            border_width = 0
+            border_color = "black"
+        if border_width > 0:
+            params.append(f"borderw={border_width}")
+            params.append(f"bordercolor={border_color}")
 
     # 阴影（shadowcolor + shadowx/y）
     if shadow:
@@ -556,13 +548,8 @@ def build_title_drawtext_filter(
         and not isinstance(pos_x, bool)
         and not isinstance(pos_y, bool)
     ):
-        # pos_x/pos_y 为百分比坐标（0-100），转换为 drawtext 表达式
-        # 例如 pos_x=50 → x=(w-text_w)*0.50（水平居中偏50%）
-        #      pos_y=30 → y=(h-text_h)*0.30
-        pct_x = max(0.0, min(100.0, float(pos_x))) / 100.0
-        pct_y = max(0.0, min(100.0, float(pos_y))) / 100.0
-        params.append(f"x=(w-text_w)*{pct_x:.4f}")
-        params.append(f"y=(h-text_h)*{pct_y:.4f}")
+        params.append(f"x={int(pos_x)}")
+        params.append(f"y={int(pos_y)}")
     else:
         # 三档预设位置：top / center / bottom
         # x 始终水平居中：(w-text_w)/2
@@ -586,7 +573,7 @@ def build_broll_overlay_filter(
     video_duration: float,
     output_width: int = DEFAULT_OUTPUT_WIDTH,
     output_height: int = DEFAULT_OUTPUT_HEIGHT,
-) -> tuple[str, str | None]:
+) -> str:
     """构建 B-roll 叠加滤镜链。
 
     支持两种模式:
@@ -594,182 +581,121 @@ def build_broll_overlay_filter(
     - pip: 在对口型视频上叠加画中画 B-roll
 
     Args:
-        b_roll_segments: B-roll 片段配置列表（原始顺序，决定 FFmpeg -i 输入顺序）
+        b_roll_segments: B-roll 片段配置列表
         video_duration: 对口型视频总时长(秒)
-        output_width: 输出宽度（默认 1280；AI 数字人竖屏传 720）
-        output_height: 输出高度（默认 720；AI 数字人竖屏传 1280）
+        output_width: 输出宽度
+        output_height: 输出高度
 
     Returns:
-        (filter_complex_str, final_label)
-        - filter_complex_str: filter_complex 片段字符串（末尾无分号）
-        - final_label: 最终输出 pad 标签名，如 "vout"；无 B-roll 时返回 None
+        FFmpeg filter_complex 滤镜字符串片段
     """
     if not b_roll_segments:
-        return "", None
-
-    # 建立原始列表下标 → FFmpeg 输入下标的映射：
-    # cmd 中 [0:v] 是主视频，随后按 b_roll_segments 原始顺序追加 -i，
-    # 因此第 i 个 segment 的输入是 [{i+1}:v]
-    def _input_label(seg: dict[str, Any]) -> str:
-        # seg 必须来自 b_roll_segments；通过 id() 在原列表中查找
-        for i, s in enumerate(b_roll_segments):
-            if s is seg:
-                return f"[{i + 1}:v]"
-        # fallback: 找不到时不应发生，保守返回
-        return "[1:v]"
+        return ""
 
     parts: list[str] = []
     sorted_segments = sorted(b_roll_segments, key=lambda s: s.get("start_time", 0))
 
-    # 按模式分组
+    # 按模式分组处理
     fullscreen_segments = [s for s in sorted_segments if s.get("mode") == "fullscreen"]
     pip_segments = [s for s in sorted_segments if s.get("mode") == "pip"]
 
-    final_label = None
-
     # ── fullscreen 模式: 切分 + concat ──
     if fullscreen_segments:
-        fs_filter, fs_label = _build_fullscreen_filters(
-            fullscreen_segments, b_roll_segments, video_duration, output_width, output_height, _input_label
-        )
-        parts.append(fs_filter)
-        final_label = fs_label
-    else:
-        fs_label = None
+        parts.append(_build_fullscreen_filters(fullscreen_segments, video_duration, output_width, output_height))
 
     # ── pip 模式: overlay 滤镜 ──
     if pip_segments:
-        pip_filter, pip_label = _build_pip_filters(
-            pip_segments, output_width, output_height, _input_label, base_label=fs_label
-        )
-        parts.append(pip_filter)
-        final_label = pip_label
+        for idx, seg in enumerate(pip_segments):
+            start = seg.get("start_time", 0)
+            end = seg.get("end_time", video_duration)
+            scale = seg.get("pip_scale", 0.3)
+            position = seg.get("pip_position", "bottom_right")
+
+            pip_w = int(output_width * scale)
+            pip_h = int(output_height * scale)
+
+            # 位置映射
+            pos_map = {
+                "top_left": "10:10",
+                "top_right": "W-w-10:10",
+                "bottom_left": "10:H-h-10",
+                "bottom_right": "W-w-10:H-h-10",
+                "center": "(W-w)/2:(H-h)/2",
+            }
+            pos_expr = pos_map.get(position, pos_map["bottom_right"])
+
+            broll_input_idx = len(sorted_segments)  # placeholder for input index
+            parts.append(
+                f"[{broll_input_idx + idx}:v]scale={pip_w}:{pip_h}," f"enable='between(t,{start},{end})'[pip{idx}];"
+            )
+            # overlay onto main stream
+            if idx == 0:
+                base_label = "[vout]" if fullscreen_segments else "[0:v]"
+            else:
+                base_label = f"[pip{idx - 1}]"
+            parts.append(f"{base_label}[pip{idx}]overlay={pos_expr}:enable='between(t,{start},{end})'[vout{idx}];")
 
     result = "".join(parts)
     # 清理末尾多余分号
     if result.endswith(";"):
         result = result[:-1]
-    return result, final_label
+    return result
 
 
 def _build_fullscreen_filters(
-    sorted_fs_segments: list[dict[str, Any]],
-    all_segments: list[dict[str, Any]],
+    segments: list[dict[str, Any]],
     video_duration: float,
     output_width: int,
     output_height: int,
-    input_label_fn,
-) -> tuple[str, str]:
-    """构建 fullscreen 模式的切分 + concat 滤镜。
+) -> str:
+    """构建 fullscreen 模式的切分 + concat 滤镜.
 
-    将主视频按 B-roll 时间段切分，然后用 concat 拼接主视频片段和 B-roll 片段。
-
-    Returns:
-        (filter_str, final_label) 其中 final_label 是 concat 输出的 pad 标签
+    将对口型视频按 B-roll 时间段切分，然后用 concat 拼接 B-roll 片段。
     """
     parts: list[str] = []
     prev_end = 0.0
 
-    # 注意：这里的 idx 是 sorted_fs_segments 中的下标；
-    # 实际 FFmpeg 输入下标必须通过 input_label_fn 查询
-    for idx, seg in enumerate(sorted_fs_segments):
+    for idx, seg in enumerate(segments):
         start = seg.get("start_time", 0)
         end = seg.get("end_time", video_duration)
 
-        # 主视频片段（B-roll 之前）
+        # 保持原视频片段（B-roll 之前的部分）
         if prev_end < start:
             parts.append(f"[0:v]trim=start={prev_end}:end={start},setpts=PTS-STARTPTS[main{idx}];")
 
-        # B-roll 片段：缩放到输出分辨率并裁到对应时长
-        in_lbl = input_label_fn(seg)
+        # B-roll 片段：缩放至目标分辨率
         parts.append(
-            f"{in_lbl}scale={output_width}:{output_height}"
+            f"[{idx + 1}:v]scale={output_width}:{output_height}"
             f":force_original_aspect_ratio=decrease,"
             f"pad={output_width}:{output_height}:(ow-iw)/2:(oh-ih)/2,"
             f"trim=start=0:end={end - start},setpts=PTS-STARTPTS[br{idx}];"
         )
         prev_end = end
 
-    # 尾部主视频片段
+    # 尾部片段
     if prev_end < video_duration:
-        last_idx = len(sorted_fs_segments)
+        last_idx = len(segments)
         parts.append(f"[0:v]trim=start={prev_end}:end={video_duration},setpts=PTS-STARTPTS[main{last_idx}];")
 
     # concat 所有片段
-    segment_labels: list[str] = []
-    for idx, seg in enumerate(sorted_fs_segments):
-        start = seg.get("start_time", 0)
-        # 每段 B-roll 之前是否有主视频片段？
-        has_main_before = (idx == 0 and start > 0) or (
-            idx > 0 and sorted_fs_segments[idx - 1].get("end_time", 0) < start
-        )
-        if has_main_before:
-            segment_labels.append(f"[main{idx}]")
+    segment_labels = []
+    for idx in range(len(segments)):
+        start = segments[idx].get("start_time", 0)
+        if (idx == 0 and segments[0].get("start_time", 0) > 0) or idx > 0:
+            prev_end_prev = segments[idx - 1].get("end_time", 0) if idx > 0 else 0
+            if prev_end_prev < start:
+                segment_labels.append(f"[main{idx}]")
         segment_labels.append(f"[br{idx}]")
-    if prev_end < video_duration:
-        segment_labels.append(f"[main{len(sorted_fs_segments)}]")
 
-    final_lbl = "vout_fs"
+    if prev_end < video_duration:
+        segment_labels.append(f"[main{len(segments)}]")
+
     n = len(segment_labels)
     if n > 0:
         concat_inputs = "".join(segment_labels)
-        parts.append(f"{concat_inputs}concat=n={n}:v=1:a=0[{final_lbl}];")
+        parts.append(f"{concat_inputs}concat=n={n}:v=1:a=0[vout];")
 
-    return "".join(parts), final_lbl
-
-
-def _build_pip_filters(
-    pip_segments: list[dict[str, Any]],
-    output_width: int,
-    output_height: int,
-    input_label_fn,
-    base_label: str | None,
-) -> tuple[str, str]:
-    """构建 PIP（画中画）overlay 滤镜链。
-
-    Args:
-        pip_segments: 按时间排序的 pip 片段
-        output_width: 输出宽度
-        output_height: 输出高度
-        input_label_fn: 片段 → 输入标签的映射函数
-        base_label: 前序滤镜链输出的标签（如 fullscreen 的 vout_fs），为 None 则基于 [0:v]
-
-    Returns:
-        (filter_str, final_label)
-    """
-    parts: list[str] = []
-    cur_label = base_label  # 当前叠加到的标签
-
-    pos_map = {
-        "top_left": "10:10",
-        "top_right": "W-w-10:10",
-        "bottom_left": "10:H-h-10",
-        "bottom_right": "W-w-10:H-h-10",
-        "center": "(W-w)/2:(H-h)/2",
-    }
-
-    for idx, seg in enumerate(pip_segments):
-        start = seg.get("start_time", 0)
-        end = seg.get("end_time", 0)
-        scale = seg.get("pip_scale", 0.3)
-        position = seg.get("pip_position", "bottom_right")
-        pos_expr = pos_map.get(position, pos_map["bottom_right"])
-
-        pip_w = max(1, int(output_width * scale))
-        pip_h = max(1, int(output_height * scale))
-        enable_expr = f"enable='between(t,{start},{end})'"
-
-        in_lbl = input_label_fn(seg)
-        pip_scaled = f"pip{idx}"
-        parts.append(f"{in_lbl}scale={pip_w}:{pip_h},{enable_expr}[{pip_scaled}];")
-
-        # overlay onto the current base
-        base = f"[{cur_label}]" if cur_label else "[0:v]"
-        out_lbl = f"vout_pip{idx}" if idx < len(pip_segments) - 1 else "vout"
-        parts.append(f"{base}[{pip_scaled}]overlay={pos_expr}:{enable_expr}[{out_lbl}];")
-        cur_label = out_lbl
-
-    return "".join(parts), cur_label or "vout"
+    return "".join(parts)
 
 
 def build_cover_extract_command(

requirements.txt

@@ -13,7 +13,3 @@ pytest-cov==6.0.0
 
 # 工具
 python-dotenv==1.0.1
-
-# AI 数字人封面智能选帧（cover_frame_scorer 用 cv2/numpy 做清晰度/亮度/色彩评分）
-numpy==1.26.4
-opencv-python-headless==4.10.0.84

tests/unit/domain/test_generated_video.py

@@ -87,19 +87,28 @@ class TestGeneratedVideoCreate:
         assert v.file_url == "http://x/v"
 
     def test_create_empty_project_id(self):
-        """空 project_id 允许（AI数字人无项目场景）."""
-        v = GeneratedVideo.create("", "t1", "v", "http://x/v")
-        assert v.project_id == ""
+        """空 project_id 无效."""
+        try:
+            GeneratedVideo.create("", "t1", "v", "http://x/v")
+            assert False
+        except ValueError as e:
+            assert "project_id" in str(e)
 
     def test_create_whitespace_project_id(self):
-        """纯空白 project_id 归一化为空串."""
-        v = GeneratedVideo.create("   ", "t1", "v", "http://x/v")
-        assert v.project_id == ""
+        """纯空白 project_id 无效."""
+        try:
+            GeneratedVideo.create("   ", "t1", "v", "http://x/v")
+            assert False
+        except ValueError as e:
+            assert "project_id" in str(e)
 
     def test_create_empty_task_id(self):
-        """空 generation_task_id 允许."""
-        v = GeneratedVideo.create("p1", "", "v", "http://x/v")
-        assert v.generation_task_id == ""
+        """空 generation_task_id 无效."""
+        try:
+            GeneratedVideo.create("p1", "", "v", "http://x/v")
+            assert False
+        except ValueError as e:
+            assert "generation_task_id" in str(e)
 
     def test_create_empty_name(self):
         """空 name 无效."""

tests/unit/test_ai_avatar_emotion_tts_lipsync.py

@@ -262,8 +262,8 @@ def test_smart_cover_selects_best_frame_and_persists():
     score_patch.assert_called_once()
     # 验证使用了增大的轮询参数
     call_kwargs = mk.extract_frames.call_args
-    assert call_kwargs.kwargs.get("poll_interval") == 1.0 or call_kwargs[1].get("poll_interval") == 1.0
-    assert call_kwargs.kwargs.get("max_poll_attempts") == 15 or call_kwargs[1].get("max_poll_attempts") == 15
+    assert call_kwargs.kwargs.get("poll_interval") == 3.0 or call_kwargs[1].get("poll_interval") == 3.0
+    assert call_kwargs.kwargs.get("max_poll_attempts") == 20 or call_kwargs[1].get("max_poll_attempts") == 20
 
 
 def test_smart_cover_returns_empty_when_mediakit_unavailable():
@@ -334,8 +334,8 @@ def test_extract_frames_uses_extended_poll_params():
         cov.select_best_cover_frame("https://other/avatar.mp4", max_frames=3)
 
     call_kwargs = mk.extract_frames.call_args
-    assert call_kwargs.kwargs.get("poll_interval") == 1.0 or call_kwargs[1].get("poll_interval") == 1.0
-    assert call_kwargs.kwargs.get("max_poll_attempts") == 15 or call_kwargs[1].get("max_poll_attempts") == 15
+    assert call_kwargs.kwargs.get("poll_interval") == 3.0 or call_kwargs[1].get("poll_interval") == 3.0
+    assert call_kwargs.kwargs.get("max_poll_attempts") == 20 or call_kwargs[1].get("max_poll_attempts") == 20
     assert call_kwargs.kwargs.get("max_retries") == 1 or call_kwargs[1].get("max_retries") == 1
 
 

tests/unit/test_ai_avatar_render_routes.py

@@ -258,9 +258,8 @@ class TestBrollOverlayFilter:
     def test_empty_segments_returns_empty(self):
         from packages.domain.video_filter_builder import build_broll_overlay_filter
 
-        result, label = build_broll_overlay_filter([], 30.0)
+        result = build_broll_overlay_filter([], 30.0)
         assert result == ""
-        assert label is None
 
     def test_pip_mode_generates_overlay(self):
         from packages.domain.video_filter_builder import build_broll_overlay_filter
@@ -276,9 +275,8 @@ class TestBrollOverlayFilter:
                 "pip_scale": 0.3,
             }
         ]
-        result, label = build_broll_overlay_filter(segments, 30.0)
+        result = build_broll_overlay_filter(segments, 30.0)
         assert "overlay" in result or "scale=" in result
-        assert label == "vout"
 
     def test_fullscreen_mode_generates_concat(self):
         from packages.domain.video_filter_builder import build_broll_overlay_filter
@@ -292,9 +290,8 @@ class TestBrollOverlayFilter:
                 "end_time": 10.0,
             }
         ]
-        result, label = build_broll_overlay_filter(segments, 30.0)
+        result = build_broll_overlay_filter(segments, 30.0)
         assert "trim" in result or "concat" in result
-        assert label == "vout_fs"
 
     def test_cover_extract_command(self):
         from packages.domain.video_filter_builder import build_cover_extract_command

tests/unit/test_ai_avatar_render_service.py

@@ -547,7 +547,7 @@ class TestAiAvatarRenderService:
         with (
             patch.object(svc, "_download_video", return_value="/tmp/video.mp4"),
             patch.object(svc, "_upload_to_oss", side_effect=lambda path, key: f"https://oss/{key}"),
-            patch("subprocess.run") as mock_run,
+            patch("os.system", return_value=0),
             patch("tempfile.TemporaryDirectory") as tmpdir_mock,
             patch(
                 "app.services.ai_avatar_cover_service.generate_smart_cover", return_value="https://oss/smart_cover.jpg"
@@ -557,9 +557,6 @@ class TestAiAvatarRenderService:
                 "packages.adapters.sqlalchemy_impl.generated_video_repository.SQLAlchemyGeneratedVideoRepository"
             ) as repo_cls,
         ):
-            import subprocess as _sp
-
-            mock_run.return_value = _sp.CompletedProcess(args=[], returncode=0, stdout="", stderr="")
             import tempfile as _tf
 
             tmpdir_mock.return_value.__enter__ = MagicMock(return_value="/tmp/testdir")
@@ -608,13 +605,10 @@ class TestAiAvatarRenderService:
         with (
             patch.object(svc, "_download_video", return_value="/tmp/video.mp4"),
             patch.object(svc, "_upload_to_oss", side_effect=lambda path, key: f"https://oss/{key}"),
-            patch("subprocess.run") as mock_run,
+            patch("os.system", return_value=0),
             patch("tempfile.TemporaryDirectory") as tmpdir_mock,
             patch("app.services.ai_avatar_cover_service.generate_smart_cover", side_effect=RuntimeError("DB error")),
         ):
-            import subprocess as _sp
-
-            mock_run.return_value = _sp.CompletedProcess(args=[], returncode=0, stdout="", stderr="")
             tmpdir_mock.return_value.__enter__ = MagicMock(return_value="/tmp/testdir")
             tmpdir_mock.return_value.__exit__ = MagicMock(return_value=False)
             svc.execute_render("render-clip-fail")

tests/unit/test_cover_frame_scorer.py

@@ -43,7 +43,7 @@ class TestScoreFrame:
 
     @requires_cv2
     def test_clear_image_high_score(self):
-        """清晰、亮度适中、色彩丰富的图像应得较高分."""
+        """清晰、亮度适中、色彩丰富的图像应得高分."""
         # 创建一个清晰的渐变图像（色彩丰富、亮度适中）
         img = np.zeros((100, 100, 3), dtype=np.uint8)
         for i in range(100):
@@ -53,8 +53,7 @@ class TestScoreFrame:
         from packages.shared.cover_frame_scorer import score_frame
 
         score = score_frame(img)
-        # 渐变图清晰度中等+亮度尚可+色彩有变化，分数应明显高于模糊/全黑/全白
-        assert 40.0 <= score <= 100.0, f"清晰图像应得较高分，实际: {score}"
+        assert 50.0 <= score <= 100.0, f"清晰图像应得高分，实际: {score}"
 
     @requires_cv2
     def test_blurry_image_low_clarity(self):
@@ -77,8 +76,8 @@ class TestScoreFrame:
         from packages.shared.cover_frame_scorer import score_frame
 
         score = score_frame(img)
-        # 全黑：清晰度 0，亮度偏离130扣约24分，色彩 0 → 得分约0~7，允许cv2内部微小浮点差异
-        assert score <= 10.0, f"全黑图像应接近 0 分，实际: {score}"
+        # 全黑：清晰度 0，亮度 0，色彩 0
+        assert score <= 5.0, f"全黑图像应接近 0 分，实际: {score}"
 
     @requires_cv2
     def test_bright_image_low_brightness(self):

tests/unit/test_dedup_v2.py

@@ -180,26 +180,28 @@ class TestDetectKeyframeTimestamps:
 
     def test_cannot_open_video_raises(self):
         """无法打开视频时抛出 RuntimeError."""
+        cv2_mock = _dedup_mod.cv2
         mock_cap = MagicMock()
         mock_cap.isOpened.return_value = False
+        cv2_mock.VideoCapture.return_value = mock_cap
 
         import pytest
 
-        with patch.object(_dedup_mod.cv2, "VideoCapture", return_value=mock_cap):
-            with pytest.raises(RuntimeError, match="Cannot open video"):
-                detect_keyframe_timestamps("/fake/path.mp4")
+        with pytest.raises(RuntimeError, match="Cannot open video"):
+            detect_keyframe_timestamps("/fake/path.mp4")
 
     def test_zero_duration_returns_empty(self):
         """视频时长为 0 时返回空列表."""
+        cv2_mock = _dedup_mod.cv2
         mock_cap = MagicMock()
         mock_cap.isOpened.return_value = True
         # cv2.CAP_PROP_FPS etc. are Mock objects; configure get() to return 0 for frame_count
         mock_cap.get.return_value = 0
         mock_cap.read.return_value = (False, None)
+        cv2_mock.VideoCapture.return_value = mock_cap
 
-        with patch.object(_dedup_mod.cv2, "VideoCapture", return_value=mock_cap):
-            result = detect_keyframe_timestamps("/fake/zero.mp4")
-            assert result == []
+        result = detect_keyframe_timestamps("/fake/zero.mp4")
+        assert result == []
 
     def test_function_signature(self):
         """验证函数签名和默认参数."""

tests/unit/test_generated_video.py

@@ -47,35 +47,32 @@ class TestGeneratedVideoCreate:
         assert video.file_url == "https://example.com/video.mp4"
         assert video.user_id == "user1"
 
-    def test_create_empty_project_id_allowed(self):
-        """project_id 允许为空（AI数字人等无项目场景）。"""
-        video = GeneratedVideo.create(
-            project_id="",
-            generation_task_id="task1",
-            name="视频",
-            file_url="https://example.com/v.mp4",
-        )
-        assert video.project_id == ""
+    def test_create_empty_project_id_raises(self):
+        with pytest.raises(ValueError, match="project_id cannot be empty"):
+            GeneratedVideo.create(
+                project_id="",
+                generation_task_id="task1",
+                name="视频",
+                file_url="https://example.com/v.mp4",
+            )
 
-    def test_create_whitespace_project_id_normalized_to_empty(self):
-        """project_id 纯空白会被 strip 为空串，不抛异常。"""
-        video = GeneratedVideo.create(
-            project_id="   ",
-            generation_task_id="task1",
-            name="视频",
-            file_url="https://example.com/v.mp4",
-        )
-        assert video.project_id == ""
+    def test_create_whitespace_project_id_raises(self):
+        with pytest.raises(ValueError, match="project_id cannot be empty"):
+            GeneratedVideo.create(
+                project_id="   ",
+                generation_task_id="task1",
+                name="视频",
+                file_url="https://example.com/v.mp4",
+            )
 
-    def test_create_empty_generation_task_id_allowed(self):
-        """generation_task_id 允许为空（兼容部分异步链路）。"""
-        video = GeneratedVideo.create(
-            project_id="proj1",
-            generation_task_id="",
-            name="视频",
-            file_url="https://example.com/v.mp4",
-        )
-        assert video.generation_task_id == ""
+    def test_create_empty_generation_task_id_raises(self):
+        with pytest.raises(ValueError, match="generation_task_id cannot be empty"):
+            GeneratedVideo.create(
+                project_id="proj1",
+                generation_task_id="",
+                name="视频",
+                file_url="https://example.com/v.mp4",
+            )
 
     def test_create_empty_name_raises(self):
         with pytest.raises(ValueError, match="name cannot be empty"):

tests/unit/test_generated_video_domain.py

@@ -75,35 +75,32 @@ class TestGeneratedVideoCreate:
         assert video.file_url == "https://example.com/out.mp4"
         assert video.user_id == "user_003"
 
-    def test_create_empty_project_id_allowed(self):
-        """project_id 允许为空（AI数字人等无项目场景）。"""
-        video = GeneratedVideo.create(
-            project_id="",
-            generation_task_id="t",
-            name="n",
-            file_url="u",
-        )
-        assert video.project_id == ""
+    def test_create_empty_project_id_raises(self):
+        with pytest.raises(ValueError, match="project_id"):
+            GeneratedVideo.create(
+                project_id="",
+                generation_task_id="t",
+                name="n",
+                file_url="u",
+            )
 
-    def test_create_whitespace_project_id_normalized(self):
-        """project_id 纯空白归一化为空串。"""
-        video = GeneratedVideo.create(
-            project_id="   ",
-            generation_task_id="t",
-            name="n",
-            file_url="u",
-        )
-        assert video.project_id == ""
+    def test_create_whitespace_project_id_raises(self):
+        with pytest.raises(ValueError, match="project_id"):
+            GeneratedVideo.create(
+                project_id="   ",
+                generation_task_id="t",
+                name="n",
+                file_url="u",
+            )
 
-    def test_create_empty_generation_task_id_allowed(self):
-        """generation_task_id 允许为空。"""
-        video = GeneratedVideo.create(
-            project_id="p",
-            generation_task_id="",
-            name="n",
-            file_url="u",
-        )
-        assert video.generation_task_id == ""
+    def test_create_empty_generation_task_id_raises(self):
+        with pytest.raises(ValueError, match="generation_task_id"):
+            GeneratedVideo.create(
+                project_id="p",
+                generation_task_id="",
+                name="n",
+                file_url="u",
+            )
 
     def test_create_empty_name_raises(self):
         with pytest.raises(ValueError, match="name"):

tests/unit/test_generated_video_verification_code.py

@@ -45,25 +45,25 @@ class TestGeneratedVideo:
         assert video.duplicate_of is None
         assert video.generation_params == {}
 
-    def test_create_empty_project_id_allowed(self):
-        """project_id 允许为空（AI数字人场景），空白归一化为空串."""
-        video = GeneratedVideo.create(
-            project_id="  ",
-            generation_task_id="t1",
-            name="v.mp4",
-            file_url="https://x.com/v.mp4",
-        )
-        assert video.project_id == ""
+    def test_create_empty_project_id_raises(self):
+        """空project_id抛异常."""
+        with pytest.raises(ValueError, match="project_id"):
+            GeneratedVideo.create(
+                project_id="  ",
+                generation_task_id="t1",
+                name="v.mp4",
+                file_url="https://x.com/v.mp4",
+            )
 
-    def test_create_empty_task_id_allowed(self):
-        """generation_task_id 允许为空."""
-        video = GeneratedVideo.create(
-            project_id="p1",
-            generation_task_id="",
-            name="v.mp4",
-            file_url="https://x.com/v.mp4",
-        )
-        assert video.generation_task_id == ""
+    def test_create_empty_task_id_raises(self):
+        """空generation_task_id抛异常."""
+        with pytest.raises(ValueError, match="generation_task_id"):
+            GeneratedVideo.create(
+                project_id="p1",
+                generation_task_id="",
+                name="v.mp4",
+                file_url="https://x.com/v.mp4",
+            )
 
     def test_create_empty_name_raises(self):
         """空name抛异常."""

tests/unit/test_lipsync_speed_optimization.py

@@ -35,10 +35,7 @@ class TestFFmpegPresetOptimization:
             final_label=None,
             output_path="/tmp/output.mp4",
         )
-        # cmd 现在是 list[str]；preset 与值是相邻两个元素
-        assert "-preset" in cmd, f"期望包含 -preset，实际命令: {cmd}"
-        preset_idx = cmd.index("-preset")
-        assert cmd[preset_idx + 1] == "veryfast", f"期望 veryfast，实际: {cmd}"
+        assert "-preset veryfast" in cmd, f"期望 -preset veryfast，实际命令: {cmd}"
 
     def test_preset_veryfast_with_filter(self):
         """带滤镜场景下也必须使用 veryfast."""
@@ -52,8 +49,7 @@ class TestFFmpegPresetOptimization:
             final_label="[v]",
             output_path="/tmp/output.mp4",
         )
-        assert "-preset" in cmd
-        assert cmd[cmd.index("-preset") + 1] == "veryfast"
+        assert "-preset veryfast" in cmd
         assert "-filter_complex" in cmd
 
     def test_preset_not_fast(self):
@@ -69,11 +65,11 @@ class TestFFmpegPresetOptimization:
             output_path="/tmp/output.mp4",
         )
         # 确保是 veryfast 而不是 fast
-        assert "-preset" in cmd
-        preset_idx = cmd.index("-preset")
-        assert cmd[preset_idx + 1] == "veryfast"
-        # 禁止 fast 单独作为 preset 值（veryfast 包含 "fast" 子串，不影响）
-        assert cmd[preset_idx + 1] != "fast"
+        assert "-preset veryfast" in cmd
+        # 排除 "fast" 单独出现（veryfast 包含 fast 子串，需精确判断）
+        parts = cmd.split()
+        preset_idx = parts.index("-preset")
+        assert parts[preset_idx + 1] == "veryfast"
 
 
 # ═══════════════════════════════════════════════════════════════════════════════

tests/unit/test_sentence_timings.py

@@ -1,171 +0,0 @@
-"""Tests for sentence timing functions in lipsync_tts."""
-
-import os
-import subprocess
-import tempfile
-import unittest
-from unittest.mock import MagicMock, patch
-
-from apps.api.app.tasks.lipsync_tts import (
-    _compute_sentence_timings,
-    _estimate_sentence_timings_by_chars,
-    _split_script_into_sentences,
-)
-
-
-class TestSplitScriptIntoSentences(unittest.TestCase):
-    """Tests for _split_script_into_sentences."""
-
-    def test_empty_string(self):
-        self.assertEqual(_split_script_into_sentences(""), [])
-
-    def test_none(self):
-        self.assertEqual(_split_script_into_sentences(None), [])
-
-    def test_whitespace_only(self):
-        self.assertEqual(_split_script_into_sentences("   \n  "), [])
-
-    def test_single_sentence(self):
-        self.assertEqual(_split_script_into_sentences("你好世界。"), ["你好世界"])
-
-    def test_multiple_sentences_chinese(self):
-        result = _split_script_into_sentences("第一句。第二句！第三句？")
-        self.assertEqual(result, ["第一句", "第二句", "第三句"])
-
-    def test_english_punctuation(self):
-        result = _split_script_into_sentences("Hello World! How are you?")
-        self.assertEqual(result, ["Hello World", "How are you"])
-
-    def test_semicolons(self):
-        result = _split_script_into_sentences("第一部分；第二部分;第三部分")
-        self.assertEqual(result, ["第一部分", "第二部分", "第三部分"])
-
-    def test_newlines(self):
-        result = _split_script_into_sentences("第一行\n第二行\n第三行")
-        self.assertEqual(result, ["第一行", "第二行", "第三行"])
-
-    def test_no_trailing_punctuation(self):
-        result = _split_script_into_sentences("没有标点的句子")
-        self.assertEqual(result, ["没有标点的句子"])
-
-
-class TestEstimateSentenceTimingsByChars(unittest.TestCase):
-    """Tests for _estimate_sentence_timings_by_chars."""
-
-    def test_empty_sentences(self):
-        self.assertEqual(_estimate_sentence_timings_by_chars([], 10.0), [])
-
-    def test_zero_duration(self):
-        self.assertEqual(_estimate_sentence_timings_by_chars(["hello"], 0), [])
-
-    def test_negative_duration(self):
-        self.assertEqual(_estimate_sentence_timings_by_chars(["hello"], -5.0), [])
-
-    def test_single_sentence(self):
-        result = _estimate_sentence_timings_by_chars(["hello"], 10.0)
-        self.assertEqual(len(result), 1)
-        self.assertAlmostEqual(result[0]["start_time"], 0.0)
-        self.assertAlmostEqual(result[0]["end_time"], 10.0)
-
-    def test_two_equal_sentences(self):
-        result = _estimate_sentence_timings_by_chars(["你好", "世界"], 10.0)
-        self.assertEqual(len(result), 2)
-        self.assertAlmostEqual(result[0]["start_time"], 0.0)
-        self.assertAlmostEqual(result[0]["end_time"], 5.0)
-        self.assertAlmostEqual(result[1]["start_time"], 5.0)
-        self.assertAlmostEqual(result[1]["end_time"], 10.0)
-
-    def test_unequal_char_distribution(self):
-        result = _estimate_sentence_timings_by_chars(["ABCD", "EF"], 9.0)
-        self.assertEqual(len(result), 2)
-        self.assertAlmostEqual(result[0]["start_time"], 0.0)
-        self.assertAlmostEqual(result[0]["end_time"], 6.0)  # 4/6 * 9 = 6
-        self.assertAlmostEqual(result[1]["start_time"], 6.0)
-        self.assertAlmostEqual(result[1]["end_time"], 9.0)
-
-    def test_timing_structure(self):
-        result = _estimate_sentence_timings_by_chars(["句子一", "句子二"], 6.0)
-        for item in result:
-            self.assertIn("index", item)
-            self.assertIn("text", item)
-            self.assertIn("start_time", item)
-            self.assertIn("end_time", item)
-
-
-class TestComputeSentenceTimings(unittest.TestCase):
-    """Tests for _compute_sentence_timings."""
-
-    def test_empty_script_returns_empty(self):
-        self.assertEqual(_compute_sentence_timings(b"fake_audio", "", 10.0), [])
-
-    def test_none_script_returns_empty(self):
-        self.assertEqual(_compute_sentence_timings(b"fake_audio", None, 10.0), [])
-
-    @patch("os.unlink")
-    @patch.object(tempfile, "NamedTemporaryFile")
-    @patch.object(subprocess, "run")
-    def test_silence_detection_insufficient_fallback(self, mock_run, mock_tmpfile, mock_unlink):
-        """When silence detection finds too few points, fallback to char estimation."""
-        mock_run.return_value = MagicMock(stderr="", returncode=0)
-        mock_tmp = MagicMock()
-        mock_tmp.name = "/tmp/fake.mp3"
-        mock_tmp.__enter__ = MagicMock(return_value=mock_tmp)
-        mock_tmp.__exit__ = MagicMock(return_value=False)
-        mock_tmpfile.return_value = mock_tmp
-
-        result = _compute_sentence_timings(b"fake_audio", "第一句。第二句。第三句。", 10.0)
-
-        # Should fallback to char estimation with 3 sentences
-        self.assertEqual(len(result), 3)
-        self.assertAlmostEqual(result[0]["start_time"], 0.0)
-
-    @patch("os.unlink")
-    @patch.object(tempfile, "NamedTemporaryFile")
-    @patch.object(subprocess, "run")
-    def test_silence_detection_with_enough_points(self, mock_run, mock_tmpfile, mock_unlink):
-        """When silence detection finds enough points, use them for boundaries."""
-        mock_run.return_value = MagicMock(
-            stderr="[silencedetect] silence_end: 3.5 | silence_duration: 0.4\n"
-            "[silencedetect] silence_end: 7.0 | silence_duration: 0.3\n",
-            returncode=0,
-        )
-        mock_tmp = MagicMock()
-        mock_tmp.name = "/tmp/fake.mp3"
-        mock_tmp.__enter__ = MagicMock(return_value=mock_tmp)
-        mock_tmp.__exit__ = MagicMock(return_value=False)
-        mock_tmpfile.return_value = mock_tmp
-
-        result = _compute_sentence_timings(b"fake_audio", "第一句。第二句。第三句。", 10.0)
-
-        self.assertEqual(len(result), 3)
-        self.assertAlmostEqual(result[0]["start_time"], 0.0)
-        self.assertAlmostEqual(result[0]["end_time"], 3.5)
-        self.assertAlmostEqual(result[1]["start_time"], 3.5)
-        self.assertAlmostEqual(result[1]["end_time"], 7.0)
-        self.assertAlmostEqual(result[2]["start_time"], 7.0)
-        self.assertAlmostEqual(result[2]["end_time"], 10.0)
-
-    @patch("os.unlink")
-    @patch.object(tempfile, "NamedTemporaryFile")
-    @patch.object(subprocess, "run")
-    def test_ffmpeg_exception_fallback(self, mock_run, mock_tmpfile, mock_unlink):
-        """When ffmpeg raises an exception, fallback to char estimation."""
-        mock_run.side_effect = Exception("ffmpeg not found")
-        mock_tmp = MagicMock()
-        mock_tmp.name = "/tmp/fake.mp3"
-        mock_tmp.__enter__ = MagicMock(return_value=mock_tmp)
-        mock_tmp.__exit__ = MagicMock(return_value=False)
-        mock_tmpfile.return_value = mock_tmp
-
-        result = _compute_sentence_timings(b"fake_audio", "句子一。句子二。", 6.0)
-
-        # Should fallback to char estimation
-        self.assertEqual(len(result), 2)
-        self.assertAlmostEqual(result[0]["start_time"], 0.0)
-        self.assertAlmostEqual(result[0]["end_time"], 3.0)
-        self.assertAlmostEqual(result[1]["start_time"], 3.0)
-        self.assertAlmostEqual(result[1]["end_time"], 6.0)
-
-
-if __name__ == "__main__":
-    unittest.main()

tests/unit/test_video_filter_builder.py

@@ -902,10 +902,9 @@ class TestResolveFontPath(unittest.TestCase):
 
     @patch("os.path.isfile")
     def test_unknown_font_fallback(self, mock_isfile):
-        # DejaVuSans 已从 fallback 列表移除（不支持 CJK），用 VF 路径模拟
-        mock_isfile.side_effect = lambda p: "NotoSansSC-VF" in p
+        mock_isfile.side_effect = lambda p: "DejaVu" in p
         result = _resolve_font_path("UnknownFont")
-        self.assertIn("NotoSansSC-VF", result)
+        self.assertIn("DejaVu", result)
 
     @patch("os.path.isfile")
     def test_no_fonts_available(self, mock_isfile):
@@ -928,11 +927,9 @@ class TestResolveFontPath(unittest.TestCase):
 
     @patch("os.path.isfile")
     def test_font_fallback_skips_nonexistent(self, mock_isfile):
-        # 所有中文字体路径都不存在时，fallback 返回第一个存在的文件；
-        # DejaVuSans 已从列表移除（不支持 CJK），使用 VF 字体路径模拟存在文件
-        mock_isfile.side_effect = lambda p: "NotoSansSC-VF" in p
+        mock_isfile.side_effect = lambda p: "DejaVu" in p
         result = _resolve_font_path("不存在字体")
-        self.assertIn("NotoSansSC-VF", result)
+        self.assertIn("DejaVu", result)
 
 
 class TestDrawtextFontFileIncluded(unittest.TestCase):
@@ -1031,14 +1028,6 @@ class TestDrawtextBoldFalse(unittest.TestCase):
         self.assertIsNotNone(result)
         self.assertNotIn("font=bold", result)
 
-    def test_bold_true_does_not_use_font_bold_param(self):
-        """粗体模式不得使用 `font=bold`——该参数无效，会导致 filter_complex 解析失败（exit 234）。"""
-        result = build_title_drawtext_filter({"text": "标题", "bold": True})
-        self.assertIsNotNone(result)
-        self.assertNotIn("font=bold", result)
-        # 粗体应通过 borderw 实现
-        self.assertIn("borderw=", result)
-
 
 class TestDrawtextPositionBranches(unittest.TestCase):
     """位置相关分支覆盖。"""
@@ -1065,23 +1054,12 @@ class TestDrawtextPositionBranches(unittest.TestCase):
         self.assertIn("y=h-text_h-50", result)
 
     @patch("packages.domain.video_filter_builder._resolve_font_path")
-    def test_position_custom_with_percentage_coords(self, mock_font):
-        """自定义位置：百分比坐标转换为 drawtext 表达式."""
-        mock_font.return_value = ""
-        # pos_x=50, pos_y=30 → x=(w-text_w)*0.5000, y=(h-text_h)*0.3000
-        result = build_title_drawtext_filter({"text": "标题", "position": "custom", "pos_x": 50, "pos_y": 30})
-        self.assertIsNotNone(result)
-        self.assertIn("x=(w-text_w)*0.5000", result)
-        self.assertIn("y=(h-text_h)*0.3000", result)
-
-    @patch("packages.domain.video_filter_builder._resolve_font_path")
-    def test_position_custom_clamped_to_100(self, mock_font):
-        """自定义位置：超过100的坐标被截断到100%."""
+    def test_position_custom_with_float_coords(self, mock_font):
         mock_font.return_value = ""
         result = build_title_drawtext_filter({"text": "标题", "position": "custom", "pos_x": 100.7, "pos_y": 200.3})
         self.assertIsNotNone(result)
-        self.assertIn("x=(w-text_w)*1.0000", result)
-        self.assertIn("y=(h-text_h)*1.0000", result)
+        self.assertIn("x=100", result)
+        self.assertIn("y=200", result)
 
     @patch("packages.domain.video_filter_builder._resolve_font_path")
     def test_position_custom_bool_coords_fallback(self, mock_font):