fix(ai-avatar): 对口型/渲染轮询增加超时保护

对口型最多轮询5分钟（100次×3s），渲染最多轮询10分钟（200次×3s），
超时后停止轮询并显示错误提示，避免任务异常时前端无限轮询卡死。

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

@@ -191,6 +191,7 @@ def retry_render_job(
     return AiAvatarRenderJobResponse.model_validate(job)
 
 
+
 # ── POST /smart-cover — 智能获取封面（MediaKit 抽帧 + 评分选帧）────────
 
 
@@ -219,9 +220,7 @@ def generate_avatar_smart_cover(
     except Exception as exc:
         logger.error(
             "智能封面生成异常: user=%s video_url=%s err=%s",
-            current_user.user.id,
-            video_url[:80],
-            exc,
+            current_user.user.id, video_url[:80], exc,
             exc_info=True,
         )
         cover_url = ""
@@ -234,72 +233,3 @@ def generate_avatar_smart_cover(
         )
     logger.info("智能封面生成成功: user=%s cover_url=%s", current_user.user.id, cover_url[:120])
     return SmartCoverResponse(cover_url=cover_url, status="completed")
-
-
-# ── POST /{job_id}/smart-cover — 从最终成片智能抽封面（步骤②）────────
-
-
-@router.post("/{job_id}/smart-cover", response_model=SmartCoverResponse)
-def generate_render_smart_cover(
-    job_id: str,
-    current_user: AuthenticatedUser = Depends(get_current_user),
-    db: Session = Depends(get_db_session),
-):
-    """从最终渲染成片智能抽帧生成封面（MediaKit 抽帧 + 评分选最佳帧 + 转存 OSS）.
-
-    - 必须等渲染任务 completed 后才可调用（否则返回 400）
-    - 生成成功后自动更新 render_job 的 cover_config 与 output_cover_url
-    """
-    from app.services.ai_avatar_render_service import AiAvatarRenderService
-
-    svc = AiAvatarRenderService(db)
-    job = svc.get_render_job(job_id, current_user.user.id)
-    if job is None:
-        raise HTTPException(status_code=404, detail="渲染任务不存在")
-    if job.status != "completed":
-        raise HTTPException(status_code=400, detail="请先完成视频生成")
-    video_url = (job.output_video_url or "").strip()
-    if not video_url:
-        raise HTTPException(status_code=400, detail="渲染成片视频 URL 为空")
-
-    try:
-        # 成片已叠加标题，不传 title_config 避免双重叠加
-        cover_url = generate_smart_cover(video_url, job_id=job_id, max_frames=5)
-    except Exception as exc:
-        logger.error(
-            "渲染成片智能封面生成异常: user=%s render_id=%s video_url=%s err=%s",
-            current_user.user.id,
-            job_id,
-            video_url[:80],
-            exc,
-            exc_info=True,
-        )
-        cover_url = ""
-
-    if not cover_url:
-        return SmartCoverResponse(
-            cover_url="",
-            status="fallback_failed",
-            message="智能抽帧失败（MediaKit 不可用或抽帧异常），请稍后重试",
-        )
-
-    # 更新 render_job 的封面字段（异步写入 DB；失败不影响返回）
-    try:
-        job.cover_config = {
-            **(job.cover_config if isinstance(job.cover_config, dict) else {}),
-            "mode": "auto_frame",
-            "url": cover_url,
-        }
-        job.output_cover_url = cover_url
-        job.updated_at = datetime.now(timezone.utc)
-        db.commit()
-    except Exception as exc:
-        logger.warning("更新 render_job 封面字段失败（不影响返回）: job_id=%s err=%s", job_id, exc)
-
-    logger.info(
-        "渲染成片智能封面生成成功: user=%s render_id=%s cover_url=%s",
-        current_user.user.id,
-        job_id,
-        cover_url[:120],
-    )
-    return SmartCoverResponse(cover_url=cover_url, status="completed")

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_render_service.py

@@ -25,7 +25,6 @@ from packages.adapters.sqlalchemy_impl.models import (
     ScriptModel,
 )
 from packages.domain.video_filter_builder import (
-    build_broll_overlay_filter,
     build_title_drawtext_filter,
 )
 from packages.shared.storage import get_shared_storage_service
@@ -197,8 +196,9 @@ class AiAvatarRenderService:
         1. 下载对口型输出视频 (20%)
         2. 构建 FFmpeg 滤镜链 (40%)
         3. 执行 FFmpeg 渲染 (80%)
-        4. 上传到 OSS (95%) — 封面不再自动生成，改由前端主动抽帧
-        5. 更新任务状态 (100%)
+        4. 提取封面 (90%)
+        5. 上传到 OSS (95%)
+        6. 更新任务状态 (100%)
         """
         job = self.db.query(AiAvatarRenderJob).filter(AiAvatarRenderJob.id == job_id).first()
         if job is None:
@@ -228,49 +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()
@@ -310,23 +288,64 @@ class AiAvatarRenderService:
                 job.progress = 80
                 self.db.commit()
 
-                # 4/5. 上传成片到 OSS (95%) —— 已砍掉自动抽封面逻辑（步骤⑤）；
-                # 封面由前端在渲染完成后通过 /smart-cover 接口主动从成片抽帧，不阻塞渲染链路。
+                # 4. 提取封面 (90%)
+                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_path = ""
+
+                job.progress = 90
+                self.db.commit()
+
+                # 5. 上传到 OSS (95%)
                 output_video_url = self._upload_to_oss(output_video_path, f"ai-avatar/{job_id}/output.mp4")
                 job.output_video_url = output_video_url
 
-                # 封面透传：如果用户已在 cover_config 中选定封面 URL（mode=upload 的自定义上传 或
-                # mode=auto_frame 已有的智能封面结果），直接透传到 output_cover_url，不再重新截帧。
-                if isinstance(job.cover_config, dict):
-                    _pre_cover_url = (
-                        job.cover_config.get("url")
-                        or job.cover_config.get("imageUrl")
-                        or job.cover_config.get("cover_url")
-                        or ""
-                    )
-                    if _pre_cover_url:
-                        job.output_cover_url = _pre_cover_url
-                        logger.info("[数字人渲染] 使用用户已选定封面 URL: job_id=%s", job_id)
+                # 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）；
+                # MediaKit 不可用时回退到 FFmpeg 已按 cover_config 抽取的 cover_path
+                smart_cover_url = ""
+                if output_video_url:
+                    try:
+                        from app.services.ai_avatar_cover_service import (
+                            generate_smart_cover,
+                        )
+
+                        smart_cover_url = generate_smart_cover(
+                            output_video_url,
+                            job_id=job_id,
+                            max_frames=5,
+                            # 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
+                        )
+                    except Exception:
+                        logger.warning("智能封面(MediaKit)失败，回退 FFmpeg 封面 job_id=%s", job_id, exc_info=True)
+
+                if smart_cover_url:
+                    job.output_cover_url = smart_cover_url
+                elif cover_path:
+                    output_cover_url = self._upload_to_oss(cover_path, f"ai-avatar/{job_id}/cover.jpg")
+                    job.output_cover_url = output_cover_url
 
                 # 获取输出视频时长
                 job.output_duration = lipsync_job.output_duration
@@ -408,37 +427,6 @@ 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,
         *,
@@ -462,16 +450,7 @@ class AiAvatarRenderService:
                 cmd.extend(["-i", asset_url])
 
         if filter_complex and final_label:
-            cmd.extend(
-                [
-                    "-filter_complex",
-                    filter_complex,
-                    "-map",
-                    f"[{final_label}]",
-                    "-map",
-                    "0:a?",
-                ]
-            )
+            cmd.extend(["-filter_complex", filter_complex, "-map", f"[{final_label}]"])
         elif filter_complex:
             cmd.extend(["-filter_complex", filter_complex])
 
@@ -483,16 +462,47 @@ class AiAvatarRenderService:
                 "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
+        else:
+            timestamp = cover_config.get("timestamp", 0.0)
+            width = cover_config.get("width", 0)
+            height = cover_config.get("height", 0)
+
+        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
+
     def _upload_to_oss(self, local_path: str, oss_key: str) -> str:
         """上传文件到 OSS，返回 URL.
 

apps/api/app/services/lipsync_service.py

@@ -198,12 +198,6 @@ class LipsyncService:
         self.db.add(job)
         self.db.flush()
 
-        # ⚠️ 必须先 commit 再发 Celery 任务，避免事务竞态：
-        # worker 是独立进程+独立DB连接，任务被消费（<4ms）时若本事务还未提交，
-        # worker 查询 job 会返回 None → 静默 return 不重试，job 永远卡在 tts_processing。
-        self.db.commit()
-        self.db.refresh(job)
-
         if is_tts_mode:
             # 2a. TTS 模式：dispatch Celery 异步任务处理 TTS 合成 + MediaKit 提交
             try:
@@ -229,7 +223,6 @@ class LipsyncService:
                 job.error_message = f"Celery 任务投递失败: {exc}"
                 job.error_code = "AsyncDispatchFailed"
                 job.updated_at = datetime.now(timezone.utc)
-                self.db.commit()  # 投递失败也要落库失败状态
         else:
             # 2b. 直接音频模式：同步签名并提交 MediaKit
             video_url = self._sign_media_url(video_url)
@@ -247,15 +240,15 @@ class LipsyncService:
                 job.mediakit_task_id = result["task_id"]
                 job.status = "submitted"
                 job.submitted_at = datetime.now(timezone.utc)
-                self.db.commit()  # submitted 状态落库
             except MediaKitError as exc:
                 job.status = "failed"
                 job.error_message = str(exc)
                 job.error_code = exc.code
                 logger.error("提交对口型任务失败: %s", exc)
-                self.db.commit()
                 raise
 
+        self.db.commit()
+        self.db.refresh(job)
         return job
 
     # ── 查询任务 ──────────────────────────────────────────────────────────
@@ -320,26 +313,11 @@ class LipsyncService:
         if mk_status == STATUS_COMPLETED:
             result = status_data.get("result", {})
             job.status = STATUS_COMPLETED
-            temp_url = result.get("video_url", "")
-            # 先以临时 URL 立即返回前端（前端可立即播放），再异步 Celery 任务转存自家 OSS（步骤⑦）
-            job.output_video_url = temp_url
+            output_url = result.get("video_url", "")
+            # MediaKit 输出为临时 URL，转存自家 OSS 防止过期（失败则回退临时 URL）
+            job.output_video_url = self._persist_output_video(output_url, job_id, user_id)
             job.output_duration = result.get("duration", 0.0)
             job.completed_at = datetime.now(timezone.utc)
-            job.updated_at = datetime.now(timezone.utc)
-            self.db.commit()
-            # 异步转存到自家 OSS（注意：必须在 commit 之后 dispatch，避免 commit 失败任务已发出）
-            try:
-                from app.tasks.lipsync_tts import persist_output_video_task
-
-                persist_output_video_task.apply_async(args=(job_id, user_id, temp_url))
-            except Exception as exc:
-                logger.warning(
-                    "提交输出视频异步转存任务失败，保留临时 URL: job_id=%s err=%s",
-                    job_id,
-                    exc,
-                )
-            self.db.refresh(job)
-            return job
         elif mk_status == STATUS_FAILED:
             error = status_data.get("error", {})
             job.status = "failed"

apps/api/app/tasks/lipsync_tts.py

@@ -54,170 +54,11 @@ 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",
-    max_retries=5,  # 事务竞态重试3次（job not found）+ TTS偶发错误2次
+    max_retries=2,
     default_retry_delay=30,
-    autoretry_for=(OSError, ConnectionError),  # 网络/连接错误自动重试
-    retry_backoff=True,
-    retry_backoff_max=30,
-    soft_time_limit=180,
-    time_limit=200,
 )
 def tts_synthesize_and_submit(
     self,
@@ -262,28 +103,7 @@ def tts_synthesize_and_submit(
         )
 
         if job is None:
-            # 事务竞态防御：API 在 commit 前投递了任务，worker 消费时事务尚未提交。
-            # Celery 内置 autoretry_for 不支持"业务条件重试"，这里手动 retry 3 次，
-            # 间隔递增（1s/3s/7s），让 API 事务有时间提交。
-            # max_retries 由 self.request(retries) 维护；默认 self.max_retries=3 由装饰器 soft_time_limit 下方指定。
-            retries = getattr(self.request, "retries", 0)
-            max_retries = 3
-            if retries < max_retries:
-                backoff = (2**retries) + (retries * 1)  # 1s, 3s, 7s
-                logger.warning(
-                    "[lipsync_tts] Job not found yet (retry %d/%d, backoff %ds): job_id=%s",
-                    retries + 1,
-                    max_retries,
-                    backoff,
-                    job_id,
-                )
-                self.db.close()
-                raise self.retry(countdown=backoff, max_retries=max_retries)
-            logger.error(
-                "[lipsync_tts] Job not found after %d retries, giving up: job_id=%s",
-                max_retries,
-                job_id,
-            )
+            logger.error("[lipsync_tts] Job not found: job_id=%s", job_id)
             return
 
         # 已取消的任务不再处理
@@ -292,13 +112,6 @@ def tts_synthesize_and_submit(
             return
 
         # 1. TTS 合成
-        logger.info(
-            "[lipsync_tts] 开始 TTS 合成: job_id=%s voice_id=%s text_len=%d speed=%.2f",
-            job_id,
-            voice_id,
-            len(script_text),
-            speed,
-        )
         try:
             cosyvoice = CosyVoiceService()
             result = cosyvoice.submit_synthesize_task(
@@ -334,113 +147,38 @@ def tts_synthesize_and_submit(
             db.commit()
             return
 
-        # 2. 下载 TTS 音频到内存（用于 2.5 静音检测；不转存自家 OSS，直接使用 CosyVoice 临时 URL）
-        audio_data: bytes | None = None
-        _st_tmp_path: str | None = None
+        # 2. 下载并转存到自家 OSS
         try:
             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,
             )
-            logger.info(
-                "[lipsync_tts] TTS 音频已下载到内存: job_id=%s size=%d",
-                job_id,
-                len(audio_data) if audio_data else 0,
-            )
+            from packages.shared.storage import get_shared_storage_service
+
+            storage = get_shared_storage_service()
+            storage_key = f"lipsync-tts/{user_id}/{job_id}.mp3"
+            permanent_url = storage.upload_file(io.BytesIO(audio_data), storage_key, content_type="audio/mpeg")
+            logger.info("[lipsync_tts] TTS 音频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
+            job.audio_url = permanent_url
         except Exception as exc:
-            # 下载失败：audio_data 保持 None，2.5 静音检测会跳过；后续仍用 temp_url 提交 MediaKit
             logger.warning(
-                "[lipsync_tts] TTS 音频下载失败，跳过静音检测，直接使用临时 URL 提交: job_id=%s err=%s",
+                "[lipsync_tts] TTS 音频转存 OSS 失败，回退临时 URL: job_id=%s err=%s",
                 job_id,
                 exc,
             )
-        # TTS 音频使用 CosyVoice 临时 URL，跳过自家 OSS 转存（加速，步骤⑥）
-        job.audio_url = temp_url
-        logger.info("[lipsync_tts] TTS 音频使用 CosyVoice 临时 URL（跳过 OSS 转存）: job_id=%s", job_id)
+            job.audio_url = temp_url
 
         db.commit()
 
-        # 2.5 计算精确句子时间戳（基于 TTS 音频静音检测）
-        # 直接复用步骤 2 已下载到内存的 audio_data，避免重新下载
-        import os as _os
-
-        try:
-            import subprocess as _sp
-            import tempfile as _tmpf
-
-            if not audio_data:
-                logger.warning("[lipsync_tts] 无音频数据，跳过句子时间戳计算: job_id=%s", job_id)
-            else:
-                # 写入临时文件供 ffprobe/ffmpeg 使用
-                with _tmpf.NamedTemporaryFile(suffix=".mp3", delete=False) as _atmp:
-                    _atmp.write(audio_data)
-                    _st_tmp_path = _atmp.name
-
-                # ffprobe 获取音频时长
-                _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
-                logger.info(
-                    "[lipsync_tts] 音频时长探测: job_id=%s duration=%.2f probe_stdout=%s probe_stderr=%s",
-                    job_id,
-                    _audio_duration,
-                    _probe_result.stdout.strip()[:50],
-                    _probe_result.stderr.strip()[:100] if _probe_result.stderr else "",
-                )
-
-                if _audio_duration > 0:
-                    _timings = _compute_sentence_timings(audio_data, 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,
-                        )
-                    else:
-                        logger.warning("[lipsync_tts] 句子时间戳计算返回空结果: job_id=%s", job_id)
-                else:
-                    logger.warning(
-                        "[lipsync_tts] ffprobe 未获取到有效时长，跳过句子时间戳: job_id=%s stdout=%s stderr=%s",
-                        job_id,
-                        _probe_result.stdout.strip()[:100],
-                        _probe_result.stderr.strip()[:200] if _probe_result.stderr else "",
-                    )
-                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, exc_info=True
-            )
-        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)
@@ -483,64 +221,3 @@ def tts_synthesize_and_submit(
             logger.exception("[lipsync_tts] 回写失败状态时异常: job_id=%s", job_id)
     finally:
         db.close()
-
-
-@shared_task(
-    name="lipsync_tts.persist_output_video",
-    max_retries=2,
-    default_retry_delay=30,
-)
-def persist_output_video_task(job_id: str, user_id: str, temp_url: str):
-    """异步转存对口型输出视频到自家 OSS（步骤⑦ — 将同步阻塞挪到后台，加速前端响应）.
-
-    - MediaKit 返回 completed 后先以 temp_url 回前端（前端可立即播放临时 URL）
-    - Celery 后台下载 temp_url 并转存 OSS，成功后更新 job.output_video_url 为永久 URL
-    - 失败则保留 temp_url，不阻断主流程
-    """
-
-    try:
-        from worker_app.db import SessionLocal  # type: ignore
-    except Exception:  # noqa: BLE001
-        from app.db import SessionLocal  # type: ignore
-
-    from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
-    from packages.shared.storage import get_shared_storage_service
-
-    db = SessionLocal()
-    try:
-        job = db.query(LipsyncJobModel).filter(LipsyncJobModel.id == job_id, LipsyncJobModel.user_id == user_id).first()
-        if job is None:
-            logger.error("[lipsync_tts.persist] Job not found: job_id=%s", job_id)
-            return
-
-        if not temp_url:
-            logger.warning("[lipsync_tts.persist] temp_url 为空，跳过转存: job_id=%s", job_id)
-            return
-
-        try:
-            import httpx
-
-            with httpx.Client(timeout=180.0, follow_redirects=True) as client:
-                resp = client.get(temp_url)
-                resp.raise_for_status()
-                data = resp.content
-
-            storage = get_shared_storage_service()
-            storage_key = f"lipsync-outputs/{user_id}/{job_id}.mp4"
-            permanent_url = storage.upload_file(io.BytesIO(data), storage_key, content_type="video/mp4")
-            # 对自家 OSS URL 重签 7 天有效期预签名，供前端播放
-            final_url = _sign_media_url(permanent_url) if permanent_url else temp_url
-            job.output_video_url = final_url
-            job.updated_at = datetime.now(timezone.utc)
-            db.commit()
-            logger.info("[lipsync_tts.persist] 输出视频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
-        except Exception as exc:
-            logger.warning(
-                "[lipsync_tts.persist] 输出视频转存失败，保留临时 URL: job_id=%s err=%s",
-                job_id,
-                exc,
-            )
-    except Exception:
-        logger.exception("[lipsync_tts.persist] 未预期异常: job_id=%s", job_id)
-    finally:
-        db.close()

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

@@ -23,10 +23,9 @@ import {
   getLipsyncJob,
   submitRender,
   getRenderJob,
-  generateRenderSmartCover,
+  generateSmartCover,
 } from "./api/aiAvatar"
 import { getOrCreateDefaultProject } from "@/api/projects"
-import type { RenderJob } from "./types"
 import {
   normalizeEmotion,
   buildTitleConfigPayload,
@@ -55,6 +54,8 @@ const AiAvatarPage: React.FC = () => {
     "generating",
   )
   const [lipsyncErrorMessage, setLipsyncErrorMessage] = useState("")
+  /* ── 智能封面加载态 ── */
+  const [smartCoverLoading, setSmartCoverLoading] = useState(false)
   /* ── 渲染进度弹窗 ── */
   const [showRenderModal, setShowRenderModal] = useState(false)
   const [renderStatus, setRenderStatus] = useState<"generating" | "completed" | "failed">(
@@ -62,8 +63,6 @@ const AiAvatarPage: React.FC = () => {
   )
   const [renderProgress, setRenderProgress] = useState(0)
   const [renderErrorMessage, setRenderErrorMessage] = useState("")
-  /* ── 当前渲染任务对象（轮询更新；用于封面区判断渲染是否完成） ── */
-  const [currentRenderJob, setCurrentRenderJob] = useState<RenderJob | null>(null)
 
   /* ── 对口型轮询 ── */
   const lipsyncTimerRef = useRef<ReturnType<typeof setInterval> | null>(null)
@@ -224,12 +223,7 @@ const AiAvatarPage: React.FC = () => {
           pip_scale: seg.pip_scale,
         })) as never,
         title_config: buildTitleConfigPayload(state.titleConfig),
-        // 封面不阻塞渲染：用户未选定封面时传空 dict，后端不生成封面；渲染完成后再单独抽帧
-        cover_config:
-          state.coverConfig.smart_cover_url ||
-          (state.coverConfig.upload_url && !state.coverConfig.upload_url.startsWith("blob:"))
-            ? buildCoverConfigPayload(state.coverConfig, state.coverConfig.smart_cover_url)
-            : {},
+        cover_config: buildCoverConfigPayload(state.coverConfig, state.coverConfig.smart_cover_url),
       })
 
       // 打开渲染进度弹窗，启动轮询
@@ -237,28 +231,26 @@ const AiAvatarPage: React.FC = () => {
       setRenderStatus("generating")
       setRenderProgress(job.progress ?? 0)
       setRenderErrorMessage("")
-      setCurrentRenderJob(job as RenderJob)
 
       if (renderTimerRef.current) clearInterval(renderTimerRef.current)
+      let renderPollCount = 0
+      const RENDER_MAX_POLLS = 200 // 最多轮询 10 分钟（200 次 × 3s）
       renderTimerRef.current = setInterval(async () => {
+        renderPollCount++
+        if (renderPollCount > RENDER_MAX_POLLS) {
+          if (renderTimerRef.current) clearInterval(renderTimerRef.current)
+          renderTimerRef.current = null
+          setRenderStatus("failed")
+          setRenderErrorMessage("渲染超时（超过10分钟），请稍后在任务历史查看结果")
+          return
+        }
         try {
           const updated = await getRenderJob(job.id)
           setRenderProgress(updated.progress ?? 0)
-          setCurrentRenderJob(updated)
           if (updated.status === "completed") {
             if (renderTimerRef.current) clearInterval(renderTimerRef.current)
             renderTimerRef.current = null
             setRenderStatus("completed")
-            // 渲染完成后：如果后端已返回封面（用户预上传/预设）则同步到前端；
-            // 否则不自动设置封面，由用户在封面区点击"智能获取封面"主动抽帧（步骤③④）
-            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)
@@ -291,48 +283,38 @@ const AiAvatarPage: React.FC = () => {
     setRenderErrorMessage("")
   }, [])
 
-  /* ── 智能封面：从最终渲染成片抽帧（POST /renders/{id}/smart-cover，步骤③④） ── */
-  const handleGenerateRenderSmartCover = useCallback(
-    async (renderId: string): Promise<{ cover_url: string; message?: string }> => {
-      try {
-        const res = await generateRenderSmartCover(renderId)
-        if (res.cover_url) {
-          state.setCoverConfig((prev) => ({
-            ...prev,
-            mode: "auto_frame",
-            smart_cover_url: res.cover_url,
-            thumbnail_url: res.cover_url,
-          }))
-          message.success("智能封面已生成")
-          return { cover_url: res.cover_url }
-        }
-        const errMsg = res.message || "智能封面生成失败，请稍后重试"
-        message.error(errMsg)
-        return { cover_url: "", message: errMsg }
-      } catch (err) {
-        console.error("智能封面生成失败:", err)
-        const errMsg = err instanceof Error ? err.message : "智能封面生成失败，请重试"
-        message.error(errMsg)
-        return { cover_url: "", message: errMsg }
+  /* ── 智能封面：调后端 MediaKit 选帧接口（#1822） ── */
+  const handleSmartCover = useCallback(async () => {
+    // 基于对口型成片抽帧，必须先完成对口型
+    const videoUrl = state.lipsyncJob?.output_video_url
+    if (state.lipsyncJob?.status !== "completed" || !videoUrl) {
+      message.warning("请先生成对口型视频，完成后再智能获取封面")
+      return
+    }
+    setSmartCoverLoading(true)
+    try {
+      const res = await generateSmartCover(videoUrl, buildTitleConfigPayload(state.titleConfig), 5)
+      if (res.cover_url) {
+        state.setCoverConfig((prev) => ({
+          ...prev,
+          mode: "auto_frame",
+          smart_cover_url: res.cover_url,
+          thumbnail_url: res.cover_url,
+        }))
+        message.success("智能封面已生成")
+      } else {
+        message.error(res.message || "智能封面生成失败，请稍后重试")
       }
-    },
-    // state.setCoverConfig 是 zustand action 引用稳定，eslint 不需要检查
+    } catch (err) {
+      console.error("智能封面生成失败:", err)
+      message.error(err instanceof Error ? err.message : "智能封面生成失败，请重试")
+    } finally {
+      setSmartCoverLoading(false)
+    }
     // eslint-disable-next-line react-hooks/exhaustive-deps
-    [],
-  )
+  }, [state.lipsyncJob])
 
   /* ── 配置汇总 ── */
-  const coverStatus: "not_ready" | "pending" | "selected" = (() => {
-    if (
-      state.coverConfig.smart_cover_url ||
-      state.coverConfig.thumbnail_url ||
-      (state.coverConfig.upload_url && !state.coverConfig.upload_url.startsWith("blob:"))
-    ) {
-      return "selected"
-    }
-    if (currentRenderJob?.status === "completed") return "pending"
-    return "not_ready"
-  })()
   const summary = {
     videoName: state.selectedVideo?.name || null,
     voiceName: state.selectedVoice?.name || null,
@@ -340,7 +322,7 @@ const AiAvatarPage: React.FC = () => {
     lipsyncStatus: state.lipsyncJob?.status || null,
     brollCount: state.bRollSegments.length,
     hasTitle: state.titleConfig.title.length > 0,
-    coverStatus,
+    hasCover: state.coverConfig.enabled,
   }
 
   return (
@@ -476,8 +458,9 @@ const AiAvatarPage: React.FC = () => {
                     state.setCoverConfig((prev) => ({ ...prev, ...partial }))
                   }
                   titleConfig={state.titleConfig}
-                  renderJob={currentRenderJob}
-                  onGenerateRenderSmartCover={handleGenerateRenderSmartCover}
+                  onSmartCover={handleSmartCover}
+                  smartCoverLoading={smartCoverLoading}
+                  canSmartCover={state.lipsyncJob?.status === "completed"}
                   resolution={state.resolution}
                   onResolutionChange={state.setResolution}
                   isGenerating={state.isGenerating}
@@ -515,9 +498,8 @@ 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

@@ -94,16 +94,3 @@ export const getRenderJob = async (jobId: string): Promise<RenderJob> => {
 export const cancelRenderJob = async (jobId: string): Promise<void> => {
   await apiClient.post(`/ai-avatar/render/${jobId}/cancel`)
 }
-
-/* ── 从最终渲染成片智能抽封面（POST /ai-avatar/renders/{job_id}/smart-cover） ── */
-export const generateRenderSmartCover = async (
-  jobId: string,
-): Promise<{ cover_url: string; status: string; message: string }> => {
-  const response = await apiClient.post<{ cover_url: string; status: string; message: string }>(
-    `/ai-avatar/render/${jobId}/smart-cover`,
-    {},
-    // 抽帧+评分+转存 OSS 链路较长，120s 超时
-    { timeout: 120000 },
-  )
-  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（优先使用后端精确时间戳，降级为字数比例估算） */
+  /** 文案分句（⑤） */
   const sentences = useMemo(
-    () => splitScriptIntoSentences(scriptText, sentenceTimings, _outputDuration),
-    [scriptText, sentenceTimings, _outputDuration],
+    () => 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） */
@@ -145,7 +142,7 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
     setSelectedAsset(asset)
   }
 
-  /** 确认添加一段 B-roll（⑥ 时间取所选句子的精确起止，后端静音检测 / 前端字数比例降级） */
+  /** 确认添加一段 B-roll（⑥ 时间取所选句子的估算起止） */
   const handleConfirm = () => {
     if (!selectedAsset || !selectedSentence) return
     const startTime = selectedSentence.startTime
@@ -267,9 +264,11 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
                           >
                             <span className="aa-sentence-item__idx">{sent.index + 1}</span>
                             <span className="aa-sentence-item__text">{sent.text}</span>
-                            <span className="aa-sentence-item__time">
-                              {sent.startTime.toFixed(1)}-{sent.endTime.toFixed(1)}s
-                            </span>
+                            {outputDuration > 0 && (
+                              <span className="aa-sentence-item__time">
+                                {sent.startTime.toFixed(1)}-{sent.endTime.toFixed(1)}s
+                              </span>
+                            )}
                           </button>
                         )
                       })}
@@ -350,7 +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

@@ -1,15 +1,14 @@
 /**
- * AI数字人 — 面板5：分辨率/配置摘要/生成按钮/封面
- * v3 调整（步骤③④）：
- * - 布局顺序：分辨率 → 配置摘要卡片 → 🔘「开始生成视频」按钮 → （渲染完成后）封面区域
- * - 渲染未完成时封面区域显示占位态，按钮 disabled
- * - 「智能获取封面」从最终成片抽帧（调用 POST /renders/{id}/smart-cover），不再依赖 lipsync 状态
- * - 修复点 2 次 bug：内部维护 smartCoverLoading，不依赖外层异步 state 更新
+ * 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
+ * - 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, useState } from "react"
-import type { AiAvatarCoverConfig, AiAvatarTitleConfig, RenderJob } from "../types"
+import React, { useMemo, useRef } from "react"
+import type { AiAvatarCoverConfig, AiAvatarTitleConfig } from "../types"
 
 interface PanelCoverAndGenerateProps {
   coverConfig: AiAvatarCoverConfig
@@ -19,12 +18,10 @@ interface PanelCoverAndGenerateProps {
   onResolutionChange: (r: string) => void
   isGenerating: boolean
   onGenerate: () => void
-  /** 当前渲染任务（渲染完成后才有 output_video_url，才能抽封面） */
-  renderJob: RenderJob | null
-  /** 从最终成片智能抽帧（参数 renderId），返回 { cover_url } */
-  onGenerateRenderSmartCover: (renderId: string) => Promise<{ cover_url: string; message?: string }>
-  /** 自定义上传封面（选择本地文件后由父组件处理实际上传） */
-  onUploadCover?: (file: File) => void
+  /** 智能获取封面（MediaKit 选帧） */
+  onSmartCover: () => void
+  smartCoverLoading: boolean
+  canSmartCover: boolean
   /** 配置汇总信息 */
   summary: {
     videoName: string | null
@@ -33,8 +30,7 @@ interface PanelCoverAndGenerateProps {
     lipsyncStatus: string | null
     brollCount: number
     hasTitle: boolean
-    /** 封面状态：'not_ready'(视频未生成) / 'pending'(视频生成了但未选) / 'selected'(已选) */
-    coverStatus: "not_ready" | "pending" | "selected"
+    hasCover: boolean
   }
 }
 
@@ -70,14 +66,12 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
   onResolutionChange,
   isGenerating,
   onGenerate,
-  renderJob,
-  onGenerateRenderSmartCover,
-  onUploadCover,
+  onSmartCover,
+  smartCoverLoading,
+  canSmartCover,
   summary,
 }) => {
   const uploadInputRef = useRef<HTMLInputElement>(null)
-  // 内部维护智能封面加载态（修复点 2 次 bug：不依赖外层异步 setState 顺序）
-  const [smartCoverLoading, setSmartCoverLoading] = useState(false)
 
   /** 自定义上传封面 */
   const handleUploadClick = () => {
@@ -87,44 +81,22 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
   const handleFileChange = (e: React.ChangeEvent<HTMLInputElement>) => {
     const file = e.target.files?.[0]
     if (!file) return
-    if (onUploadCover) {
-      onUploadCover(file)
-    } else {
-      // 本地预览兜底（实际上传由父级处理；blob URL 仅作本地展示）
-      const url = URL.createObjectURL(file)
-      onCoverConfigChange({ mode: "upload", upload_url: url, thumbnail_url: url })
-    }
+    // 本地预览：生成 object URL（实际上传由父级/后端链路处理）
+    const url = URL.createObjectURL(file)
+    onCoverConfigChange({ mode: "upload", upload_url: url, thumbnail_url: url })
+    // 允许重复选择同一文件
     e.target.value = ""
   }
 
-  /** 智能获取封面（从最终成片抽帧；必须等 render 完成） */
-  const handleSmartCover = async () => {
-    if (!renderJob || renderJob.status !== "completed" || !renderJob.id) return
-    setSmartCoverLoading(true)
-    try {
-      const res = await onGenerateRenderSmartCover(renderJob.id)
-      if (res.cover_url) {
-        onCoverConfigChange({
-          mode: "auto_frame",
-          smart_cover_url: res.cover_url,
-          thumbnail_url: res.cover_url,
-        })
-      } else {
-        // 失败由父组件 message 提示，这里不重复弹窗
-        console.warn("[智能封面] 返回空 cover_url:", res.message)
-      }
-    } catch (err) {
-      console.error("[智能封面] 调用失败:", err)
-    } finally {
-      setSmartCoverLoading(false)
-    }
+  /** 智能获取封面（调后端 MediaKit 抽帧评分选最佳帧，#1822） */
+  const handleSmartCover = () => {
+    onCoverConfigChange({ mode: "auto_frame" })
+    onSmartCover()
   }
 
   const lipsync = summary.lipsyncStatus ? LIPSYNC_STATUS_LABEL[summary.lipsyncStatus] : null
+
   const canGenerate = summary.lipsyncStatus === "completed" && !isGenerating
-  // 渲染已完成 → 封面区可用
-  const isRenderCompleted = renderJob?.status === "completed"
-  const canSmartCover = isRenderCompleted && !smartCoverLoading
 
   /** 封面图实际展示的 url：智能封面 > 自定义上传 > 空 */
   const coverUrl =
@@ -148,7 +120,7 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
       wordBreak: "break-word",
       whiteSpace: "pre-wrap",
       color: titleConfig.color || "#ffffff",
-      fontSize: `${(titleConfig.size || 48) * 0.35}px`,
+      fontSize: `${titleConfig.size}px`,
       fontFamily: getFontFamily(titleConfig.font),
       fontWeight: titleConfig.bold ? "bold" : "normal",
       fontStyle: titleConfig.italic ? "italic" : "normal",
@@ -156,6 +128,7 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
       pointerEvents: "none",
     }
 
+    // 位置
     const pos = titleConfig.position || "bottom"
     if (pos === "top") {
       style.top = "40px"
@@ -163,6 +136,7 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
       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%)"
@@ -170,35 +144,67 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
       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 {
-      style.textShadow = "none"
+      // 默认给轻微阴影保证白字在亮背景可读
+      style.textShadow = "0 2px 6px rgba(0,0,0,0.6)"
     }
 
     return style
   }, [titleConfig])
 
-  /** 封面区占位文字 */
-  const coverPlaceholder = isRenderCompleted ? "暂无封面" : "视频生成后可选择封面"
-
-  /** 封面摘要状态文本 */
-  const coverSummaryNode = (() => {
-    if (summary.coverStatus === "selected") {
-      return <span className="aa-config-summary__value">已选择</span>
-    }
-    if (summary.coverStatus === "pending") {
-      return <span className="aa-config-summary__value">待选择</span>
-    }
-    return <span className="aa-config-summary__empty">生成视频后可选</span>
-  })()
-
   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} />
+        ) : (
+          <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">
+        <button
+          type="button"
+          className={`aa-btn aa-btn--ghost${coverConfig.mode === "auto_frame" ? " active" : ""}`}
+          onClick={handleSmartCover}
+          disabled={smartCoverLoading || !canSmartCover}
+          title={canSmartCover ? "基于对口型成片智能选帧" : "请先完成对口型生成"}
+        >
+          {smartCoverLoading ? "⏳ 智能选帧中…" : "🎬 智能获取封面"}
+        </button>
+        <button
+          type="button"
+          className={`aa-btn aa-btn--ghost${coverConfig.mode === "upload" ? " active" : ""}`}
+          onClick={handleUploadClick}
+        >
+          📷 自定义上传
+        </button>
+        <input
+          ref={uploadInputRef}
+          type="file"
+          accept="image/*"
+          style={{ display: "none" }}
+          onChange={handleFileChange}
+        />
+      </div>
+
       {/* 分辨率选择 */}
       <div className="aa-form-field">
         <label className="aa-label">分辨率</label>
@@ -206,7 +212,6 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           className="aa-select"
           value={resolution}
           onChange={(e) => onResolutionChange(e.target.value)}
-          disabled={isGenerating}
         >
           {RESOLUTION_OPTIONS.map((opt) => (
             <option key={opt.value} value={opt.value}>
@@ -267,7 +272,11 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           </div>
           <div className="aa-config-summary__row">
             <span>封面</span>
-            {coverSummaryNode}
+            {summary.hasCover ? (
+              <span className="aa-config-summary__value">已开启</span>
+            ) : (
+              <span className="aa-config-summary__empty">未配置</span>
+            )}
           </div>
         </div>
 
@@ -285,60 +294,6 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
             请先完成对口型生成
           </div>
         )}
-        {isGenerating && (
-          <div style={{ marginTop: 8, fontSize: 11, color: "#8c8ca1", textAlign: "center" }}>
-            视频生成中，请稍候…
-          </div>
-        )}
-      </div>
-
-      {/* 封面区域（视频生成后才激活；步骤③④要求：按钮在封面上方，完成后再显示封面区） */}
-      <div className="aa-cover-section" style={{ marginTop: 16 }}>
-        <div className="aa-label" style={{ marginBottom: 8 }}>
-          封面
-        </div>
-        {/* 封面预览（竖屏 9:16） */}
-        <div className="aa-cover-preview" style={{ opacity: isRenderCompleted ? 1 : 0.5 }}>
-          {hasCoverImage ? (
-            <img src={coverUrl!} alt="封面预览" draggable={false} />
-          ) : (
-            <span className="aa-cover-preview__placeholder">{coverPlaceholder}</span>
-          )}
-          {smartCoverLoading && <div className="aa-cover-preview__loading">⏳ 智能选帧中…</div>}
-          {showTitleOverlay && (
-            <div style={titleOverlayStyle} aria-hidden="true">
-              {titleConfig.title}
-            </div>
-          )}
-        </div>
-
-        <div className="aa-cover-actions">
-          <button
-            type="button"
-            className={`aa-btn aa-btn--ghost${coverConfig.mode === "auto_frame" ? " active" : ""}`}
-            onClick={handleSmartCover}
-            disabled={!canSmartCover}
-            title={isRenderCompleted ? "从成片智能选帧" : "请先生成视频"}
-          >
-            {smartCoverLoading ? "⏳ 智能选帧中…" : "🎬 智能获取封面"}
-          </button>
-          <button
-            type="button"
-            className={`aa-btn aa-btn--ghost${coverConfig.mode === "upload" ? " active" : ""}`}
-            onClick={handleUploadClick}
-            disabled={!isRenderCompleted || smartCoverLoading}
-            title={isRenderCompleted ? "自定义上传封面" : "请先生成视频"}
-          >
-            📷 自定义上传
-          </button>
-          <input
-            ref={uploadInputRef}
-            type="file"
-            accept="image/*"
-            style={{ display: "none" }}
-            onChange={handleFileChange}
-          />
-        </div>
       </div>
     </div>
   )

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: `${(titleConfig.size || 48) * 0.35}px`,
+        fontSize: `${(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/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
 }
@@ -122,7 +110,7 @@ export const DEFAULT_TITLE_CONFIG: AiAvatarTitleConfig = {
   title: "",
   position: "bottom",
   font: "思源黑体",
-  size: 48,
+  size: 28,
   bold: true,
   italic: false,
   stroke: false,

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) || 48,
+    font_size: Math.round(cfg.size) || 36,
     font_color: cfg.color || "#ffffff",
     position,
     bold: !!cfg.bold,
@@ -67,14 +67,9 @@ export function buildCoverConfigPayload(
     // build_cover_extract_command 读取 timestamp（截帧秒数）
     timestamp: cfg.frame_time || 0,
   }
-  // 智能封面 URL（后端字段名为 url/imageUrl/cover_url 都兼容，优先 url）
-  if (smartCoverUrl) {
-    payload.url = smartCoverUrl
-    payload.cover_url = smartCoverUrl
-  }
+  if (smartCoverUrl) payload.cover_url = smartCoverUrl
   // 自定义上传：blob: 本地预览地址无法给后端，仅 OSS URL 可用
   if (cfg.mode === "upload" && cfg.upload_url && !cfg.upload_url.startsWith("blob:")) {
-    payload.url = cfg.upload_url
     payload.upload_url = cfg.upload_url
   }
   return payload

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

@@ -1,8 +1,5 @@
 /**
- * AI数字人 — 文案分句 & B-roll 时间计算
- *
- * 优先使用后端基于 TTS 音频静音检测计算的精确 sentence_timings；
- * 后端未返回（如对口型还在生成中）时，降级为前端按字数比例估算。
+ * AI数字人 — 文案分句 & B-roll 时间自动估算（#1809 ⑤⑥）
  */
 
 export interface ScriptSentence {
@@ -14,63 +11,28 @@ export interface ScriptSentence {
   charCount: number
   /** 累计起始字数（用于时间估算） */
   startChar: number
-  /** 对口型视频内起始时间（秒）——后端精确值或前端估算 */
+  /** 估算的对口型视频内起始时间（秒） */
   startTime: number
-  /** 对口型视频内结束时间（秒）——后端精确值或前端估算 */
+  /** 估算的对口型视频内结束时间（秒） */
   endTime: number
 }
 
 /**
  * 按句号/问号/感叹号/分号/换行分句（兼容中英文标点）。
- * 空文案返回空数组。时间优先使用后端 sentence_timings；否则按字数线性估算。
- *
- * @param sentenceTimings 后端返回的精确句子时间戳（来自 lipsync_job.sentence_timings）。
- *   非空且有效时优先采用，跳过前端估算。
+ * 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?: { index: number; text: string; start_time: number; end_time: number }[] | null,
-  outputDuration: number = 0,
+  outputDuration: number,
 ): ScriptSentence[] {
   const text = (scriptText || "").trim()
   if (!text) return []
 
-  // 1. 先做基础分句（仅用于降级估算 / 没有 sentenceTimings 时）
   const rawParts = text
     .split(/[。！？!?；;\n\r]+/)
     .map((part) => part.trim())
     .filter((part) => part.length > 0)
 
-  // 2. 优先使用后端精确时间戳
-  //    校验：必须是数组、条数一致、每条都有 start_time/end_time，否则降级估算
-  if (Array.isArray(sentenceTimings) && sentenceTimings.length === rawParts.length) {
-    const valid = sentenceTimings.every(
-      (t) =>
-        t &&
-        typeof t.start_time === "number" &&
-        typeof t.end_time === "number" &&
-        t.end_time >= t.start_time,
-    )
-    if (valid) {
-      let accChar = 0
-      return sentenceTimings.map((t, i) => {
-        const part = rawParts[i] ?? t.text ?? ""
-        const charCount = part.replace(/\s/g, "").length
-        const sentence: ScriptSentence = {
-          index: t.index ?? i,
-          text: part,
-          charCount,
-          startChar: accChar,
-          startTime: round1(t.start_time),
-          endTime: round1(t.end_time),
-        }
-        accChar += charCount
-        return sentence
-      })
-    }
-  }
-
-  // 3. 降级：按字数比例线性估算
   const totalChars = rawParts.reduce((sum, part) => sum + part.replace(/\s/g, "").length, 0)
   const duration = outputDuration > 0 ? outputDuration : 0
 

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/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",
 }
 
 
@@ -420,44 +416,21 @@ def _escape_drawtext_text(text: str) -> str:
     return result
 
 
-# 粗体字体文件映射：服务器镜像只保留了 NotoSansSC-VF.ttf（可变字体，已删除
-# NotoSansCJK-Bold.ttc 以避免 Mono 变体问题，见 worker-base.Dockerfile），
-# 因此无法通过 fontfile 切换到 Bold 字重。这里保留路径列表作为未来扩展，
-# 实际加粗通过 borderw 黑色描边实现（见下）。
-DRAWTEXT_BOLD_FONT_SEARCH_PATHS: list[str] = [
-    "/usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc",
-    "/usr/share/fonts/noto-cjk/NotoSansCJK-Bold.ttc",
-    "/usr/share/fonts/google-noto-cjk/NotoSansCJK-Bold.ttc",
-    "/usr/share/fonts/truetype/noto/NotoSansSC-Bold.ttf",
-    "/usr/share/fonts/noto/NotoSansSC-Bold.ttf",
-]
-
-
-def _resolve_font_path(font_name: str, bold: bool = False) -> str:
+def _resolve_font_path(font_name: str) -> str:
     """解析字体名到服务器实际字体文件路径。
 
     查找策略：
     1. 通过 DRAWTEXT_FONT_MAP 映射前端字体名到服务器关键字
-    2. bold=True 时优先查找粗体变体；找不到回退常规字重
-    3. 在 DRAWTEXT_FONT_SEARCH_PATHS 中查找匹配路径
-    4. 未找到则返回空字符串（drawtext 使用内置默认字体）
+    2. 在 DRAWTEXT_FONT_SEARCH_PATHS 中查找匹配路径
+    3. 未找到则返回空字符串（drawtext 使用内置默认字体）
     """
     keyword = DRAWTEXT_FONT_MAP.get(font_name, font_name)
     import os
 
-    if bold:
-        for path in DRAWTEXT_BOLD_FONT_SEARCH_PATHS:
-            if keyword.lower() in path.lower() and os.path.isfile(path):
-                return path
-        # 粗体文件找不到时，再查常规字重（后面会用描边兜底加粗）
     for path in DRAWTEXT_FONT_SEARCH_PATHS:
         if keyword.lower() in path.lower() and os.path.isfile(path):
             return path
     # fallback：遍历搜索任意可用字体
-    if bold:
-        for path in DRAWTEXT_BOLD_FONT_SEARCH_PATHS:
-            if os.path.isfile(path):
-                return path
     for path in DRAWTEXT_FONT_SEARCH_PATHS:
         if os.path.isfile(path):
             return path
@@ -504,13 +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 48)
+    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")
@@ -518,8 +491,8 @@ def build_title_drawtext_filter(
     # ── 构建 drawtext 参数 ──
     params: list[str] = []
 
-    # 字体文件：粗体优先使用 Bold 字体文件，避免同色描边造成字形偏移/重影
-    font_path = _resolve_font_path(font_name, bold=bold)
+    # 字体文件
+    font_path = _resolve_font_path(font_name)
     if font_path:
         escaped_path = font_path.replace("\\", "\\\\").replace(":", "\\\\:").replace("'", "\\\\'")
         params.append(f"fontfile='{escaped_path}'")
@@ -531,28 +504,26 @@ def build_title_drawtext_filter(
     params.append(f"fontsize={font_size}")
     params.append(f"fontcolor={font_color}")
 
+    # 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）
-    # 之前用 borderw=3 + font_color 同色描边模拟粗体，会在小字号/竖屏视频上造成
-    # 字形偏移、边缘重影，看起来像文字被打印了两次（用户截图中的标题"曝光曝光…"）。
-    # 修复：粗体改用黑色细描边（borderw=2, 黑色），视觉上清晰加粗且不产生偏移。
-    # 用户显式开启 stroke 时按用户配置走；粗体+无stroke 默认黑色细描边。
-    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：黑色细描边，模拟粗体同时保证不重影
-        border_width = 2
-        border_color = "000000"
-    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:
@@ -577,13 +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
@@ -607,7 +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:
@@ -615,182 +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(

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
@@ -318,120 +315,3 @@ class TestBrollOverlayFilter:
             "/tmp/cover.jpg",
         )
         assert "scale=" in cmd
-
-
-def _make_mock_auth_user(user_id="user-1"):
-    """构造 AuthenticatedUser：current_user.user.id."""
-    auth = MagicMock()
-    auth.user.id = user_id
-    return auth
-
-
-class TestRenderSmartCoverRoute:
-    """POST /renders/{job_id}/smart-cover — 从成片智能抽封面（步骤②）."""
-
-    def test_smart_cover_job_not_found_returns_404(self):
-        """渲染任务不存在 → 404."""
-        from app.api.routes.ai_avatar_render import generate_render_smart_cover
-        from fastapi import HTTPException
-
-        mock_service = MagicMock()
-        mock_service.get_render_job.return_value = None
-        mock_db = MagicMock()
-        mock_user = _make_mock_auth_user()
-
-        # 函数内部 `from app.services.ai_avatar_render_service import AiAvatarRenderService`
-        with patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service):
-            with pytest.raises(HTTPException) as exc_info:
-                generate_render_smart_cover(job_id="render-missing", current_user=mock_user, db=mock_db)
-
-        assert exc_info.value.status_code == 404
-        assert "不存在" in exc_info.value.detail
-        mock_service.get_render_job.assert_called_once_with("render-missing", "user-1")
-
-    def test_smart_cover_job_not_completed_returns_400(self):
-        """任务未 completed（如 processing）→ 400."""
-        from app.api.routes.ai_avatar_render import generate_render_smart_cover
-        from fastapi import HTTPException
-
-        mock_service = MagicMock()
-        mock_job = _make_mock_render_job(status="processing", output_video_url="https://oss/video.mp4")
-        mock_service.get_render_job.return_value = mock_job
-        mock_db = MagicMock()
-        mock_user = _make_mock_auth_user()
-
-        with patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service):
-            with pytest.raises(HTTPException) as exc_info:
-                generate_render_smart_cover(job_id="render-1", current_user=mock_user, db=mock_db)
-
-        assert exc_info.value.status_code == 400
-        assert "先完成视频生成" in exc_info.value.detail
-
-    def test_smart_cover_empty_video_url_returns_400(self):
-        """已 completed 但 output_video_url 为空/空白 → 400."""
-        from app.api.routes.ai_avatar_render import generate_render_smart_cover
-        from fastapi import HTTPException
-
-        mock_service = MagicMock()
-        mock_job = _make_mock_render_job(status="completed", output_video_url="   ")
-        mock_service.get_render_job.return_value = mock_job
-        mock_db = MagicMock()
-        mock_user = _make_mock_auth_user()
-
-        with patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service):
-            with pytest.raises(HTTPException) as exc_info:
-                generate_render_smart_cover(job_id="render-1", current_user=mock_user, db=mock_db)
-
-        assert exc_info.value.status_code == 400
-        assert "URL 为空" in exc_info.value.detail
-
-    def test_smart_cover_success_updates_db_and_returns_url(self):
-        """抽帧成功 → 更新 job.cover_config / output_cover_url 并 commit，返回 completed."""
-        from app.api.routes.ai_avatar_render import generate_render_smart_cover
-
-        mock_service = MagicMock()
-        mock_job = _make_mock_render_job(
-            status="completed",
-            output_video_url="https://oss/final.mp4",
-        )
-        mock_job.cover_config = {"mode": "manual"}
-        mock_service.get_render_job.return_value = mock_job
-        mock_db = MagicMock()
-        mock_user = _make_mock_auth_user()
-
-        with (
-            patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service),
-            patch(
-                "app.api.routes.ai_avatar_render.generate_smart_cover", return_value="https://oss/cover.jpg"
-            ) as mock_gen,
-        ):
-            result = generate_render_smart_cover(job_id="render-1", current_user=mock_user, db=mock_db)
-
-        mock_gen.assert_called_once_with("https://oss/final.mp4", job_id="render-1", max_frames=5)
-        assert result.status == "completed"
-        assert result.cover_url == "https://oss/cover.jpg"
-        assert mock_job.output_cover_url == "https://oss/cover.jpg"
-        assert mock_job.cover_config["mode"] == "auto_frame"
-        assert mock_job.cover_config["url"] == "https://oss/cover.jpg"
-        mock_db.commit.assert_called_once()
-
-    def test_smart_cover_extract_failure_returns_fallback_failed(self):
-        """generate_smart_cover 抛异常 → fallback_failed，不抛错不写 DB."""
-        from app.api.routes.ai_avatar_render import generate_render_smart_cover
-
-        mock_service = MagicMock()
-        mock_job = _make_mock_render_job(status="completed", output_video_url="https://oss/final.mp4")
-        mock_service.get_render_job.return_value = mock_job
-        mock_db = MagicMock()
-        mock_user = _make_mock_auth_user()
-
-        with (
-            patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service),
-            patch("app.api.routes.ai_avatar_render.generate_smart_cover", side_effect=RuntimeError("mediakit down")),
-        ):
-            result = generate_render_smart_cover(job_id="render-1", current_user=mock_user, db=mock_db)
-
-        assert result.status == "fallback_failed"
-        assert result.cover_url == ""
-        # 失败时不写 cover_config / 不 commit
-        mock_db.commit.assert_not_called()

tests/unit/test_ai_avatar_render_service.py

@@ -628,58 +628,3 @@ class TestAiAvatarRenderService:
         err = AiAvatarRenderError("测试错误", code="TestCode")
         assert err.code == "TestCode"
         assert str(err) == "测试错误"
-
-
-class TestAiAvatarRenderCoverPassthrough:
-    """execute_render 中封面透传逻辑（320~329 行）：cover_config 含 url/imageUrl/cover_url 时直接透传到 output_cover_url."""
-
-    def _run_execute(self, mock_job, mock_lipsync_job):
-        """驱动 execute_render 跑到完成阶段的通用脚手架（mock IO 部分）."""
-        from app.services.ai_avatar_render_service import AiAvatarRenderService
-
-        mock_db = _make_mock_db()
-        mock_filter = MagicMock()
-        # query.filter 返回同一个 filter 两次（render_job 查询、lipsync 查询）
-        mock_filter.first.side_effect = [mock_job, mock_lipsync_job]
-        mock_query = MagicMock()
-        mock_query.filter.return_value = mock_filter
-        mock_db.query.return_value = mock_query
-
-        svc = AiAvatarRenderService(mock_db)
-        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("tempfile.TemporaryDirectory") as tmpdir_mock,
-            patch("app.services.ai_avatar_cover_service.generate_smart_cover", return_value=""),
-            patch("packages.domain.generated_video.GeneratedVideo.create", return_value=MagicMock()),
-            patch(
-                "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")
-            tmpdir_mock.return_value.__exit__ = MagicMock(return_value=False)
-            repo_cls.return_value = MagicMock()
-            svc.execute_render(mock_job.id)
-        return mock_db, mock_job
-
-    def test_cover_url_in_cover_config_passthrough_to_output_cover(self):
-        """cover_config.url 存在 → 透传到 output_cover_url."""
-        mock_job = _make_mock_render_job(job_id="render-cov-1", status="pending")
-        mock_job.cover_config = {"mode": "upload", "url": "https://oss/user-cover.jpg"}
-        mock_lipsync_job = _make_mock_lipsync_job(status="completed", output_duration=10.0)
-        _, job = self._run_execute(mock_job, mock_lipsync_job)
-        assert job.output_cover_url == "https://oss/user-cover.jpg"
-
-    def test_cover_imageurl_fallback_also_passthrough(self):
-        """cover_config.imageUrl（老字段）存在 → 也透传到 output_cover_url."""
-        mock_job = _make_mock_render_job(job_id="render-cov-2", status="pending")
-        mock_job.cover_config = {"mode": "upload", "imageUrl": "https://oss/user-cover2.jpg"}
-        mock_lipsync_job = _make_mock_lipsync_job(status="completed", output_duration=10.0)
-        _, job = self._run_execute(mock_job, mock_lipsync_job)
-        assert job.output_cover_url == "https://oss/user-cover2.jpg"

tests/unit/test_lipsync_speed_optimization.py

@@ -288,47 +288,3 @@ class TestCancelJobTtsProcessing:
 
         result = svc.cancel_job("job-1", "user-1")
         assert result.status == "cancelled"
-
-
-class TestCreateJobCommitOrder:
-    """验证事务顺序修复：create_job 必须先 commit 再发 Celery 任务，避免 worker 消费时 job 不可见。"""
-
-    def test_commit_called_before_apply_async_in_tts_mode(self):
-        """TTS 模式：db.commit() 必须在 apply_async() 之前调用，防止 worker 查不到 job 永远卡在 tts_processing。"""
-        svc, client, cosy = _make_service_with_mocks()
-        call_order: list[str] = []
-
-        def track_commit():
-            call_order.append("commit")
-
-        def track_apply_async(*args, **kwargs):
-            call_order.append("apply_async")
-
-        svc.db.commit.side_effect = track_commit
-
-        with patch("app.services.lipsync_service.tts_synthesize_and_submit") as mock_task:
-            mock_task.apply_async = MagicMock(side_effect=track_apply_async)
-            svc.create_job(
-                user_id="user-1",
-                video_url="https://example.com/video.mp4",
-                voice_id="v-1",
-                script_text="测试",
-            )
-
-        # 至少有一次 commit 在 apply_async 之前
-        assert "commit" in call_order, "db.commit 必须被调用"
-        assert "apply_async" in call_order, "apply_async 必须被调用"
-        assert call_order.index("commit") < call_order.index(
-            "apply_async"
-        ), f"事务顺序错误：commit 必须在 apply_async 之前，实际顺序 {call_order}"
-
-    def test_job_not_found_retry_mechanism_exists(self):
-        """worker 侧 job not found 必须有重试机制（self.retry），而不是静默 return。"""
-        import inspect
-
-        from app.tasks.lipsync_tts import tts_synthesize_and_submit
-
-        source = inspect.getsource(tts_synthesize_and_submit.run)
-        assert (
-            "self.retry" in source or "retry" in source
-        ), "tts_synthesize_and_submit 在 job not found 时必须重试，防止静默失败"

tests/unit/test_lipsync_tts.py

@@ -185,10 +185,7 @@ class TestTtsSynthesizeAndSubmit:
         mk_client.submit_lipsync.assert_called_once()
         call_kwargs = mk_client.submit_lipsync.call_args.kwargs
         assert call_kwargs["client_token"] == "job-1"
-        # CosyVoice 临时 URL 经 _sign_media_url 透传（mock 统一追加 ?signed），
-        # 自家 OSS 才会被重签，外部 URL 原样透传；job.audio_url 存原始临时 URL
-        assert call_kwargs["audio_url"] == "https://tts/raw.mp3?signed"
-        assert job.audio_url == "https://tts/raw.mp3"
+        assert call_kwargs["audio_url"].endswith("?signed")
         session.commit.assert_called()
         session.close.assert_called_once()
 
@@ -390,242 +387,3 @@ class TestSignMediaUrl:
 
         assert result == "https://anything.example.com/a.mp3"
         fake_storage.get_download_url.assert_not_called()
-
-
-class TestPersistOutputVideoTask:
-    """persist_output_video_task：下载 MediaKit 临时视频 → 上传自有 OSS → 更新 DB."""
-
-    def _make_persist_job(self, **kwargs):
-        job = MagicMock()
-        job.id = kwargs.get("job_id", "job-1")
-        job.user_id = kwargs.get("user_id", "user-1")
-        job.output_video_url = kwargs.get("output_video_url", "https://temp.mk/output.mp4")
-        job.updated_at = None
-        return job
-
-    def _persist_patches(self, *, job, video_bytes=b"FAKEMP4", download_side_effect=None, upload_url=None):
-        """统一 patch：SessionLocal、httpx.Client、storage、_sign_media_url."""
-        fake_app_db = ModuleType("app.db")
-        fake_worker_db = ModuleType("worker_app.db")
-        session, factory = _build_session(job)
-        fake_app_db.SessionLocal = factory
-        fake_worker_db.SessionLocal = factory
-
-        # httpx.Client 上下文管理器
-        fake_response = MagicMock()
-        fake_response.content = video_bytes
-        fake_response.raise_for_status = MagicMock()
-        fake_client = MagicMock()
-        fake_client.get.return_value = fake_response
-        fake_client_cm = MagicMock()
-        fake_client_cm.__enter__ = MagicMock(return_value=fake_client)
-        fake_client_cm.__exit__ = MagicMock(return_value=False)
-        FakeHttpxClient = MagicMock(return_value=fake_client_cm)
-        if download_side_effect is not None:
-            fake_client.get.side_effect = download_side_effect
-
-        # storage
-        storage = MagicMock()
-        storage.public_url = "https://oss.example.com/"
-        storage.upload_file.return_value = upload_url or "https://oss.example.com/lipsync-outputs/user-1/job-1.mp4"
-
-        fake_httpx = ModuleType("httpx")
-        fake_httpx.Client = FakeHttpxClient
-
-        patches = [
-            patch.dict(
-                sys.modules,
-                {"app.db": fake_app_db, "worker_app.db": fake_worker_db, "httpx": fake_httpx},
-            ),
-            patch("packages.shared.storage.get_shared_storage_service", return_value=storage),
-            patch("app.tasks.lipsync_tts._sign_media_url", side_effect=lambda url: url + "?signed" if url else url),
-        ]
-        return session, fake_client, storage, patches
-
-    def test_success_download_upload_updates_db(self):
-        """正常路径：下载 temp_url → 上传 OSS → 签名 → 写回 DB commit."""
-        from app.tasks.lipsync_tts import persist_output_video_task
-
-        job = self._make_persist_job(output_video_url="https://temp.mk/x.mp4")
-        session, fake_client, storage, patches = self._persist_patches(
-            job=job, video_bytes=b"VIDEODATA", upload_url="https://oss.example.com/lipsync-outputs/u1/j1.mp4"
-        )
-        entered = [p.__enter__() for p in patches]
-        try:
-            persist_output_video_task("job-1", "user-1", "https://temp.mk/x.mp4")
-        finally:
-            for p in reversed(patches):
-                p.__exit__(None, None, None)
-
-        fake_client.get.assert_called_once_with("https://temp.mk/x.mp4")
-        storage.upload_file.assert_called_once()
-        # 上传的 key 必须是 lipsync-outputs/{user_id}/{job_id}.mp4
-        key_arg = (
-            storage.upload_file.call_args.args[1]
-            if storage.upload_file.call_args.args
-            else storage.upload_file.call_args.kwargs.get("key")
-        )
-        # upload_file(data, key, content_type=...)
-        call_args = storage.upload_file.call_args.args
-        assert call_args[1] == "lipsync-outputs/user-1/job-1.mp4"
-        # output_video_url 被替换为签名后的永久 URL
-        assert job.output_video_url == "https://oss.example.com/lipsync-outputs/u1/j1.mp4?signed"
-        assert job.updated_at is not None
-        session.commit.assert_called_once()
-        session.close.assert_called_once()
-
-    def test_download_failure_keeps_temp_url_no_commit(self):
-        """下载失败（raise）→ 记录 warning、保留 temp_url、不抛异常."""
-        from app.tasks.lipsync_tts import persist_output_video_task
-
-        job = self._make_persist_job(output_video_url="https://temp.mk/x.mp4")
-        session, fake_client, storage, patches = self._persist_patches(
-            job=job, download_side_effect=RuntimeError("network down")
-        )
-        entered = [p.__enter__() for p in patches]
-        try:
-            persist_output_video_task("job-1", "user-1", "https://temp.mk/x.mp4")
-        finally:
-            for p in reversed(patches):
-                p.__exit__(None, None, None)
-
-        storage.upload_file.assert_not_called()
-        # output_video_url 保持原值（temp_url）
-        assert job.output_video_url == "https://temp.mk/x.mp4"
-        # 内层 except 不会 commit
-        # 注：若内部发生 commit 说明测试失败
-        session.close.assert_called_once()
-
-    def test_empty_temp_url_skips_persist(self):
-        """temp_url 为空 → 直接返回，不下载不上传."""
-        from app.tasks.lipsync_tts import persist_output_video_task
-
-        job = self._make_persist_job(output_video_url="")
-        session, fake_client, storage, patches = self._persist_patches(job=job)
-        entered = [p.__enter__() for p in patches]
-        try:
-            persist_output_video_task("job-1", "user-1", "")
-        finally:
-            for p in reversed(patches):
-                p.__exit__(None, None, None)
-
-        fake_client.get.assert_not_called()
-        storage.upload_file.assert_not_called()
-        session.commit.assert_not_called()
-        session.close.assert_called_once()
-
-    def test_job_not_found_returns_early(self):
-        """DB 中找不到 job → 直接返回，不抛错."""
-        from app.tasks.lipsync_tts import persist_output_video_task
-
-        session, fake_client, storage, patches = self._persist_patches(job=None)
-        entered = [p.__enter__() for p in patches]
-        try:
-            persist_output_video_task("missing", "user-1", "https://temp.mk/x.mp4")
-        finally:
-            for p in reversed(patches):
-                p.__exit__(None, None, None)
-
-        fake_client.get.assert_not_called()
-        storage.upload_file.assert_not_called()
-        session.commit.assert_not_called()
-        session.close.assert_called_once()
-
-
-class TestLipsyncServiceRefreshCompletedAsyncPersist:
-    """refresh_job_status 在 completed 分支异步转存的单元测试（补 0% 覆盖的 316~335 行）."""
-
-    def test_refresh_completed_dispatches_persist_task(self):
-        """completed 分支：设置 temp_url → commit → dispatch persist_output_video_task.apply_async."""
-        from app.services.lipsync_service import LipsyncService
-
-        mock_job = MagicMock()
-        mock_job.id = "job-1"
-        mock_job.user_id = "user-1"
-        mock_job.mediakit_task_id = "mk-1"
-        mock_job.status = "submitted"
-        mock_job.output_video_url = ""
-        mock_job.output_duration = 0.0
-
-        mock_db = MagicMock()
-        mock_query = MagicMock()
-        mock_filter = MagicMock()
-        mock_filter.first.return_value = mock_job
-        mock_query.filter.return_value = mock_filter
-        mock_db.query.return_value = mock_query
-
-        mock_client = MagicMock()
-        mock_client.get_task_status.return_value = {
-            "status": "completed",
-            "result": {"video_url": "https://temp.mk/out.mp4", "duration": 25.5},
-        }
-
-        fake_persist_task = MagicMock()
-        svc = LipsyncService(mock_db, client=mock_client, cosyvoice_service=MagicMock())
-        with patch.dict("sys.modules", {}):
-            # 直接 patch 懒 import 路径
-            with patch("app.tasks.lipsync_tts.persist_output_video_task", fake_persist_task, create=False):
-                # 但懒 import 发生在函数内部 from app.tasks.lipsync_tts import persist_output_video_task
-                # 通过 patch sys.modules 的方式提供
-                import sys as _sys
-
-                fake_mod = MagicMock()
-                fake_mod.persist_output_video_task = fake_persist_task
-                _sys.modules["app.tasks.lipsync_tts"] = fake_mod
-                try:
-                    result = svc.refresh_job_status("job-1", "user-1")
-                finally:
-                    _sys.modules.pop("app.tasks.lipsync_tts", None)
-
-        assert result.status == "completed"
-        assert result.output_video_url == "https://temp.mk/out.mp4"
-        assert result.output_duration == 25.5
-        mock_db.commit.assert_called()
-        # 必须在 commit 之后 dispatch
-        fake_persist_task.apply_async.assert_called_once()
-        kwargs = fake_persist_task.apply_async.call_args.kwargs
-        assert kwargs["args"] == ("job-1", "user-1", "https://temp.mk/out.mp4")
-
-    def test_refresh_completed_dispatch_exception_does_not_break_return(self):
-        """apply_async 抛异常（如 Celery 不可用）→ 捕获 warning，仍返回 completed job."""
-        from app.services.lipsync_service import LipsyncService
-
-        mock_job = MagicMock()
-        mock_job.id = "job-2"
-        mock_job.user_id = "user-1"
-        mock_job.mediakit_task_id = "mk-2"
-        mock_job.status = "submitted"
-        mock_job.output_video_url = ""
-        mock_job.output_duration = 0.0
-
-        mock_db = MagicMock()
-        mock_query = MagicMock()
-        mock_filter = MagicMock()
-        mock_filter.first.return_value = mock_job
-        mock_query.filter.return_value = mock_filter
-        mock_db.query.return_value = mock_query
-
-        mock_client = MagicMock()
-        mock_client.get_task_status.return_value = {
-            "status": "completed",
-            "result": {"video_url": "https://temp.mk/out2.mp4", "duration": 10.0},
-        }
-
-        fake_persist_task = MagicMock()
-        fake_persist_task.apply_async.side_effect = ConnectionError("celery down")
-
-        svc = LipsyncService(mock_db, client=mock_client, cosyvoice_service=MagicMock())
-        import sys as _sys
-
-        fake_mod = MagicMock()
-        fake_mod.persist_output_video_task = fake_persist_task
-        _sys.modules["app.tasks.lipsync_tts"] = fake_mod
-        try:
-            result = svc.refresh_job_status("job-2", "user-1")
-        finally:
-            _sys.modules.pop("app.tasks.lipsync_tts", None)
-
-        # 即便 dispatch 失败，主流程不受影响：仍然返回 completed + temp_url
-        assert result.status == "completed"
-        assert result.output_video_url == "https://temp.mk/out2.mp4"
-        fake_persist_task.apply_async.assert_called_once()

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):
@@ -1031,37 +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)
-
-    @patch("packages.domain.video_filter_builder._resolve_font_path")
-    def test_bold_default_uses_black_stroke_when_no_bold_font(self, mock_font):
-        """默认 bold=true 且无 Bold 字体文件时，使用黑色细描边（borderw=2 + 黑），
-        不得使用与文字同色的 borderw>=3（否则会造成竖屏小字号重影）。"""
-        mock_font.return_value = ""  # 无粗体字体
-        result = build_title_drawtext_filter({"text": "标题"})
-        self.assertIsNotNone(result)
-        self.assertIn("borderw=2", result)
-        # 黑描边：要么是 black 关键字，要么是 000000
-        self.assertTrue("bordercolor=black" in result or "bordercolor=000000" in result)
-        self.assertNotIn("borderw=3", result)
-
-    @patch("packages.domain.video_filter_builder._resolve_font_path")
-    def test_bold_with_user_stroke_preserves_user_color(self, mock_font):
-        """用户显式开启 stroke 时，stroke 颜色/宽度优先于默认粗体黑边。"""
-        mock_font.return_value = ""
-        result = build_title_drawtext_filter(
-            {"text": "标题", "bold": True, "stroke": {"width": 4, "color": "#ffffff"}}
-        )
-        self.assertIsNotNone(result)
-        self.assertIn("borderw=4", result)
-        self.assertIn("bordercolor=ffffff", result)  # 去掉 # 前缀
-
 
 class TestDrawtextPositionBranches(unittest.TestCase):
     """位置相关分支覆盖。"""
@@ -1088,23 +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):