chore(ci): 重新触发 staging 部署（首次执行服务器拉 ACR 网络超时，镜像已就绪）

Co-authored-by: xiaoxia <dev@xiaoxiajianji.com>
Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>

alembic/versions/083_cover_title_config.py

@@ -0,0 +1,26 @@
+"""#2001 爆款标题样式面板升级: ai_avatar_render_jobs 新增 cover_title_config
+
+Revision ID: 083_cover_title_config
+Revises: 082_atom_clip_ai_tags
+Create Date: 2026-09-20
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "083_cover_title_config"
+down_revision = "082_atom_clip_ai_tags"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "ai_avatar_render_jobs",
+        sa.Column("cover_title_config", sa.JSON(), nullable=False, server_default=sa.text("'{}'")),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("ai_avatar_render_jobs", "cover_title_config")

alembic/versions/084_lipsync_jobs_style.py

@@ -0,0 +1,26 @@
+"""lipsync_jobs 新增 style 字段（TTS 语气风格）
+
+Revision ID: 084_lipsync_jobs_style
+Revises: 083_cover_title_config
+Create Date: 2026-09-21
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "084_lipsync_jobs_style"
+down_revision = "083_cover_title_config"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("style", sa.String(length=32), nullable=False, server_default=""),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("lipsync_jobs", "style")

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

@@ -63,6 +63,7 @@ def create_render_job(
             b_roll_segments=[s.model_dump() for s in body.b_roll_segments],
             title_config=body.title_config,
             cover_config=body.cover_config,
+            cover_title_config=body.cover_title_config,
             project_id=body.project_id,
         )
     except AiAvatarRenderError as exc:

apps/api/app/api/routes/gpu_lipsync.py

@@ -93,7 +93,7 @@ def register_worker(
     svc: GpuLipsyncService = Depends(_get_svc),
     _token: str = Depends(_verify_gpu_token),
 ):
-    svc.register_worker(
+    worker, cancel_task = svc.register_worker(
         worker_id=body.worker_id,
         hostname=body.hostname,
         gpu_name=body.gpu_name,
@@ -101,7 +101,7 @@ def register_worker(
         capabilities=body.capabilities,
         task_id=body.task_id,
     )
-    return GpuWorkerRegisterResponse(ok=True, server_time=datetime.now(UTC), message="ok")
+    return GpuWorkerRegisterResponse(ok=True, server_time=datetime.now(UTC), message="ok", cancel_task=cancel_task)
 
 
 # ── GET /lipsync/poll — Worker 轮询拉任务 ─────────────────────────

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

@@ -111,9 +111,9 @@ def create_lipsync_job(
             voice_id=body.voice_id,
             script_text=body.script_text,
             speed=body.speed,
-            volume=body.volume if hasattr(body, "volume") and body.volume is not None else 50,
+            style=body.style or "",
+            volume=body.volume if body.volume is not None else 50,
             emotion=body.emotion,
-            style=body.style if hasattr(body, "style") else "",
             enable_video_loop=body.enable_video_loop,
             project_id=body.project_id,
         )
@@ -213,8 +213,9 @@ def preview_tts(
             voice_id=body.voice_id,
             script_text=body.script_text,
             speed=body.speed,
+            style=body.style or "",
+            volume=body.volume if body.volume is not None else 50,
             emotion=body.emotion,
-            style=getattr(body, "style", "") or "",
         )
     except MediaKitError as exc:
         if _points_deducted > 0 and _points_svc is not None:
@@ -345,13 +346,13 @@ def cancel_lipsync_job(
     current_user: AuthenticatedUser = Depends(get_current_user),
     svc: LipsyncService = Depends(_get_service),
 ):
-    """取消对口型任务（仅 pending/tts_processing/submitted 状态可取消）."""
+    """取消对口型任务（仅 pending/tts_processing/submitted/processing 状态可取消）."""
     job = svc.cancel_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="任务不存在")
     if job.status != "cancelled":
         raise HTTPException(
             status_code=400,
-            detail=f"任务状态 {job.status} 不可取消，仅 pending/tts_processing/submitted 可取消",
+            detail=f"任务状态 {job.status} 不可取消，仅 pending/tts_processing/submitted/processing 可取消",
         )
     return job

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

@@ -203,12 +203,13 @@ def synthesize(
         # job.voice_id 统一存解析后的 CosyVoice voice_id
         actual_voice_id = resolved_profile.voice_id
 
-    # 语速/情绪/风格/音量等合成参数随 metadata 落库，workflow 提交 CosyVoice 时读取透传
+    # 语速/情绪等合成参数随 metadata 落库，workflow 提交 CosyVoice 时读取透传
     synthesis_meta = {
         "speed": request.speed,
-        "volume": request.volume if request.volume is not None else 50,
         "emotion": request.emotion or "",
         "style": request.style or "",
+        "volume": request.volume if request.volume is not None else 50,
+        "pitch": request.pitch if request.pitch is not None else 1.0,
         "language": request.language or "zh-CN",
     }
     if request.metadata_:
@@ -656,10 +657,11 @@ def preview_tts(
             text=request.text,
             voice_id=actual_voice_id,
             speed=request.speed,
+            style=request.style or "",
+            volume=request.volume if request.volume is not None else 50,
+            pitch=request.pitch,
             emotion=request.emotion,
-            style=getattr(request, "style", "") or "",
             language=getattr(request, "language", "zh-CN"),
-            pitch=getattr(request, "pitch", 1.0),
         )
     except (CosyVoiceError, ValueError) as e:
         # 合成失败退费

apps/api/app/schemas/ai_avatar_render.py

@@ -53,9 +53,14 @@ class CreateAiAvatarRenderRequest(BaseModel):
     script_id: str = Field("", description="文案 ID（选自文案库时传；手动输入文案直生场景可留空）")
     b_roll_segments: list[BRollSegment] = Field(default_factory=list, description="B-roll 片段列表")
     title_config: dict[str, Any] = Field(
-        default_factory=dict, description="标题配置（可含 title_image_dataurl：前端 Canvas 渲染的标题 PNG dataURL）"
+        default_factory=dict,
+        description="标题配置（可含 title_image_dataurl：前端 Canvas 渲染的标题 PNG dataURL；含 line_overrides 逐行样式）",
     )
     cover_config: dict[str, Any] = Field(default_factory=dict, description="封面配置")
+    cover_title_config: dict[str, Any] = Field(
+        default_factory=dict,
+        description="封面独立标题配置（#2001），结构同 title_config；为空时封面不叠标题",
+    )
     project_id: str = Field("", description="项目 ID")
 
     @field_validator("lipsync_job_id")
@@ -83,6 +88,7 @@ class AiAvatarRenderJobResponse(BaseModel):
     b_roll_segments: list[dict[str, Any]]
     title_config: dict[str, Any]
     cover_config: dict[str, Any]
+    cover_title_config: dict[str, Any] = Field(default_factory=dict, description="封面独立标题配置")
     status: str
     progress: int
     output_video_url: str

apps/api/app/schemas/gpu_lipsync.py

@@ -36,6 +36,7 @@ class GpuWorkerRegisterResponse(BaseModel):
     ok: bool = True
     server_time: datetime
     message: str = "ok"
+    cancel_task: bool = Field(False, description="当前心跳任务是否已被用户取消；为 true 时 Worker 应终止推理")
 
 
 # ── 轮询任务 ────────────────────────────────────────────────────

apps/api/app/schemas/lipsync.py

@@ -29,6 +29,7 @@ class LipsyncJobResponse(BaseModel):
     voice_id: str = ""
     script_text: str = ""
     speed: float = 1.0
+    style: str = ""
     emotion: str = ""
     mediakit_task_id: str
     status: str
@@ -67,14 +68,14 @@ class CreateLipsyncJobRequest(BaseModel):
     voice_id: str = Field("", description="音色 ID（预置音色或克隆音色 profile UUID）")
     script_text: str = Field("", description="要合成的文案（直生模式必填，最长 5000 字符）")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速（0.5-2.0），默认 1.0")
+    style: Optional[str] = Field(
+        None,
+        description="语气风格（natural/sweet/excited/professional/news/livestream），可选；优先级高于 emotion",
+    )
     volume: Optional[int] = Field(None, ge=0, le=100, description="音量（0-100），默认 50")
     emotion: str = Field(
         "",
-        description="情绪（英文枚举 neutral/happy/sad/angry/surprised/fearful/disgusted，或中文 中立/开心/难过/生气/惊讶/恐惧/厌恶；空为默认自然）",
-    )
-    style: Optional[str] = Field(
-        "",
-        description="语气风格（natural/excited/professional/gentle/news/livestream），可选；与 emotion 互斥，style 优先级更高",
+        description="[deprecated] 旧情绪参数，内部映射为 style",
     )
 
     enable_video_loop: bool = Field(
@@ -128,14 +129,15 @@ class AiAvatarTtsPreviewRequest(BaseModel):
     voice_id: str = Field(..., min_length=1, max_length=128, description="音色 ID")
     script_text: str = Field(..., min_length=1, max_length=5000, description="要合成的文案")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速（0.5-2.0），默认 1.0")
+    style: Optional[str] = Field(
+        None,
+        description="语气风格（natural/sweet/excited/professional/news/livestream），可选；优先级高于 emotion",
+    )
+    volume: Optional[int] = Field(None, ge=0, le=100, description="音量（0-100），默认 50")
     emotion: str = Field(
         "neutral",
         max_length=32,
-        description="情绪（英文枚举 neutral/happy/sad/angry/surprised/fearful/disgusted，或中文 中立/开心/难过/生气/惊讶/恐惧/厌恶；默认 neutral）",
-    )
-    style: Optional[str] = Field(
-        "",
-        description="语气风格（natural/excited/professional/gentle/news/livestream），可选；与 emotion 互斥，style 优先级更高",
+        description="[deprecated] 旧情绪参数，内部映射为 style；默认 neutral",
     )
 
 

apps/api/app/schemas/tts.py

@@ -16,14 +16,15 @@ class TTSSynthesizeRequest(BaseModel):
     output_name: str = Field("", description="输出文件名")
     language: str = Field("zh-CN", description="语言")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
+    style: Optional[str] = Field(
+        None,
+        description="语气风格（natural/sweet/excited/professional/news/livestream），可选；优先级高于 emotion",
+    )
     volume: Optional[int] = Field(None, ge=0, le=100, description="音量（0-100），默认 50")
+    pitch: Optional[float] = Field(None, ge=0.5, le=2.0, description="音调（0.5-2.0），默认 1.0")
     emotion: str = Field(
         "",
-        description="情绪（中文/英文：自然/兴奋/沉稳/亲切/开心/悲伤/愤怒/惊讶/恐惧/厌恶 等；通过 instruction 自然语言指令控制）",
-    )
-    style: Optional[str] = Field(
-        "",
-        description="语气风格（natural/excited/professional/gentle/news/livestream），可选；与 emotion 互斥，style 优先级更高",
+        description="[deprecated] 旧情绪参数，内部映射为 style；新接入请使用 style",
     )
     voice_model: str = Field("", description="语音模型名称")
     voice_clone_profile_id: str = Field("", description="关联的音色克隆档案 ID")
@@ -118,13 +119,14 @@ class TTSPreviewRequest(BaseModel):
     text: str = Field(..., min_length=1, max_length=200, description="合成文本，限制 200 字")
     voice_id: str = Field(..., min_length=1, description="音色 ID")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
-    emotion: str = Field("", description="情绪（中文/英文：自然/兴奋/沉稳/亲切/开心/悲伤/愤怒/惊讶/恐惧/厌恶 等）")
     style: Optional[str] = Field(
-        "",
-        description="语气风格（natural/excited/professional/gentle/news/livestream），可选；与 emotion 互斥，style 优先级更高",
+        None,
+        description="语气风格（natural/sweet/excited/professional/news/livestream），可选；优先级高于 emotion",
     )
+    volume: Optional[int] = Field(None, ge=0, le=100, description="音量（0-100），默认 50")
+    emotion: str = Field("", description="[deprecated] 旧情绪参数，内部映射为 style")
     language: str = Field("zh-CN", description="语言（zh-CN/en-US 等）")
-    pitch: float = Field(1.0, ge=0.5, le=2.0, description="音调（0.5-2.0），1.0 为默认值")
+    pitch: float = Field(1.0, ge=0.5, le=2.0, description="音调（0.5-2.0），默认 1.0")
 
 
 class TTSPreviewResponse(BaseModel):

apps/api/app/services/ai_avatar_render_service.py

@@ -61,6 +61,7 @@ class AiAvatarRenderService:
         b_roll_segments: list[dict[str, Any]] | None = None,
         title_config: dict[str, Any],
         cover_config: dict[str, Any],
+        cover_title_config: dict[str, Any] | None = None,
         project_id: str = "",
     ) -> AiAvatarRenderJob:
         """创建渲染任务.
@@ -112,6 +113,7 @@ class AiAvatarRenderService:
             b_roll_segments=[s if isinstance(s, dict) else s.model_dump() for s in (b_roll_segments or [])],
             title_config=title_config,
             cover_config=cover_config,
+            cover_title_config=cover_title_config or {},
             status="pending",
         )
         self.db.add(job)

apps/api/app/services/gpu_lipsync_service.py

@@ -50,13 +50,16 @@ class GpuLipsyncService:
         free_vram_mb: int = 0,
         capabilities: str = "musetalk",
         task_id: Optional[str] = None,
-    ) -> GpuWorkerModel:
+    ) -> tuple[GpuWorkerModel, bool]:
         """Worker 注册/心跳。
 
         task_id 非空时（Worker 推理期间的任务级心跳），同步把对应 processing
         任务的 last_heartbeat_at 续到当前时间，使长推理不会被
         ``_recover_timed_out_tasks`` 误回退。任务已结束 / 不属于该 worker
         （如已被超时回收重新派发）时忽略，不报错。
+
+        返回 ``(worker, cancel_task)``：当心跳任务已被用户取消时
+        ``cancel_task=True``，Worker 应尽快终止推理并释放 GPU。
         """
         now = datetime.now(UTC)
         worker = self.db.query(GpuWorkerModel).filter(GpuWorkerModel.worker_id == worker_id).one_or_none()
@@ -77,10 +80,11 @@ class GpuLipsyncService:
             worker.free_vram_mb = free_vram_mb
             worker.capabilities = capabilities or worker.capabilities
             worker.last_heartbeat_at = now
+        cancel_task = False
         if task_id:
-            self._touch_task_heartbeat(task_id, worker_id, now)
+            cancel_task = self._touch_task_heartbeat(task_id, worker_id, now)
         self.db.commit()
-        return worker
+        return worker, cancel_task
 
     # ── 轮询拉任务（Worker 调用） ──────────────────────────────────
 
@@ -166,6 +170,11 @@ class GpuLipsyncService:
             task.result_duration = duration_seconds or 0.0
             task.error_msg = ""
             task.finished_at = now
+        elif task.status == "cancelled":
+            # 用户已取消的任务，Worker 终止后上报失败，保持 cancelled 状态不回退
+            task.finished_at = now
+            task.error_msg = (error_msg or "用户取消")[:2000]
+            logger.info("GPU 任务 %s 已被用户取消，保持 cancelled 状态", task_id)
         else:
             # 失败：若仍可重试（已尝试次数 < MAX_ATTEMPTS）→ 回退 pending；否则 → failed
             if task.attempt < MAX_ATTEMPTS:
@@ -250,15 +259,21 @@ class GpuLipsyncService:
     def _result_key(self, task_id: str) -> str:
         return f"{self.RESULT_PREFIX}{task_id}.mp4"
 
-    def _touch_task_heartbeat(self, task_id: str, worker_id: str, now: datetime) -> None:
+    def _touch_task_heartbeat(self, task_id: str, worker_id: str, now: datetime) -> bool:
         """Worker 推理期间的任务级心跳：只刷新属于该 worker 且仍在 processing 的任务。
 
         任务不存在 / 已被超时回收重新派发 / 已完成 → 静默忽略（此时旧 worker 的
         结果上报会被结果接口按最终态处理）。
+
+        返回 ``cancel_task``：任务已被用户取消时为 True，Worker 应终止推理。
         """
         task = self.db.get(GpuLipsyncTaskModel, task_id)
         if task is None:
-            return
+            return False
+        # 任务已被用户取消 → 通知 Worker 终止推理
+        if task.status == "cancelled":
+            logger.info("任务心跳检测到已取消 task=%s worker=%s，通知 Worker 终止", task_id, worker_id)
+            return True
         if task.status != "processing" or task.worker_id != worker_id:
             logger.info(
                 "忽略过期任务心跳 task=%s worker=%s（status=%s owner=%s）",
@@ -267,10 +282,11 @@ class GpuLipsyncService:
                 task.status,
                 task.worker_id,
             )
-            return
+            return False
         task.last_heartbeat_at = now
         task.updated_at = now
         self.db.flush()
+        return False
 
     def _touch_worker(self, worker_id: str, now: datetime) -> None:
         if not worker_id:
@@ -371,9 +387,7 @@ class GpuLipsyncService:
             task = self.db.get(GpuLipsyncTaskModel, task_id)
             if task is None:
                 return None
-            if task.status == "done":
-                return task
-            if task.status == "failed":
+            if task.status in ("done", "failed", "cancelled"):
                 return task
             # pending/processing 继续等
             if time.monotonic() >= deadline:

apps/api/app/services/lipsync_service.py

@@ -111,6 +111,8 @@ class LipsyncService:
         script_text: str,
         speed: float,
         emotion: str,
+        style: str = "",
+        volume: int = 50,
     ) -> str:
         """TTS 直生：调 CosyVoice 合成音频并转存 OSS，返回可公网访问的音频 URL.
 
@@ -124,7 +126,9 @@ class LipsyncService:
                 text=script_text,
                 voice_id=actual_voice_id,
                 speed=speed,
-                emotion=emotion,  # normalize 在 CosyVoiceService 内部完成
+                style=style,
+                volume=volume,
+                emotion=emotion,
                 language="zh",
             )
         except CosyVoiceError as exc:
@@ -424,9 +428,9 @@ class LipsyncService:
         voice_id: str = "",
         script_text: str = "",
         speed: float = 1.0,
+        style: str = "",
         volume: int = 50,
         emotion: str = "",
-        style: str = "",
         enable_video_loop: bool = True,
         project_id: str = "",
     ) -> LipsyncJobModel:
@@ -474,6 +478,7 @@ class LipsyncService:
             voice_id=voice_id or "",
             script_text=script_text or "",
             speed=speed,
+            style=style or "",
             emotion=emotion or "",
             # 音频直传（含预合成）直接进入 pending（后续同步改为 submitted）；TTS 模式进入 tts_processing
             status="tts_processing" if is_tts_mode else "pending",
@@ -495,9 +500,9 @@ class LipsyncService:
                         voice_id,
                         script_text,
                         speed,
-                        emotion or "",
-                        volume,
                         style or "",
+                        volume,
+                        emotion or "",
                     )
                 )
             except Exception as exc:
@@ -531,8 +536,9 @@ class LipsyncService:
         voice_id: str,
         script_text: str,
         speed: float = 1.0,
-        emotion: str = "neutral",
         style: str = "",
+        volume: int = 50,
+        emotion: str = "neutral",
     ) -> dict:
         """同步做 TTS 合成 + 下载 + ffprobe + 句子时间戳计算.
 
@@ -557,7 +563,6 @@ class LipsyncService:
                 speed=speed,
                 emotion=emotion,  # normalize 在 CosyVoiceService 内部完成
                 language="zh",
-                style=style,
             )
         except CosyVoiceError as exc:
             raise MediaKitError(f"TTS 合成失败: {exc}", code="TTSSynthesisFailed") from exc
@@ -778,12 +783,32 @@ class LipsyncService:
     # ── 取消任务 ──────────────────────────────────────────────────────────
 
     def cancel_job(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
-        """取消任务（仅 pending/tts_processing/submitted 状态可取消）."""
+        """取消任务（pending/tts_processing/submitted/processing 状态可取消）.
+
+        当 job 走 GPU 路径（mediakit_task_id 以 "gpu:" 开头）且状态为 processing 时，
+        同步将关联的 GpuLipsyncTask 标记为 cancelled，以便 Worker 心跳时检测到取消信号。
+        """
         job = self.get_job(job_id, user_id)
         if job is None:
             return None
 
-        if job.status in ("pending", "tts_processing", "submitted"):
+        if job.status in ("pending", "tts_processing", "submitted", "processing"):
+            # GPU 路径：同步标记关联的 GPU 任务为 cancelled
+            if job.status == "processing" and job.mediakit_task_id and job.mediakit_task_id.startswith("gpu:"):
+                gpu_task_id = job.mediakit_task_id[4:]  # 去掉 "gpu:" 前缀
+                try:
+                    from packages.adapters.sqlalchemy_impl.models import GpuLipsyncTaskModel
+
+                    gpu_task = self.db.get(GpuLipsyncTaskModel, gpu_task_id)
+                    if gpu_task and gpu_task.status == "processing":
+                        gpu_task.status = "cancelled"
+                        gpu_task.error_msg = "用户取消"
+                        gpu_task.updated_at = datetime.now(UTC)
+                        gpu_task.finished_at = datetime.now(UTC)
+                        logger.info("GPU 任务 %s 已被用户取消（通过 job_id=%s）", gpu_task_id, job_id)
+                except Exception as exc:
+                    logger.warning("标记 GPU 任务取消失败（不影响 job 取消）: %s", exc)
+
             job.status = "cancelled"
             job.updated_at = datetime.now(UTC)
             self.db.commit()

apps/api/app/tasks/lipsync_gpu.py

@@ -98,6 +98,14 @@ def lipsync_gpu_process_async(self, job_id: str, user_id: str, gpu_task_id: str)
             _fallback_to_mediakit(db, job)
             return
 
+        if final_task.status == "cancelled":
+            # 用户已取消任务，不回退 MediaKit，直接标记 job 为 cancelled
+            job.status = "cancelled"
+            job.updated_at = datetime.now(UTC)
+            db.commit()
+            logger.info("[lipsync_gpu_async] GPU 任务已被用户取消: job_id=%s", job_id)
+            return
+
         if final_task.status != "done":
             logger.warning(
                 "[lipsync_gpu_async] GPU 失败，回退 MediaKit: job_id=%s gpu_task=%s status=%s",

apps/api/app/tasks/lipsync_tts.py

@@ -82,9 +82,9 @@ def tts_synthesize_and_submit(
     voice_id: str,
     script_text: str,
     speed: float,
-    emotion: str,
-    volume: int = 50,
     style: str = "",
+    volume: int = 50,
+    emotion: str = "",
 ):
     """异步执行 TTS 合成 + OSS 转存 + MediaKit 提交.
 
@@ -161,10 +161,10 @@ def tts_synthesize_and_submit(
                 text=script_text,
                 voice_id=voice_id,
                 speed=speed,
+                style=style,
+                volume=volume,
                 emotion=emotion,
                 language="zh",
-                volume=volume,
-                style=style,
             )
         except CosyVoiceError as exc:
             logger.error("[lipsync_tts] TTS 合成失败: job_id=%s err=%s", job_id, exc)

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

@@ -7,8 +7,19 @@ export interface GenerateCoverTitleConfig {
   font_color?: string
   position?: string
   bold?: boolean
-  stroke?: boolean
-  shadow?: boolean
+  italic?: boolean
+  stroke?: boolean | { enabled?: boolean; width?: number; color?: string }
+  shadow?:
+    | boolean
+    | { enabled?: boolean; offset_x?: number; offset_y?: number; blur?: number; color?: string }
+  line_height?: number
+  margin_top?: number
+  max_chars_per_line?: number
+  background?: { enabled?: boolean; color?: string; padding?: number; radius?: number }
+  line_overrides?: Array<Record<string, unknown>>
+  cover_title_config?: Record<string, unknown>
+  pos_x?: number
+  pos_y?: number
 }
 
 export interface GenerateCoverRequest {

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

@@ -81,7 +81,7 @@ export interface CreateGenerationTaskRequest {
   tts_voice_source?: "preset" | "clone"
   /** #1970：智能降重开关（默认 true） */
   dedup_enabled?: boolean
-  /** 标题烧录配置 */
+  /** 标题烧录配置（#2001 扩展：描边/阴影参数/行距/自动换行/背景/逐行/封面） */
   title_config?: {
     text?: string
     font?: string
@@ -89,8 +89,34 @@ export interface CreateGenerationTaskRequest {
     font_color?: string
     position?: string
     bold?: boolean
-    stroke?: boolean
-    shadow?: boolean
+    italic?: boolean
+    stroke?: boolean | { enabled?: boolean; width?: number; color?: string }
+    shadow?:
+      | boolean
+      | {
+          enabled?: boolean
+          offset_x?: number
+          offset_y?: number
+          blur?: number
+          color?: string
+        }
+    line_height?: number
+    margin_top?: number
+    max_chars_per_line?: number
+    background?: { enabled?: boolean; color?: string; padding?: number; radius?: number }
+    line_overrides?: Array<{
+      line_index: number
+      text?: string
+      size?: number
+      color?: string
+      bold?: boolean
+      italic?: boolean
+      stroke?: boolean
+      highlights?: Array<{ word: string; color?: string; bold?: boolean; scale?: number }>
+    }>
+    cover_title_config?: Record<string, unknown>
+    pos_x?: number
+    pos_y?: number
   }
   /** 关联的草稿 ID（编辑流程数据链路用） */
   source_edit_plan_id?: string

apps/web/src/components/title/constants.ts

@@ -0,0 +1,361 @@
+/**
+ * 标题样式相关常量（#2001）
+ * - 字体列表（新增4款爆款字体）
+ * - 色板（常用标题字色/描边色/背景色）
+ * - 预设样式方案（10 个，含抖音爆款黄）
+ */
+import type { TitleStyleConfig } from "./types"
+
+/* ── 字体选项（#2001：新增优设标题黑/阿里普惠体Bold/抖音美好体/思源黑体Heavy） ── */
+export interface FontOption {
+  value: string
+  label: string
+  /** CSS font-family 栈 */
+  family: string
+  /** 爆款/常用标签 */
+  tag?: "hot" | "new"
+}
+
+export const FONT_OPTIONS: FontOption[] = [
+  {
+    value: "优设标题黑",
+    label: "优设标题黑",
+    family:
+      '"YouShe Title Black","YouSheBiaoTiHei","Source Han Sans SC Heavy","Noto Sans SC","PingFang SC",sans-serif',
+    tag: "hot",
+  },
+  {
+    value: "阿里普惠体Bold",
+    label: "阿里普惠体Bold",
+    family:
+      '"Alibaba PuHuiTi Bold","Alibaba PuHuiTi","Source Han Sans SC","PingFang SC",sans-serif',
+    tag: "hot",
+  },
+  {
+    value: "抖音美好体",
+    label: "抖音美好体",
+    family: '"Douyin Sans","DouyinSans","Source Han Sans SC","PingFang SC",sans-serif',
+    tag: "hot",
+  },
+  {
+    value: "思源黑体Heavy",
+    label: "思源黑体Heavy",
+    family:
+      '"Source Han Sans SC Heavy","Noto Sans SC","Source Han Sans CN Heavy","PingFang SC",sans-serif',
+    tag: "new",
+  },
+  {
+    value: "思源黑体",
+    label: "思源黑体",
+    family: '"Source Han Sans SC","Noto Sans SC","PingFang SC","Microsoft YaHei",sans-serif',
+  },
+  {
+    value: "思源宋体",
+    label: "思源宋体",
+    family: '"Source Han Serif SC","Noto Serif SC","Songti SC","SimSun",serif',
+  },
+  {
+    value: "苹方",
+    label: "苹方",
+    family: '"PingFang SC",-apple-system,"Helvetica Neue",sans-serif',
+  },
+  {
+    value: "微软雅黑",
+    label: "微软雅黑",
+    family: '"Microsoft YaHei","PingFang SC",sans-serif',
+  },
+  {
+    value: "楷体",
+    label: "楷体",
+    family: '"KaiTi","STKaiti","DFKai-SB",serif',
+  },
+]
+
+/** 根据中文名取 font-family 栈（找不到回退思源黑体） */
+export function getFontFamily(font: string): string {
+  const f = FONT_OPTIONS.find((x) => x.value === font)
+  if (f) return f.family
+  return FONT_OPTIONS[4].family // 思源黑体
+}
+
+/* ── 色板 ── */
+/** 标题字色（常用爆款色） */
+export const TITLE_COLOR_PALETTE: string[] = [
+  "#ffffff",
+  "#000000",
+  "#ffd700", // 抖音黄
+  "#ff2d55", // 抖音红
+  "#ff4081",
+  "#00e5ff",
+  "#d4a843",
+  "#ffa500",
+  "#52c41a",
+  "#1890ff",
+  "#7c3aed",
+  "#ff6b35",
+]
+
+/** 描边色（黑/白/灰为主） */
+export const STROKE_COLOR_PALETTE: string[] = [
+  "#000000",
+  "#ffffff",
+  "#333333",
+  "#555555",
+  "#8b0000",
+  "#001f3f",
+]
+
+/** 背景色（带透明度） */
+export const BG_COLOR_PALETTE: string[] = [
+  "rgba(0,0,0,0.5)",
+  "rgba(0,0,0,0.7)",
+  "rgba(0,0,0,0.3)",
+  "rgba(255,215,0,0.9)",
+  "rgba(255,45,85,0.85)",
+  "rgba(124,58,237,0.85)",
+  "rgba(24,144,255,0.85)",
+  "rgba(82,196,26,0.85)",
+]
+
+/* ── 预设样式方案（10 个，含抖音爆款黄） ── */
+export interface TitlePreset {
+  key: string
+  label: string
+  emoji: string
+  /** 应用时覆盖到 TitleStyleConfig 的字段（其他字段保持当前值） */
+  style: Partial<TitleStyleConfig>
+}
+
+const BASE: Partial<TitleStyleConfig> = {
+  line_overrides: [],
+  cover_title_config: null,
+}
+
+export const TITLE_PRESETS: TitlePreset[] = [
+  {
+    key: "douyin_hot",
+    label: "抖音爆款黄",
+    emoji: "🔥",
+    style: {
+      ...BASE,
+      font: "优设标题黑",
+      size: 80,
+      color: "#ffd700",
+      bold: true,
+      italic: false,
+      stroke: true,
+      stroke_width: 8,
+      stroke_color: "#000000",
+      shadow: true,
+      shadow_offset_x: 3,
+      shadow_offset_y: 3,
+      shadow_blur: 6,
+      shadow_color: "rgba(0,0,0,0.6)",
+      bg_enabled: false,
+      line_height: 1.25,
+      max_chars_per_line: 8,
+    },
+  },
+  {
+    key: "classic_white",
+    label: "经典白字黑描边",
+    emoji: "⚪",
+    style: {
+      ...BASE,
+      font: "思源黑体Heavy",
+      size: 56,
+      color: "#ffffff",
+      bold: true,
+      italic: false,
+      stroke: true,
+      stroke_width: 5,
+      stroke_color: "#000000",
+      shadow: false,
+      bg_enabled: false,
+      line_height: 1.2,
+      max_chars_per_line: 10,
+    },
+  },
+  {
+    key: "red_bold",
+    label: "醒目红字",
+    emoji: "🔴",
+    style: {
+      ...BASE,
+      font: "优设标题黑",
+      size: 72,
+      color: "#ff2d55",
+      bold: true,
+      italic: false,
+      stroke: true,
+      stroke_width: 6,
+      stroke_color: "#ffffff",
+      shadow: true,
+      shadow_offset_x: 2,
+      shadow_offset_y: 2,
+      shadow_blur: 5,
+      shadow_color: "rgba(0,0,0,0.5)",
+      bg_enabled: false,
+      line_height: 1.2,
+      max_chars_per_line: 9,
+    },
+  },
+  {
+    key: "black_gold",
+    label: "黑金质感",
+    emoji: "🟡",
+    style: {
+      ...BASE,
+      font: "思源宋体",
+      size: 52,
+      color: "#d4a843",
+      bold: true,
+      italic: false,
+      stroke: false,
+      shadow: true,
+      shadow_offset_x: 2,
+      shadow_offset_y: 2,
+      shadow_blur: 8,
+      shadow_color: "rgba(0,0,0,0.8)",
+      bg_enabled: false,
+      line_height: 1.25,
+      max_chars_per_line: 10,
+    },
+  },
+  {
+    key: "neon_blue",
+    label: "霓虹发光",
+    emoji: "💙",
+    style: {
+      ...BASE,
+      font: "阿里普惠体Bold",
+      size: 60,
+      color: "#00e5ff",
+      bold: true,
+      italic: false,
+      stroke: false,
+      shadow: true,
+      shadow_offset_x: 0,
+      shadow_offset_y: 0,
+      shadow_blur: 16,
+      shadow_color: "#00e5ff",
+      bg_enabled: false,
+      line_height: 1.2,
+      max_chars_per_line: 10,
+    },
+  },
+  {
+    key: "bg_black",
+    label: "黑底白字",
+    emoji: "⬛",
+    style: {
+      ...BASE,
+      font: "思源黑体Heavy",
+      size: 52,
+      color: "#ffffff",
+      bold: true,
+      italic: false,
+      stroke: false,
+      shadow: false,
+      bg_enabled: true,
+      bg_color: "rgba(0,0,0,0.7)",
+      bg_padding: 16,
+      bg_radius: 8,
+      line_height: 1.3,
+      max_chars_per_line: 10,
+    },
+  },
+  {
+    key: "bg_yellow",
+    label: "黄底黑字",
+    emoji: "🟨",
+    style: {
+      ...BASE,
+      font: "抖音美好体",
+      size: 56,
+      color: "#000000",
+      bold: true,
+      italic: false,
+      stroke: false,
+      shadow: false,
+      bg_enabled: true,
+      bg_color: "rgba(255,215,0,0.95)",
+      bg_padding: 14,
+      bg_radius: 6,
+      line_height: 1.2,
+      max_chars_per_line: 9,
+    },
+  },
+  {
+    key: "sweet_pink",
+    label: "温柔甜美粉",
+    emoji: "🌸",
+    style: {
+      ...BASE,
+      font: "阿里普惠体Bold",
+      size: 50,
+      color: "#ff4081",
+      bold: false,
+      italic: false,
+      stroke: true,
+      stroke_width: 4,
+      stroke_color: "#ffffff",
+      shadow: true,
+      shadow_offset_x: 2,
+      shadow_offset_y: 2,
+      shadow_blur: 4,
+      shadow_color: "rgba(255,64,129,0.4)",
+      bg_enabled: false,
+      line_height: 1.3,
+      max_chars_per_line: 11,
+    },
+  },
+  {
+    key: "business_dark",
+    label: "商务深色",
+    emoji: "💼",
+    style: {
+      ...BASE,
+      font: "思源黑体",
+      size: 44,
+      color: "#ffffff",
+      bold: false,
+      italic: false,
+      stroke: false,
+      shadow: true,
+      shadow_offset_x: 1,
+      shadow_offset_y: 1,
+      shadow_blur: 3,
+      shadow_color: "rgba(0,0,0,0.8)",
+      bg_enabled: true,
+      bg_color: "rgba(24,144,255,0.85)",
+      bg_padding: 12,
+      bg_radius: 4,
+      line_height: 1.3,
+      max_chars_per_line: 12,
+    },
+  },
+  {
+    key: "minimal_clean",
+    label: "极简无描边",
+    emoji: "✨",
+    style: {
+      ...BASE,
+      font: "苹方",
+      size: 48,
+      color: "#ffffff",
+      bold: true,
+      italic: false,
+      stroke: false,
+      shadow: false,
+      bg_enabled: false,
+      line_height: 1.3,
+      max_chars_per_line: 10,
+    },
+  },
+]
+
+/** 根据 key 获取预设 */
+export function getTitlePreset(key: string): TitlePreset | undefined {
+  return TITLE_PRESETS.find((p) => p.key === key)
+}

apps/web/src/components/title/types.ts

@@ -0,0 +1,118 @@
+/**
+ * 共享标题样式配置类型（#2001 爆款标题样式配置面板升级）
+ *
+ * 设计原则：
+ * 1. 向后兼容：保留旧的 bold/stroke/shadow 布尔字段，新增细粒度字段
+ *    （stroke_width/stroke_color/shadow_offset_x-y-blur-color/bg_enabled-color-padding-radius/line_height/margin_top/max_chars_per_line）。
+ * 2. 后端契约：字段名使用 snake_case，与 title_config dict 直接对齐。
+ * 3. line_overrides 支持逐行覆盖（选中某行单独设置颜色/字号/关键词高亮/加粗/斜体）。
+ * 4. cover_title_config 为封面独立标题样式，null 表示封面沿用主标题样式。
+ */
+
+/** 关键词高亮配置 */
+export interface TitleKeywordHighlight {
+  /** 要高亮的词 */
+  word: string
+  /** 高亮颜色（可选，默认主色反转） */
+  color?: string
+  /** 是否加粗（默认 true） */
+  bold?: boolean
+  /** 额外字号放大倍数（1.0=不变，1.3=放大 30%） */
+  scale?: number
+}
+
+/** 单行覆盖配置 */
+export interface TitleLineOverride {
+  /** 行索引（0-based，按 / 或自动换行后的行序） */
+  line_index: number
+  /** 覆盖后的文字（可选，默认沿用原行） */
+  text?: string
+  /** 覆盖字号（可选） */
+  size?: number
+  /** 覆盖字色（可选） */
+  color?: string
+  /** 覆盖加粗（可选） */
+  bold?: boolean
+  /** 覆盖斜体（可选） */
+  italic?: boolean
+  /** 覆盖描边开关（可选） */
+  stroke?: boolean
+  /** 关键词高亮列表 */
+  highlights?: TitleKeywordHighlight[]
+}
+
+/**
+ * 标题样式配置（不含 title 文字本身，不含 auto_subtitle）。
+ *
+ * cover_title_config 使用 Partial<Omit<...,"cover_title_config">> 递归避免无限类型。
+ */
+export interface TitleStyleConfig {
+  /* ── 基础 ── */
+  font: string
+  size: number
+  color: string
+  bold: boolean
+  italic: boolean
+  position: "top" | "center" | "bottom" | "custom"
+  pos_x?: number
+  pos_y?: number
+
+  /* ── 排版（P0） ── */
+  /** 行距倍数（默认 1.2） */
+  line_height: number
+  /** 顶部边距（position=top 时距画面顶部距离，px @720p，默认 24） */
+  margin_top: number
+  /** 每行最大字符数（4-20，超出自动换行；0=不自动换行，使用 / 手动分行） */
+  max_chars_per_line: number
+
+  /* ── 描边参数化（P0） ── */
+  stroke: boolean
+  stroke_width: number
+  stroke_color: string
+
+  /* ── 阴影参数化（P1） ── */
+  shadow: boolean
+  shadow_offset_x: number
+  shadow_offset_y: number
+  shadow_blur: number
+  shadow_color: string
+
+  /* ── 背景色块（P1） ── */
+  bg_enabled: boolean
+  bg_color: string
+  bg_padding: number
+  bg_radius: number
+
+  /* ── 逐行独立样式（P1） ── */
+  line_overrides: TitleLineOverride[]
+
+  /* ── 封面独立标题配置（P1）：null=沿用主标题样式 ── */
+  cover_title_config: null | Partial<Omit<TitleStyleConfig, "cover_title_config">>
+}
+
+/** 默认样式（经典白字黑描边，保持老版本观感） */
+export const DEFAULT_TITLE_STYLE: TitleStyleConfig = {
+  font: "思源黑体",
+  size: 48,
+  color: "#ffffff",
+  bold: true,
+  italic: false,
+  position: "bottom",
+  line_height: 1.2,
+  margin_top: 24,
+  max_chars_per_line: 0,
+  stroke: true,
+  stroke_width: 4,
+  stroke_color: "#000000",
+  shadow: false,
+  shadow_offset_x: 2,
+  shadow_offset_y: 2,
+  shadow_blur: 4,
+  shadow_color: "rgba(0,0,0,0.8)",
+  bg_enabled: false,
+  bg_color: "rgba(0,0,0,0.5)",
+  bg_padding: 12,
+  bg_radius: 8,
+  line_overrides: [],
+  cover_title_config: null,
+}

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

@@ -30,11 +30,7 @@ import {
 } from "./api/aiAvatar"
 import { getOrCreateDefaultProject } from "@/api/projects"
 import type { RenderJob, SentenceTiming } from "./types"
-import {
-  normalizeEmotion,
-  buildTitleConfigPayload,
-  buildCoverConfigPayload,
-} from "./utils/contract"
+import { buildTitleConfigPayload, buildCoverConfigPayload } from "./utils/contract"
 import { renderTitleToPngDataUrl, getVideoResolution } from "./utils/titleCanvas"
 
 /** 面板折叠状态 */
@@ -94,7 +90,7 @@ const AiAvatarPage: React.FC = () => {
       state.resetTtsPreview()
     }
     // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [state.scriptText, state.selectedVoice?.voice_id, state.speed, state.emotion, state.style])
+  }, [state.scriptText, state.selectedVoice?.voice_id, state.speed, state.style])
 
   const _clearTtsProgressTimer = useCallback(() => {
     if (ttsProgressTimerRef.current) {
@@ -148,7 +144,7 @@ const AiAvatarPage: React.FC = () => {
         voice_id: state.selectedVoice!.voice_id,
         script_text: state.scriptText,
         speed: state.speed,
-        emotion: normalizeEmotion(state.emotion),
+        style: state.style,
       })
       _clearTtsProgressTimer()
       setTtsProgress(100)
@@ -175,14 +171,7 @@ const AiAvatarPage: React.FC = () => {
       })
     }
     // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [
-    state.selectedVideo,
-    state.selectedVoice,
-    state.scriptText,
-    state.speed,
-    state.emotion,
-    state.style,
-  ])
+  }, [state.selectedVideo, state.selectedVoice, state.scriptText, state.speed, state.style])
 
   const handleRetryTts = useCallback(() => {
     handleGenerateTts()
@@ -261,7 +250,6 @@ const AiAvatarPage: React.FC = () => {
           script_text: state.scriptText,
           video_url: videoUrl,
           speed: state.speed,
-          emotion: normalizeEmotion(state.emotion),
           style: state.style,
         }
       }
@@ -306,7 +294,7 @@ const AiAvatarPage: React.FC = () => {
     state.selectedVoice,
     state.scriptText,
     state.speed,
-    state.emotion,
+
     state.style,
     state.ttsPreview,
   ])
@@ -607,8 +595,6 @@ const AiAvatarPage: React.FC = () => {
                   onVoiceSourceChange={state.setVoiceSource}
                   selectedVoice={state.selectedVoice}
                   onSelectVoice={state.setSelectedVoice}
-                  emotion={state.emotion}
-                  onEmotionChange={state.setEmotion}
                   style={state.style}
                   onStyleChange={state.setStyle}
                   speed={state.speed}

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

@@ -29,15 +29,7 @@ function formatTime(seconds: number): string {
   return `${m}:${s.toString().padStart(2, "0")}`
 }
 
-/** 字体名 → CSS font-family 映射（与 titleCanvas 字体链对齐） */
-const FONT_FAMILY_MAP: Record<string, string> = {
-  思源黑体:
-    "'Noto Sans CJK SC', 'Source Han Sans CN', 'PingFang SC', 'Microsoft YaHei', sans-serif",
-  思源宋体: "'Noto Serif SC', 'Source Han Serif SC', 'SimSun', serif",
-  楷体: "KaiTi, 'STKaiti', serif",
-  黑体: "'Heiti SC', 'SimHei', 'Microsoft YaHei', sans-serif",
-}
-const getFontFamily = (font: string): string => FONT_FAMILY_MAP[font] || FONT_FAMILY_MAP["思源黑体"]
+import { getFontFamily as getFontFamilyByKey } from "@/components/title/constants"
 
 export function PanelLipsyncPreview({
   lipsyncJob,
@@ -83,63 +75,105 @@ export function PanelLipsyncPreview({
   const previewScale = containerWidth > 0 ? containerWidth / 720 : 0.35
   const ps = useCallback((v: number) => Math.round(v * previewScale * 100) / 100, [previewScale])
 
-  /** 标题叠加样式（字号/padding/描边/阴影均按 previewScale 缩放，保持与成片视觉一致） */
-  const titleOverlayStyle: React.CSSProperties | null =
+  /** 标题叠加样式（新字段全支持：描边宽色/阴影参数化/背景块/行距/顶部边距/自动换行） */
+  const titleOverlayData =
     titleConfig?.title && containerWidth > 0
       ? (() => {
+          const c = titleConfig as AiAvatarTitleConfig & {
+            stroke_width?: number
+            stroke_color?: string
+            shadow_offset_x?: number
+            shadow_offset_y?: number
+            shadow_blur?: number
+            shadow_color?: string
+            line_height?: number
+            margin_top?: number
+            max_chars_per_line?: number
+            bg_enabled?: boolean
+            bg_color?: string
+            bg_padding?: number
+            bg_radius?: number
+            cover_title_config?: Record<string, unknown> | null
+            line_overrides?: unknown[]
+          }
           const baseSize = titleConfig.size || 48
           const fontSize = ps(baseSize)
-          // 描边宽度基准 ≈ size * 0.06，最小 1.5px @720p
-          const strokeW = Math.max(ps(1.5), +(baseSize * 0.06 * previewScale).toFixed(2))
-          // 阴影按比例缩放
-          const shadowBlur = ps(4)
-          const shadowOffsetY = ps(2)
-          // padding / top 边距按比例（基准 8px 对应预览小窗，成片基准 16px，这里 8px 对应约 0.33 缩放）
-          const padV = ps(16) * 0.5 // ≈ 8px in ~240px container
-          const padH = ps(24) * 0.5
-
+          const strokeW = c.stroke ? ps(c.stroke_width ?? 4) : 0
+          const strokeC = c.stroke_color || "#000000"
+          const shBlur = ps(c.shadow_blur ?? 4)
+          const shOffX = ps(c.shadow_offset_x ?? 2)
+          const shOffY = ps(c.shadow_offset_y ?? 2)
+          const shColor = c.shadow_color || "rgba(0,0,0,0.8)"
+          const lh = c.line_height ?? 1.2
+          const mTop = ps(c.margin_top ?? 24)
+          const bgPad = ps(c.bg_padding ?? 12)
+          const bgR = ps(c.bg_radius ?? 8)
+          const maxChars = c.max_chars_per_line ?? 0
+          const rawText = titleConfig.title || ""
+          const lines = (() => {
+            const manual = rawText
+              .split(/[/／]/)
+              .map((l) => l.trim())
+              .filter(Boolean)
+            if (!maxChars || maxChars <= 0) return manual
+            const out: string[] = []
+            manual.forEach((seg) => {
+              for (let i = 0; i < seg.length; i += maxChars) out.push(seg.slice(i, i + maxChars))
+            })
+            return out
+          })()
+          const padV = ps(16) * 0.5
+          const textShadow = titleConfig.shadow
+            ? `${shOffX}px ${shOffY}px ${shBlur}px ${shColor}`
+            : undefined
           const style: React.CSSProperties = {
             position: "absolute",
             color: titleConfig.color || "#ffffff",
-            fontFamily: getFontFamily(titleConfig.font || "思源黑体"),
+            fontFamily: getFontFamilyByKey(titleConfig.font || "source_sans_sc"),
             fontSize: `${fontSize}px`,
             fontWeight: titleConfig.bold ? 700 : 400,
             fontStyle: titleConfig.italic ? "italic" : "normal",
             textAlign: "center",
-            width: "90%",
-            lineHeight: 1.2,
-            padding: `${ps(4)}px ${padH}px`,
-            textShadow: titleConfig.shadow
-              ? `0 ${shadowOffsetY}px ${shadowBlur}px rgba(0,0,0,0.8), 0 0 ${ps(2)}px rgba(0,0,0,0.5)`
-              : undefined,
-            WebkitTextStroke: titleConfig.stroke ? `${strokeW}px #000` : undefined,
-            boxSizing: "border-box",
-            wordBreak: "break-word",
+            lineHeight: lh,
+            WebkitTextStroke:
+              titleConfig.stroke && strokeW > 0 ? `${strokeW}px ${strokeC}` : undefined,
+            paintOrder: "stroke fill",
+            textShadow,
             whiteSpace: "pre-wrap",
+            padding: c.bg_enabled ? `${bgPad}px ${bgPad}px` : 0,
+            background: c.bg_enabled ? c.bg_color || "rgba(0,0,0,0.5)" : "transparent",
+            borderRadius: c.bg_enabled ? `${bgR}px` : 0,
+            boxSizing: "border-box",
+            display: "inline-block",
+            maxWidth: "94%",
+          }
+          const wrap: React.CSSProperties = {
+            position: "absolute",
+            left: "50%",
+            width: "100%",
+            display: "flex",
+            justifyContent: "center",
+            pointerEvents: onTitlePositionChange ? "auto" : "none",
           }
-
           if (
             titleConfig.position === "custom" &&
             titleConfig.pos_x != null &&
             titleConfig.pos_y != null
           ) {
-            style.left = `${titleConfig.pos_x}%`
-            style.top = `${titleConfig.pos_y}%`
-            style.transform = "translateX(-50%) translateY(-50%)"
+            wrap.left = `${titleConfig.pos_x}%`
+            wrap.top = `${titleConfig.pos_y}%`
+            wrap.transform = "translate(-50%, -50%)"
           } else if (titleConfig.position === "top") {
-            style.left = "50%"
-            style.top = padV
-            style.transform = "translateX(-50%)"
+            wrap.top = `${padV + mTop}px`
+            wrap.transform = "translateX(-50%)"
           } else if (titleConfig.position === "bottom") {
-            style.left = "50%"
-            style.bottom = padV
-            style.transform = "translateX(-50%)"
+            wrap.bottom = `${padV}px`
+            wrap.transform = "translateX(-50%)"
           } else {
-            style.left = "50%"
-            style.top = "50%"
-            style.transform = "translateX(-50%) translateY(-50%)"
+            wrap.top = "50%"
+            wrap.transform = "translate(-50%, -50%)"
           }
-          return style
+          return { style, wrap, lines }
         })()
       : null
 
@@ -252,25 +286,23 @@ export function PanelLipsyncPreview({
           {isDone && lipsyncJob?.output_video_url ? (
             <div style={{ position: "relative", width: "100%", height: "100%" }}>
               <video src={lipsyncJob.output_video_url} controls />
-              {titleOverlayStyle && (
+              {titleOverlayData && (
                 <div
                   ref={titleDragRef}
                   style={{
-                    ...titleOverlayStyle,
+                    ...titleOverlayData.wrap,
                     cursor: onTitlePositionChange ? "grab" : "default",
-                    pointerEvents: onTitlePositionChange ? "auto" : "none",
                   }}
                   onPointerDown={handleTitlePointerDown}
                   onPointerMove={handleTitlePointerMove}
                   onPointerUp={handleTitlePointerUp}
                   onPointerCancel={handleTitlePointerUp}
                 >
-                  {titleConfig!.title.split(/[/／]/).map((part, i) => (
-                    <span key={i}>
-                      {i > 0 && <br />}
-                      {part}
-                    </span>
-                  ))}
+                  <div style={titleOverlayData.style}>
+                    {titleOverlayData.lines.map((part: string, i: number) => (
+                      <div key={i}>{part}</div>
+                    ))}
+                  </div>
                 </div>
               )}
             </div>

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

@@ -13,7 +13,8 @@ import TitleStylePanel from "@/pages/generate/components/title/TitleStylePanel"
 import TitleLibraryAutoComplete from "@/pages/generate/components/title/TitleLibraryAutoComplete"
 import type { TitleOption } from "@/pages/generate/components/title/TitleLibraryAutoComplete"
 import type { TitleSettings } from "@/pages/generate/types"
-import { POSITION_OPTIONS, FONT_OPTIONS, TITLE_PRESETS } from "@/pages/generate/constants"
+import { POSITION_OPTIONS } from "@/pages/generate/constants"
+import { FONT_OPTIONS, TITLE_PRESETS } from "@/components/title/constants"
 import type { AiAvatarTitleConfig } from "../types"
 // #1894: 标题数据源切换到文案库，取 script.title 作为候选
 import { getScripts } from "@/api/scripts"
@@ -49,9 +50,60 @@ const PanelTitleConfig: React.FC<PanelTitleConfigProps> = ({ titleConfig, onUpda
       .catch(() => setTitleOptions([]))
   }, [])
 
-  /** AiAvatarTitleConfig → TitleSettings（补齐 aiAutoSelect / 自由坐标字段） */
-  const titleSettings: TitleSettings = useMemo(
-    () => ({
+  /** AiAvatarTitleConfig (snake_case) → TitleSettings (camelCase) */
+  const titleSettings: TitleSettings = useMemo(() => {
+    const c = titleConfig as AiAvatarTitleConfig & {
+      stroke_width?: number
+      stroke_color?: string
+      shadow_offset_x?: number
+      shadow_offset_y?: number
+      shadow_blur?: number
+      shadow_color?: string
+      line_height?: number
+      margin_top?: number
+      max_chars_per_line?: number
+      bg_enabled?: boolean
+      bg_color?: string
+      bg_padding?: number
+      bg_radius?: number
+      cover_title_config?: {
+        title?: string
+        font?: string
+        size?: number
+        font_size?: number
+        color?: string
+        font_color?: string
+        bold?: boolean
+        italic?: boolean
+        position?: string
+        stroke?: { enabled: boolean; width?: number; color?: string } | boolean
+        stroke_width?: number
+        stroke_color?: string
+        shadow?:
+          | {
+              enabled: boolean
+              offset_x?: number
+              offset_y?: number
+              blur?: number
+              color?: string
+            }
+          | boolean
+        shadow_offset_x?: number
+        shadow_offset_y?: number
+        shadow_blur?: number
+        shadow_color?: string
+        background?: { enabled: boolean; color?: string; padding?: number; radius?: number }
+        bg_enabled?: boolean
+        bg_color?: string
+        bg_padding?: number
+        bg_radius?: number
+        line_height?: number
+        margin_top?: number
+        max_chars_per_line?: number
+      } | null
+      line_overrides?: unknown[]
+    }
+    return {
       aiAutoSelect: false,
       title: titleConfig.title,
       position: titleConfig.position,
@@ -64,24 +116,219 @@ const PanelTitleConfig: React.FC<PanelTitleConfigProps> = ({ titleConfig, onUpda
       color: titleConfig.color,
       posX: null,
       posY: null,
-    }),
-    [titleConfig],
-  )
+      lineHeight: c.line_height ?? 1.2,
+      marginTop: c.margin_top ?? 24,
+      maxCharsPerLine: c.max_chars_per_line ?? 0,
+      strokeWidth: c.stroke_width ?? 4,
+      strokeColor: c.stroke_color ?? "#000000",
+      shadowOffsetX: c.shadow_offset_x ?? 2,
+      shadowOffsetY: c.shadow_offset_y ?? 2,
+      shadowBlur: c.shadow_blur ?? 4,
+      shadowColor: c.shadow_color ?? "rgba(0,0,0,0.8)",
+      bgEnabled: !!c.bg_enabled,
+      bgColor: c.bg_color ?? "rgba(0,0,0,0.5)",
+      bgPadding: c.bg_padding ?? 12,
+      bgRadius: c.bg_radius ?? 8,
+      lineOverrides: Array.isArray(c.line_overrides) ? c.line_overrides : [],
+      coverTitle: (() => {
+        const ct = c.cover_title_config as
+          | null
+          | (AiAvatarTitleConfig & {
+              font_size?: number
+              font_color?: string
+              stroke?: { enabled?: boolean; width?: number; color?: string } | boolean
+              stroke_width?: number
+              stroke_color?: string
+              shadow?:
+                | {
+                    enabled?: boolean
+                    offset_x?: number
+                    offset_y?: number
+                    blur?: number
+                    color?: string
+                  }
+                | boolean
+              shadow_offset_x?: number
+              shadow_offset_y?: number
+              shadow_blur?: number
+              shadow_color?: string
+              background?: { enabled?: boolean; color?: string; padding?: number; radius?: number }
+              bg_enabled?: boolean
+              bg_color?: string
+              bg_padding?: number
+              bg_radius?: number
+            })
+        if (!ct) return null
+        const ctStroke = ct.stroke as
+          { enabled?: boolean; width?: number; color?: string } | boolean | undefined
+        const ctShadow = ct.shadow as
+          | {
+              enabled?: boolean
+              offset_x?: number
+              offset_y?: number
+              blur?: number
+              color?: string
+            }
+          | boolean
+          | undefined
+        const ctBg = ct.background as
+          { enabled?: boolean; color?: string; padding?: number; radius?: number } | undefined
+        return {
+          title: ct.title,
+          font: ct.font,
+          size: ct.font_size ?? ct.size,
+          color: ct.font_color ?? ct.color,
+          bold: ct.bold,
+          italic: ct.italic,
+          position: ct.position,
+          stroke:
+            typeof ctStroke === "object" && ctStroke ? ctStroke.enabled !== false : !!ctStroke,
+          strokeWidth:
+            (typeof ctStroke === "object" && ctStroke ? ctStroke.width : undefined) ??
+            ct.stroke_width ??
+            4,
+          strokeColor:
+            (typeof ctStroke === "object" && ctStroke ? ctStroke.color : undefined) ??
+            ct.stroke_color ??
+            "#000000",
+          shadow:
+            typeof ctShadow === "object" && ctShadow ? ctShadow.enabled !== false : !!ctShadow,
+          shadowOffsetX:
+            (typeof ctShadow === "object" && ctShadow ? ctShadow.offset_x : undefined) ??
+            ct.shadow_offset_x ??
+            2,
+          shadowOffsetY:
+            (typeof ctShadow === "object" && ctShadow ? ctShadow.offset_y : undefined) ??
+            ct.shadow_offset_y ??
+            2,
+          shadowBlur:
+            (typeof ctShadow === "object" && ctShadow ? ctShadow.blur : undefined) ??
+            ct.shadow_blur ??
+            4,
+          shadowColor:
+            (typeof ctShadow === "object" && ctShadow ? ctShadow.color : undefined) ??
+            ct.shadow_color ??
+            "rgba(0,0,0,0.8)",
+          bgEnabled: ctBg?.enabled ?? !!ct.bg_enabled,
+          bgColor: ctBg?.color ?? ct.bg_color ?? "rgba(0,0,0,0.5)",
+          bgPadding: ctBg?.padding ?? ct.bg_padding ?? 12,
+          bgRadius: ctBg?.radius ?? ct.bg_radius ?? 8,
+        }
+      })(),
+    }
+  }, [titleConfig])
 
-  /** 应用预设：与智能剪辑一致，只覆盖 color/bold/italic/stroke/shadow，不改变字号 */
+  /** 应用预设：覆盖新细粒度字段（颜色/描边/阴影/字号/字体等） */
   const handleApplyPreset = (presetKey: string) => {
     const preset = TITLE_PRESETS.find((p) => p.key === presetKey)
     if (!preset) return
     setActivePreset(presetKey)
+    const st = preset.style || {}
     onUpdate({
-      color: preset.style.color,
-      bold: preset.style.bold,
-      italic: preset.style.italic,
-      stroke: preset.style.stroke,
-      shadow: preset.style.shadow,
+      font: st.font,
+      size: st.size,
+      color: st.color,
+      bold: st.bold,
+      italic: st.italic,
+      stroke: st.stroke,
+      stroke_width: st.stroke_width,
+      stroke_color: st.stroke_color,
+      shadow: st.shadow,
+      shadow_offset_x: st.shadow_offset_x,
+      shadow_offset_y: st.shadow_offset_y,
+      shadow_blur: st.shadow_blur,
+      shadow_color: st.shadow_color,
+      bg_enabled: st.bg_enabled,
+      bg_color: st.bg_color,
+      bg_padding: st.bg_padding,
+      bg_radius: st.bg_radius,
+      line_overrides: [],
+      cover_title_config: null,
     })
   }
 
+  /** 字段 patch 透传：TitleStylePanel 的 onUpdateStyle（camelCase → snake_case） */
+  const handleUpdateStyle = (patch: Partial<TitleSettings>) => {
+    const snake: Record<string, unknown> = {}
+    const map: Record<string, string> = {
+      lineHeight: "line_height",
+      marginTop: "margin_top",
+      maxCharsPerLine: "max_chars_per_line",
+      strokeWidth: "stroke_width",
+      strokeColor: "stroke_color",
+      shadowOffsetX: "shadow_offset_x",
+      shadowOffsetY: "shadow_offset_y",
+      shadowBlur: "shadow_blur",
+      shadowColor: "shadow_color",
+      bgEnabled: "bg_enabled",
+      bgColor: "bg_color",
+      bgPadding: "bg_padding",
+      bgRadius: "bg_radius",
+      lineOverrides: "line_overrides",
+      coverTitle: "cover_title_config",
+    }
+    Object.entries(patch).forEach(([k, v]) => {
+      if (k === "coverTitle" && v && typeof v === "object") {
+        const ct = v as {
+          title?: string
+          font?: string
+          size?: number
+          color?: string
+          bold?: boolean
+          italic?: boolean
+          position?: string
+          stroke?: boolean
+          strokeWidth?: number
+          strokeColor?: string
+          shadow?: boolean
+          shadowOffsetX?: number
+          shadowOffsetY?: number
+          shadowBlur?: number
+          shadowColor?: string
+          bgEnabled?: boolean
+          bgColor?: string
+          bgPadding?: number
+          bgRadius?: number
+          lineHeight?: number
+          marginTop?: number
+          maxCharsPerLine?: number
+        }
+        snake.cover_title_config = {
+          title: ct.title,
+          font: ct.font,
+          font_size: ct.size,
+          font_color: ct.color,
+          bold: ct.bold,
+          italic: ct.italic,
+          position: ct.position,
+          stroke: ct.stroke
+            ? { enabled: true, width: ct.strokeWidth ?? 4, color: ct.strokeColor ?? "#000" }
+            : { enabled: false },
+          shadow: ct.shadow
+            ? {
+                enabled: true,
+                offset_x: ct.shadowOffsetX ?? 2,
+                offset_y: ct.shadowOffsetY ?? 2,
+                blur: ct.shadowBlur ?? 4,
+                color: ct.shadowColor ?? "rgba(0,0,0,0.8)",
+              }
+            : { enabled: false },
+          background: ct.bgEnabled
+            ? { enabled: true, color: ct.bgColor, padding: ct.bgPadding, radius: ct.bgRadius }
+            : { enabled: false },
+          line_height: ct.lineHeight,
+          margin_top: ct.marginTop,
+          max_chars_per_line: ct.maxCharsPerLine,
+        }
+      } else if (map[k]) {
+        snake[map[k]] = v
+      } else {
+        snake[k] = v
+      }
+    })
+    onUpdate(snake)
+  }
+
   return (
     <div className="aa-title-config">
       {/* 主标题输入 — TextArea 多行 + 标题库选择 */}
@@ -125,8 +372,13 @@ const PanelTitleConfig: React.FC<PanelTitleConfigProps> = ({ titleConfig, onUpda
         onToggleStroke={() => onUpdate({ stroke: !titleConfig.stroke })}
         onToggleShadow={() => onUpdate({ shadow: !titleConfig.shadow })}
         onApplyPreset={handleApplyPreset}
+        onUpdateStyle={handleUpdateStyle}
+        showCoverToggle
+        previewWidth={280}
         activePreset={activePreset}
-        titlePresets={TITLE_PRESETS}
+        titlePresets={
+          TITLE_PRESETS as unknown as React.ComponentProps<typeof TitleStylePanel>["titlePresets"]
+        }
         POSITION_OPTIONS={POSITION_OPTIONS}
         FONT_OPTIONS={FONT_OPTIONS}
       />

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

@@ -1,20 +1,17 @@
 /**
  * AI数字人 — 配音库面板（面板3）
- * 音色来源切换（系统预设 / 我的音色）、音色选择与试听、情绪/语速/语言参数
+ * 音色来源切换（系统预设 / 我的音色）、音色选择与试听、风格/语速/语言参数
  */
 import { useEffect, useRef, useState } from "react"
 import { message } from "antd"
 import { fetchVoices } from "@/api/voices/voices"
 import { previewTts } from "@/api/tts"
-import { normalizeEmotion } from "../utils/contract"
 import TtsStyleSelector from "@/components/voice/TtsStyleSelector"
 import type { TtsStyle } from "@/api/tts/styles"
 import type { UnifiedVoiceItem } from "@/api/voices/types"
 import {
   type VoiceSource,
-  type VoiceEmotion,
   type VoiceLanguage,
-  VOICE_EMOTION_OPTIONS,
   PRESET_VOICE_LANGUAGE_OPTIONS,
   CLONE_VOICE_LANGUAGE_OPTIONS,
 } from "../types"
@@ -24,8 +21,6 @@ interface PanelVoiceSelectorProps {
   onVoiceSourceChange: (source: VoiceSource) => void
   selectedVoice: UnifiedVoiceItem | null
   onSelectVoice: (voice: UnifiedVoiceItem) => void
-  emotion: VoiceEmotion
-  onEmotionChange: (e: VoiceEmotion) => void
   style: TtsStyle
   onStyleChange: (s: TtsStyle) => void
   speed: number
@@ -39,8 +34,6 @@ export function PanelVoiceSelector({
   onVoiceSourceChange,
   selectedVoice,
   onSelectVoice,
-  emotion,
-  onEmotionChange,
   style,
   onStyleChange,
   speed,
@@ -152,14 +145,12 @@ export function PanelVoiceSelector({
         return
       }
       const targetId = voice.voice_clone_profile_id || voice.id
-      // DEBUG: 打印请求参数，帮助定位 /tts/preview 失败原因
       setPreviewingId(voice.id)
       try {
         const res = await previewTts({
           text: VOICE_PREVIEW_TEXT,
           voice_id: targetId,
           speed: speed, // 透传用户选择的语速（#1822）
-          emotion: normalizeEmotion(emotion), // 情绪中文→英文枚举
           style,
         })
         if (!res.audio_url) {
@@ -171,7 +162,6 @@ export function PanelVoiceSelector({
         playAudioUrl(voice.id, res.audio_url)
       } catch (err) {
         setPreviewingId(null)
-        // DEBUG: 打印详细错误信息
         console.error("[AI数字人-克隆试听] previewTts 失败:", {
           status: (err as { response?: { status?: number } })?.response?.status,
           data: (err as { response?: { data?: unknown } })?.response?.data,
@@ -280,23 +270,6 @@ export function PanelVoiceSelector({
       {/* 配音参数 */}
       <div className="aa-voice-params">
         <div className="aa-voice-params__row">
-          <div className="aa-voice-params__field">
-            <label className="aa-label" htmlFor="aa-voice-emotion">
-              情绪
-            </label>
-            <select
-              id="aa-voice-emotion"
-              className="aa-select"
-              value={emotion}
-              onChange={(e) => onEmotionChange(e.target.value as VoiceEmotion)}
-            >
-              {VOICE_EMOTION_OPTIONS.map((opt) => (
-                <option key={opt.value} value={opt.value}>
-                  {opt.label}
-                </option>
-              ))}
-            </select>
-          </div>
           <div className="aa-voice-params__field">
             <label className="aa-label" htmlFor="aa-voice-language">
               语言

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

@@ -6,7 +6,6 @@ import type { AssetItem } from "@/api/assets"
 import type { UnifiedVoiceItem } from "@/api/voices/types"
 import {
   type VoiceSource,
-  type VoiceEmotion,
   type VoiceLanguage,
   type Script,
   type LipsyncJob,
@@ -35,7 +34,6 @@ export function useAiAvatar() {
   /* ── 面板2：配音库 ── */
   const [voiceSource, setVoiceSource] = useState<VoiceSource>("preset")
   const [selectedVoice, setSelectedVoice] = useState<UnifiedVoiceItem | null>(null)
-  const [emotion, setEmotion] = useState<VoiceEmotion>("neutral")
   const [style, setStyle] = useState<TtsStyle>(DEFAULT_TTS_STYLE)
   const [speed, setSpeed] = useState(1.0)
   const [language, setLanguage] = useState<VoiceLanguage>("zh")
@@ -115,8 +113,6 @@ export function useAiAvatar() {
     setVoiceSource,
     selectedVoice,
     setSelectedVoice,
-    emotion,
-    setEmotion,
     style,
     setStyle,
     speed,

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

@@ -100,7 +100,7 @@ export interface BRollSegment {
   pip_scale: number
 }
 
-/* ── 标题配置 ── */
+/* ── 标题配置（#2001 升级：细粒度描边/阴影/背景/排版/逐行/封面独立标题） ── */
 export interface AiAvatarTitleConfig {
   title: string
   position: string
@@ -115,6 +115,42 @@ export interface AiAvatarTitleConfig {
   /** 自定义位置坐标（position=custom 时生效，百分比 0-100） */
   pos_x?: number
   pos_y?: number
+
+  /* ── 排版 ── */
+  line_height: number
+  margin_top: number
+  max_chars_per_line: number
+
+  /* ── 描边参数化 ── */
+  stroke_width: number
+  stroke_color: string
+
+  /* ── 阴影参数化 ── */
+  shadow_offset_x: number
+  shadow_offset_y: number
+  shadow_blur: number
+  shadow_color: string
+
+  /* ── 背景色块 ── */
+  bg_enabled: boolean
+  bg_color: string
+  bg_padding: number
+  bg_radius: number
+
+  /* ── 逐行覆盖 ── */
+  line_overrides: Array<{
+    line_index: number
+    text?: string
+    size?: number
+    color?: string
+    bold?: boolean
+    italic?: boolean
+    stroke?: boolean
+    highlights?: Array<{ word: string; color?: string; bold?: boolean; scale?: number }>
+  }>
+
+  /* ── 封面独立标题（null=沿用主标题） ── */
+  cover_title_config: null | Partial<AiAvatarTitleConfig>
 }
 
 /* ── 封面配置 ── */
@@ -149,12 +185,27 @@ export const DEFAULT_TITLE_CONFIG: AiAvatarTitleConfig = {
   size: 48,
   bold: true,
   italic: false,
-  stroke: false,
+  stroke: true,
   shadow: false,
   color: "#ffffff",
   auto_subtitle: true,
   pos_x: undefined,
   pos_y: undefined,
+  line_height: 1.2,
+  margin_top: 24,
+  max_chars_per_line: 0,
+  stroke_width: 4,
+  stroke_color: "#000000",
+  shadow_offset_x: 2,
+  shadow_offset_y: 2,
+  shadow_blur: 4,
+  shadow_color: "rgba(0,0,0,0.8)",
+  bg_enabled: false,
+  bg_color: "rgba(0,0,0,0.5)",
+  bg_padding: 12,
+  bg_radius: 8,
+  line_overrides: [],
+  cover_title_config: null,
 }
 
 export const DEFAULT_COVER_CONFIG: AiAvatarCoverConfig = {

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

@@ -67,6 +67,37 @@ export function buildTitleConfigPayload(
   const text = (cfg.title || "").trim()
   if (!text) return {}
   const position = cfg.position || "bottom"
+  const anyCfg = cfg as AiAvatarTitleConfig & {
+    stroke_width?: number
+    stroke_color?: string
+    shadow_offset_x?: number
+    shadow_offset_y?: number
+    shadow_blur?: number
+    shadow_color?: string
+    line_height?: number
+    margin_top?: number
+    max_chars_per_line?: number
+    bg_enabled?: boolean
+    bg_color?: string
+    bg_padding?: number
+    bg_radius?: number
+    line_overrides?: unknown[]
+    cover_title_config?: Record<string, unknown> | null
+  }
+  const strokeWidth = anyCfg.stroke_width != null ? anyCfg.stroke_width : 4
+  const strokeColor = anyCfg.stroke_color || "#000000"
+  const shadowOffsetX = anyCfg.shadow_offset_x != null ? anyCfg.shadow_offset_x : 2
+  const shadowOffsetY = anyCfg.shadow_offset_y != null ? anyCfg.shadow_offset_y : 2
+  const shadowBlur = anyCfg.shadow_blur != null ? anyCfg.shadow_blur : 4
+  const shadowColor = anyCfg.shadow_color || "rgba(0,0,0,0.8)"
+  const lineHeight = anyCfg.line_height != null ? anyCfg.line_height : 1.2
+  const marginTop = anyCfg.margin_top != null ? anyCfg.margin_top : 24
+  const maxCharsPerLine = anyCfg.max_chars_per_line ?? 0
+  const bgEnabled = !!anyCfg.bg_enabled
+  const bgColor = anyCfg.bg_color || "rgba(0,0,0,0.5)"
+  const bgPadding = anyCfg.bg_padding != null ? anyCfg.bg_padding : 12
+  const bgRadius = anyCfg.bg_radius != null ? anyCfg.bg_radius : 8
+
   const payload: Record<string, unknown> = {
     text,
     enabled: true,
@@ -75,16 +106,68 @@ export function buildTitleConfigPayload(
     font_color: cfg.color || "#ffffff",
     position,
     bold: !!cfg.bold,
-    stroke: cfg.stroke ? { enabled: true, width: 2, color: "#000000" } : { enabled: false },
-    shadow: cfg.shadow
-      ? { enabled: true, color: "#000000", offset_x: 2, offset_y: 2 }
+    italic: !!cfg.italic,
+    stroke: cfg.stroke
+      ? { enabled: true, width: strokeWidth, color: strokeColor }
       : { enabled: false },
+    shadow: cfg.shadow
+      ? {
+          enabled: true,
+          color: shadowColor,
+          offset_x: shadowOffsetX,
+          offset_y: shadowOffsetY,
+          blur: shadowBlur,
+        }
+      : { enabled: false },
+    line_height: lineHeight,
+    margin_top: marginTop,
+    max_chars_per_line: maxCharsPerLine,
+    background: bgEnabled
+      ? { enabled: true, color: bgColor, padding: bgPadding, radius: bgRadius }
+      : { enabled: false },
+    line_overrides: Array.isArray(anyCfg.line_overrides) ? anyCfg.line_overrides : [],
   }
   // 自定义坐标（custom 位置）
   if (position === "custom" && typeof cfg.pos_x === "number" && typeof cfg.pos_y === "number") {
     payload.pos_x = cfg.pos_x
     payload.pos_y = cfg.pos_y
   }
+  // 封面独立标题配置
+  if (anyCfg.cover_title_config) {
+    const ctc = anyCfg.cover_title_config
+    payload.cover_title_config = {
+      title: ctc.title,
+      font: ctc.font,
+      font_size: ctc.size,
+      font_color: ctc.color,
+      position: ctc.position,
+      bold: ctc.bold,
+      italic: ctc.italic,
+      stroke: ctc.stroke
+        ? { enabled: true, width: ctc.stroke_width ?? 4, color: ctc.stroke_color ?? "#000000" }
+        : { enabled: false },
+      shadow: ctc.shadow
+        ? {
+            enabled: true,
+            color: ctc.shadow_color ?? shadowColor,
+            offset_x: ctc.shadow_offset_x ?? 2,
+            offset_y: ctc.shadow_offset_y ?? 2,
+            blur: ctc.shadow_blur ?? 4,
+          }
+        : { enabled: false },
+      line_height: ctc.line_height ?? lineHeight,
+      margin_top: ctc.margin_top ?? marginTop,
+      max_chars_per_line: ctc.max_chars_per_line ?? maxCharsPerLine,
+      background: ctc.bg_enabled
+        ? {
+            enabled: true,
+            color: ctc.bg_color ?? bgColor,
+            padding: ctc.bg_padding ?? bgPadding,
+            radius: ctc.bg_radius ?? bgRadius,
+          }
+        : { enabled: false },
+    }
+  }
   // 前端 Canvas 渲染好的 PNG dataURL（所见即所得，后端优先 overlay 此图片图层）
   if (titleImageDataUrl) {
     payload.title_image_dataurl = titleImageDataUrl

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

@@ -9,37 +9,78 @@
  *       按 videoWidth / 720 得到 scale，所有长度类参数乘以 scale，
  *       保证 1080p / 4K 成片里标题视觉大小与预览一致。
  */
+import { getFontFamily } from "@/components/title/constants"
 import type { AiAvatarTitleConfig } from "../types"
 
 export interface RenderTitlePngOptions {
-  /** 标题配置 */
   titleConfig: AiAvatarTitleConfig
-  /** 视频宽度（像素），默认 720 */
   videoWidth?: number
-  /** 视频高度（像素），默认 1280 */
   videoHeight?: number
+  useCoverTitle?: boolean
 }
 
-/**
- * 将标题渲染为透明背景 PNG 的 dataURL（data:image/png;base64,...）
- * Canvas 尺寸与视频一致，保证叠加时 1:1 像素对齐。
- *
- * 标题为空时返回 null。
- */
-export function renderTitleToPngDataUrl(opts: RenderTitlePngOptions): string | null {
-  const { titleConfig, videoWidth = 720, videoHeight = 1280 } = opts
-  if (!titleConfig) return null
-  const rawTitle = (titleConfig.title || "").trim()
-  if (!rawTitle) return null
-
-  // 按 / 或 ／ 分割为多行
-  const lines = rawTitle
+function autoWrapLines(rawTitle: string, maxCharsPerLine: number): string[] {
+  const manual = rawTitle
     .split(/[/／]/)
     .map((l) => l.trim())
     .filter((l) => l.length > 0)
+  if (!maxCharsPerLine || maxCharsPerLine <= 0) return manual
+  const out: string[] = []
+  manual.forEach((seg) => {
+    for (let i = 0; i < seg.length; i += maxCharsPerLine) {
+      out.push(seg.slice(i, i + maxCharsPerLine))
+    }
+  })
+  return out
+}
+
+export function renderTitleToPngDataUrl(opts: RenderTitlePngOptions): string | null {
+  const { titleConfig, videoWidth = 720, videoHeight = 1280, useCoverTitle } = opts
+  if (!titleConfig) return null
+
+  type TitleCfgExt = AiAvatarTitleConfig & {
+    stroke_width?: number
+    stroke_color?: string
+    shadow_offset_x?: number
+    shadow_offset_y?: number
+    shadow_blur?: number
+    shadow_color?: string
+    line_height?: number
+    margin_top?: number
+    max_chars_per_line?: number
+    bg_enabled?: boolean
+    bg_color?: string
+    bg_padding?: number
+    bg_radius?: number
+    line_overrides?: Array<{
+      line_index: number
+      text?: string
+      size?: number
+      color?: string
+      bold?: boolean
+      italic?: boolean
+      stroke?: boolean
+      highlights?: Array<{ word: string; color?: string; bold?: boolean; scale?: number }>
+    }>
+    cover_title_config?: Partial<AiAvatarTitleConfig> | null
+    pos_x?: number
+    pos_y?: number
+  }
+  const cfg: TitleCfgExt =
+    useCoverTitle && titleConfig.cover_title_config
+      ? ({
+          ...(titleConfig as TitleCfgExt),
+          ...(titleConfig.cover_title_config as object),
+        } as TitleCfgExt)
+      : (titleConfig as TitleCfgExt)
+
+  const rawTitle = (cfg.title || "").trim()
+  if (!rawTitle) return null
+
+  const maxCharsPerLine = cfg.max_chars_per_line ?? 0
+  const lines = autoWrapLines(rawTitle, maxCharsPerLine)
   if (lines.length === 0) return null
 
-  // 分辨率缩放系数：基准 720p，所有长度类参数乘以 scale
   const scale = videoWidth / 720
   const r = (v: number) => Math.round(v * scale)
 
@@ -49,86 +90,183 @@ export function renderTitleToPngDataUrl(opts: RenderTitlePngOptions): string | n
   const ctx = canvas.getContext("2d")
   if (!ctx) return null
 
-  const baseSize = Math.max(12, Math.round(titleConfig.size || 48))
+  const baseSize = Math.max(12, Math.round(cfg.size || 48))
   const size = r(baseSize)
-  const bold = !!titleConfig.bold
-  const italic = !!titleConfig.italic
-  const color = titleConfig.color || "#ffffff"
-  const stroke = !!titleConfig.stroke
-  const shadow = !!titleConfig.shadow
+  const bold = !!cfg.bold
+  const italic = !!cfg.italic
+  const color = cfg.color || "#ffffff"
+  const stroke = !!cfg.stroke
+  const shadow = !!cfg.shadow
 
-  // 字体族 fallback 链：优先中文字体
-  const fontFamily =
-    '"Noto Sans CJK SC","Source Han Sans CN","PingFang SC","Microsoft YaHei",sans-serif'
-  const fontParts: string[] = []
-  if (italic) fontParts.push("italic")
-  if (bold) fontParts.push("bold")
-  fontParts.push(`${size}px`, fontFamily)
-  ctx.font = fontParts.join(" ")
+  const strokeWidthBase = cfg.stroke_width != null ? cfg.stroke_width : 4
+  const strokeColor = cfg.stroke_color || "#000000"
+  const shadowOffsetXBase = cfg.shadow_offset_x != null ? cfg.shadow_offset_x : 2
+  const shadowOffsetYBase = cfg.shadow_offset_y != null ? cfg.shadow_offset_y : 2
+  const shadowBlurBase = cfg.shadow_blur != null ? cfg.shadow_blur : 4
+  const shadowColor = cfg.shadow_color || "rgba(0,0,0,0.8)"
+  const lineHeightScale = cfg.line_height != null ? cfg.line_height : 1.2
+  const marginTopBase = cfg.margin_top != null ? cfg.margin_top : 24
+  const bgEnabled = !!cfg.bg_enabled
+  const bgColor = cfg.bg_color || "rgba(0,0,0,0.5)"
+  const bgPaddingBase = cfg.bg_padding != null ? cfg.bg_padding : 12
+  const bgRadiusBase = cfg.bg_radius != null ? cfg.bg_radius : 8
+
+  const fontKey = cfg.font || "思源黑体"
+  const fontFamily = getFontFamily(fontKey)
+  const setFont = (sz: number, bd: boolean, it: boolean) => {
+    const parts: string[] = []
+    if (it) parts.push("italic")
+    if (bd) parts.push("bold")
+    parts.push(`${sz}px`, fontFamily)
+    ctx.font = parts.join(" ")
+  }
+  setFont(size, bold, italic)
   ctx.fillStyle = color
   ctx.textAlign = "center"
   ctx.textBaseline = "middle"
 
-  // 阴影（shadow=true 时开启）——按 scale 缩放
-  if (shadow) {
-    ctx.shadowColor = "rgba(0,0,0,0.8)"
-    ctx.shadowBlur = r(4)
-    ctx.shadowOffsetX = 0
-    ctx.shadowOffsetY = r(2)
-  }
+  const lineGap = size * lineHeightScale
+  const totalTextH = lines.length * lineGap - (lineGap - size)
+  let maxLineW = 0
+  lines.forEach((l: string) => {
+    const m = ctx.measureText(l).width
+    if (m > maxLineW) maxLineW = m
+  })
 
-  // 位置计算：与 PanelLipsyncPreview 的 CSS 对齐（按 scale 缩放 PAD）
   const PAD = r(16)
   let centerX = videoWidth / 2
-  const position = titleConfig.position || "bottom"
-  const lineGap = size * 1.2
-  const totalTextH = lines.length * lineGap - (lineGap - size) // 所有行的总高度
-  // 文本块顶部 y（textBaseline=middle 时首行基线）
+  const position = cfg.position || "bottom"
   let firstLineY: number
-  if (
-    position === "custom" &&
-    typeof titleConfig.pos_x === "number" &&
-    typeof titleConfig.pos_y === "number"
-  ) {
-    centerX = (Math.max(0, Math.min(100, titleConfig.pos_x)) / 100) * videoWidth
-    const centerY = (Math.max(0, Math.min(100, titleConfig.pos_y)) / 100) * videoHeight
+  if (position === "custom" && typeof cfg.pos_x === "number" && typeof cfg.pos_y === "number") {
+    centerX = (Math.max(0, Math.min(100, cfg.pos_x)) / 100) * videoWidth
+    const centerY = (Math.max(0, Math.min(100, cfg.pos_y)) / 100) * videoHeight
     firstLineY = centerY - totalTextH / 2 + size / 2
   } else if (position === "top") {
-    // 顶部：y = size/2 + PAD
-    firstLineY = size / 2 + PAD
+    firstLineY = size / 2 + PAD + r(marginTopBase)
   } else if (position === "center") {
     firstLineY = videoHeight / 2 - totalTextH / 2 + size / 2
   } else {
-    // bottom（默认）
     firstLineY = videoHeight - totalTextH - PAD + size / 2
   }
 
-  // 描边参数：描边 lineWidth 按 scale 缩放（基准 size * 0.06，最小 2px @720p）
-  const doStroke = stroke
-  const strokeWidth = Math.max(r(2), Math.round(size * 0.06))
-  // 逐行绘制
-  lines.forEach((line, idx) => {
+  if (shadow) {
+    ctx.shadowColor = shadowColor
+    ctx.shadowBlur = r(shadowBlurBase)
+    ctx.shadowOffsetX = r(shadowOffsetXBase)
+    ctx.shadowOffsetY = r(shadowOffsetYBase)
+  } else {
+    ctx.shadowColor = "rgba(0,0,0,0)"
+    ctx.shadowBlur = 0
+    ctx.shadowOffsetX = 0
+    ctx.shadowOffsetY = 0
+  }
+
+  const bgPad = r(bgPaddingBase)
+  const bgR = r(bgRadiusBase)
+  const bgW = maxLineW + bgPad * 2
+  const bgH = totalTextH + bgPad * 2
+  const bgX = centerX - bgW / 2
+  const bgY = firstLineY - size / 2 - bgPad
+
+  if (bgEnabled) {
+    ctx.save()
+    ctx.shadowColor = "rgba(0,0,0,0)"
+    ctx.shadowBlur = 0
+    ctx.shadowOffsetX = 0
+    ctx.shadowOffsetY = 0
+    ctx.fillStyle = bgColor
+    if (
+      bgR > 0 &&
+      (
+        ctx as CanvasRenderingContext2D & {
+          roundRect?: (x: number, y: number, w: number, h: number, r: number) => void
+        }
+      ).roundRect
+    ) {
+      ;(
+        ctx as CanvasRenderingContext2D & {
+          roundRect?: (x: number, y: number, w: number, h: number, r: number) => void
+        }
+      ).roundRect(bgX, bgY, bgW, bgH, bgR)
+      ctx.fill()
+    } else {
+      ctx.fillRect(bgX, bgY, bgW, bgH)
+    }
+    ctx.restore()
+  }
+
+  const sw = stroke ? Math.max(r(1), r(strokeWidthBase)) : 0
+  const lineOverrides = cfg.line_overrides || []
+  lines.forEach((line: string, idx: number) => {
     const y = firstLineY + idx * lineGap
-    if (doStroke) {
-      const prevShadowColor = ctx.shadowColor
-      const prevShadowBlur = ctx.shadowBlur
-      // 描边不要带阴影（避免黑色描边发虚）
+    const override = lineOverrides.find((lo) => lo.line_index === idx)
+    const lineSize = override?.size ? r(Math.max(12, Math.round(override.size))) : size
+    const lineColor = override?.color || color
+    const lineBold = override?.bold != null ? !!override.bold : bold
+    const lineItalic = override?.italic != null ? !!override.italic : italic
+    const lineStroke = override?.stroke != null ? !!override.stroke : stroke
+
+    setFont(lineSize, lineBold, lineItalic)
+    ctx.fillStyle = lineColor
+
+    if (shadow) {
+      ctx.shadowColor = shadowColor
+      ctx.shadowBlur = r(shadowBlurBase)
+      ctx.shadowOffsetX = r(shadowOffsetXBase)
+      ctx.shadowOffsetY = r(shadowOffsetYBase)
+    } else {
       ctx.shadowColor = "rgba(0,0,0,0)"
       ctx.shadowBlur = 0
-      ctx.lineWidth = strokeWidth
-      ctx.strokeStyle = "#000000"
+      ctx.shadowOffsetX = 0
+      ctx.shadowOffsetY = 0
+    }
+
+    const lineSw = override?.size
+      ? Math.max(r(1), Math.round(lineSize * (strokeWidthBase / baseSize)))
+      : sw
+
+    if (lineStroke && lineSw > 0) {
+      ctx.save()
+      ctx.shadowColor = "rgba(0,0,0,0)"
+      ctx.shadowBlur = 0
+      ctx.shadowOffsetX = 0
+      ctx.shadowOffsetY = 0
+      ctx.lineWidth = lineSw
+      ctx.strokeStyle = strokeColor
       ctx.lineJoin = "round"
       ctx.strokeText(line, centerX, y)
-      // 恢复阴影
-      if (shadow) {
-        ctx.shadowColor = "rgba(0,0,0,0.8)"
-        ctx.shadowBlur = r(4)
-      } else {
-        ctx.shadowColor = prevShadowColor
-        ctx.shadowBlur = prevShadowBlur
-      }
+      ctx.restore()
     }
+
     ctx.fillText(line, centerX, y)
+
+    if (override?.highlights?.length) {
+      const fullW = ctx.measureText(line).width
+      const charW = line.length > 0 ? fullW / line.length : lineSize
+      override.highlights.forEach((hl) => {
+        if (!hl.word) return
+        const pos = line.indexOf(hl.word)
+        if (pos < 0) return
+        const hlX = centerX - fullW / 2 + pos * charW + (charW * hl.word.length) / 2
+        const hlColor = hl.color || "#ffd700"
+        const hlScale = hl.scale || 1
+        const hlSize = lineSize * hlScale
+        const hlBold = hl.bold != null ? !!hl.bold : true
+        ctx.save()
+        ctx.shadowColor = "rgba(0,0,0,0)"
+        ctx.shadowBlur = 0
+        setFont(hlSize, hlBold, lineItalic)
+        ctx.fillStyle = hlColor
+        if (lineStroke && lineSw > 0) {
+          ctx.lineWidth = Math.max(r(1), Math.round(hlSize * (strokeWidthBase / baseSize)))
+          ctx.strokeStyle = strokeColor
+          ctx.lineJoin = "round"
+          ctx.strokeText(hl.word, hlX, y)
+        }
+        ctx.fillText(hl.word, hlX, y)
+        ctx.restore()
+      })
+    }
   })
 
   try {
@@ -138,9 +276,6 @@ export function renderTitleToPngDataUrl(opts: RenderTitlePngOptions): string | n
   }
 }
 
-/**
- * 获取视频真实分辨率（HTMLVideoElement + loadedmetadata，超时 3 秒兜底 720×1280）。
- */
 export function getVideoResolution(
   videoUrl: string,
   timeoutMs = 3000,

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

@@ -640,6 +640,7 @@ const GeneratePage: React.FC = () => {
             onToggleStroke={styleUpdaters.toggleStroke}
             onToggleShadow={styleUpdaters.toggleShadow}
             onApplyPreset={styleUpdaters.applyPreset}
+            onUpdateStyle={styleUpdaters.updateStyle}
             activePreset={styleUpdaters.activePreset}
             titlePresets={styleUpdaters.titlePresets}
             bgm={bgm}

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

@@ -51,8 +51,14 @@ export interface GenerateStepContentProps {
   onToggleStroke: () => void
   onToggleShadow: () => void
   onApplyPreset: (presetKey: string) => void
+  onUpdateStyle?: (patch: Partial<TitleSettings>) => void
   activePreset: string | null
-  titlePresets: { key: string; label: string; previewStyle: React.CSSProperties }[]
+  titlePresets: Array<{
+    key: string
+    label: string
+    emoji?: string
+    style: Record<string, unknown>
+  }>
   /* ── 封面 ── */
   coverSettings: CoverConfig
   onCoverSettingsChange: (settings: CoverConfig) => void
@@ -119,6 +125,7 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     onToggleStroke,
     onToggleShadow,
     onApplyPreset,
+    onUpdateStyle,
     activePreset,
     titlePresets,
     coverSettings,
@@ -191,6 +198,7 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           onToggleStroke={onToggleStroke}
           onToggleShadow={onToggleShadow}
           onApplyPreset={onApplyPreset}
+          onUpdateStyle={onUpdateStyle}
           activePreset={activePreset}
           titlePresets={titlePresets}
           previewCount={previewCount}

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

@@ -12,7 +12,8 @@ import React, { useMemo, useState } from "react"
 import { Input, message } from "antd"
 import { LoadingOutlined } from "@ant-design/icons"
 import type { TitleSettings } from "../types"
-import { POSITION_OPTIONS, FONT_OPTIONS } from "../constants"
+import { POSITION_OPTIONS } from "../constants"
+import { FONT_OPTIONS } from "@/components/title/constants"
 import { useStep4Title } from "../hooks/useStep4Title"
 import AiTitleGenerator from "./title/AiTitleGenerator"
 import TitleLibraryAutoComplete from "./title/TitleLibraryAutoComplete"
@@ -33,8 +34,15 @@ interface Step4TitleSettingsProps {
   onToggleStroke: () => void
   onToggleShadow: () => void
   onApplyPreset: (presetKey: string) => void
+  onUpdateStyle?: (patch: Partial<TitleSettings>) => void
   activePreset: string | null
-  titlePresets: { key: string; label: string; previewStyle: React.CSSProperties }[]
+  titlePresets: Array<{
+    key: string
+    label: string
+    emoji?: string
+    style?: Record<string, unknown>
+    previewStyle?: React.CSSProperties
+  }>
   /* ── 批量生成（#1677）── */
   /** 生成数量 */
   previewCount?: number
@@ -84,6 +92,7 @@ const Step4TitleSettings: React.FC<Step4TitleSettingsProps> = (props) => {
     onToggleStroke,
     onToggleShadow,
     onApplyPreset,
+    onUpdateStyle,
     activePreset,
     titlePresets,
     previewCount = 1,
@@ -285,6 +294,8 @@ const Step4TitleSettings: React.FC<Step4TitleSettingsProps> = (props) => {
         onToggleStroke={onToggleStroke}
         onToggleShadow={onToggleShadow}
         onApplyPreset={onApplyPreset}
+        onUpdateStyle={onUpdateStyle}
+        showCoverToggle
         activePreset={activePreset}
         titlePresets={titlePresets}
         POSITION_OPTIONS={POSITION_OPTIONS}

apps/web/src/pages/generate/components/title/TitleMiniPreview.tsx

@@ -0,0 +1,221 @@
+/**
+ * 标题迷你 Canvas 预览（#2001）
+ *
+ * 渲染一张指定宽度的小 Canvas 预览标题效果，用于：
+ * - 预设卡片缩略图
+ * - 样式面板顶部的实时预览
+ *
+ * 与 titleCanvas.ts 渲染逻辑保持一致，但：
+ * - 固定分辨率（width × 宽高比约 2:1）
+ * - 不调用 ffmpeg，只做视觉预览
+ * - 支持背景色块、描边宽度/颜色、阴影参数化、行距、自动换行
+ */
+import React, { useEffect, useRef } from "react"
+import type { TitleSettings } from "../../types"
+import { getFontFamily } from "../../constants"
+
+interface Props {
+  settings: TitleSettings
+  width?: number
+  sampleText?: string
+  /** 背景（预览用，默认深色渐变模拟视频底） */
+  background?: string
+  /** 高度（可选，默认 width/2） */
+  height?: number
+}
+
+/** 按 maxCharsPerLine 自动换行 */
+function wrapLines(text: string, maxChars: number): string[] {
+  const manual = text
+    .split(/[/／\n]/)
+    .map((l) => l.trim())
+    .filter(Boolean)
+  if (!maxChars || maxChars <= 0) return manual
+  const out: string[] = []
+  for (const line of manual) {
+    if (line.length <= maxChars) {
+      out.push(line)
+      continue
+    }
+    let cur = ""
+    for (const ch of line) {
+      cur += ch
+      if (cur.length >= maxChars) {
+        out.push(cur)
+        cur = ""
+      }
+    }
+    if (cur) out.push(cur)
+  }
+  return out
+}
+
+const TitleMiniPreview: React.FC<Props> = ({
+  settings,
+  width = 200,
+  sampleText,
+  background = "linear-gradient(135deg,#1f2937,#111827)",
+  height,
+}) => {
+  const canvasRef = useRef<HTMLCanvasElement>(null)
+  const h = height ?? Math.round(width / 1.8)
+  const text = (sampleText || settings.title || "预览标题").trim() || "预览标题"
+
+  useEffect(() => {
+    const cvs = canvasRef.current
+    if (!cvs) return
+    const dpr = window.devicePixelRatio || 1
+    cvs.width = width * dpr
+    cvs.height = h * dpr
+    cvs.style.width = `${width}px`
+    cvs.style.height = `${h}px`
+    const ctx = cvs.getContext("2d")
+    if (!ctx) return
+    ctx.scale(dpr, dpr)
+    ctx.clearRect(0, 0, width, h)
+
+    // 背景
+    ctx.fillStyle = "#111827"
+    ctx.fillRect(0, 0, width, h)
+
+    // 分辨率缩放：以 360 宽为基准（对应 720p 的一半）
+    const scale = width / 360
+    const r = (v: number) => Math.round(v * scale)
+
+    // 字体
+    const size = r(settings.size)
+    const ff = getFontFamily(settings.font)
+    const parts: string[] = []
+    if (settings.italic) parts.push("italic")
+    if (settings.bold) parts.push("bold")
+    parts.push(`${size}px`, ff)
+    ctx.font = parts.join(" ")
+    ctx.textAlign = "center"
+    ctx.textBaseline = "middle"
+    ctx.fillStyle = settings.color
+    ctx.lineJoin = "round"
+
+    // 阴影
+    const shadowEnabled = !!settings.shadow
+    const prevShadow = {
+      c: ctx.shadowColor,
+      b: ctx.shadowBlur,
+      ox: ctx.shadowOffsetX,
+      oy: ctx.shadowOffsetY,
+    }
+    if (shadowEnabled) {
+      ctx.shadowColor = settings.shadowColor ?? "rgba(0,0,0,0.8)"
+      ctx.shadowBlur = r(settings.shadowBlur ?? 4)
+      ctx.shadowOffsetX = r(settings.shadowOffsetX ?? 2)
+      ctx.shadowOffsetY = r(settings.shadowOffsetY ?? 2)
+    }
+
+    // 换行
+    const lines = wrapLines(text, settings.maxCharsPerLine ?? 0)
+    const lineH = size * (settings.lineHeight ?? 1.2)
+    const totalH = lines.length * lineH
+    let startY: number
+    if (settings.position === "top") {
+      startY = size / 2 + r(settings.marginTop ?? 24)
+    } else if (settings.position === "center") {
+      startY = h / 2 - totalH / 2 + size / 2
+    } else {
+      // bottom
+      startY = h - totalH - r(16) + size / 2
+    }
+    let centerX = width / 2
+    if (settings.position === "custom" && settings.posX != null) {
+      centerX = (settings.posX / 100) * width
+    }
+
+    // 背景块
+    if (settings.bgEnabled) {
+      const pad = r(settings.bgPadding ?? 12)
+      const rad = r(settings.bgRadius ?? 8)
+      let maxLineW = 0
+      for (const l of lines) {
+        const m = ctx.measureText(l)
+        if (m.width > maxLineW) maxLineW = m.width
+      }
+      const bw = maxLineW + pad * 2
+      const bh = totalH + pad * 2
+      const bx = centerX - bw / 2
+      const by = startY - size / 2 - pad + (size - lineH) / 2
+      ctx.shadowColor = "rgba(0,0,0,0)"
+      ctx.shadowBlur = 0
+      ctx.fillStyle = settings.bgColor ?? "rgba(0,0,0,0.5)"
+      roundRect(ctx, bx, by, bw, bh, rad)
+      ctx.fill()
+      // 恢复阴影
+      if (shadowEnabled) {
+        ctx.shadowColor = settings.shadowColor ?? "rgba(0,0,0,0.8)"
+        ctx.shadowBlur = r(settings.shadowBlur ?? 4)
+        ctx.shadowOffsetX = r(settings.shadowOffsetX ?? 2)
+        ctx.shadowOffsetY = r(settings.shadowOffsetY ?? 2)
+      }
+    }
+
+    // 描边（先画，再画填充）
+    const strokeEnabled = !!settings.stroke && (settings.strokeWidth ?? 0) > 0
+    lines.forEach((line, i) => {
+      const y = startY + i * lineH
+      if (strokeEnabled) {
+        ctx.shadowColor = "rgba(0,0,0,0)"
+        ctx.shadowBlur = 0
+        ctx.lineWidth = r(settings.strokeWidth ?? 4)
+        ctx.strokeStyle = settings.strokeColor ?? "#000000"
+        ctx.strokeText(line, centerX, y)
+        // 恢复阴影
+        if (shadowEnabled) {
+          ctx.shadowColor = settings.shadowColor ?? "rgba(0,0,0,0.8)"
+          ctx.shadowBlur = r(settings.shadowBlur ?? 4)
+          ctx.shadowOffsetX = r(settings.shadowOffsetX ?? 2)
+          ctx.shadowOffsetY = r(settings.shadowOffsetY ?? 2)
+        }
+      }
+      ctx.fillText(line, centerX, y)
+    })
+
+    // 恢复
+    ctx.shadowColor = prevShadow.c
+    ctx.shadowBlur = prevShadow.b
+    ctx.shadowOffsetX = prevShadow.ox
+    ctx.shadowOffsetY = prevShadow.oy
+  }, [settings, width, h, text])
+
+  return (
+    <canvas
+      ref={canvasRef}
+      style={{
+        borderRadius: 6,
+        display: "block",
+        maxWidth: "100%",
+        background,
+      }}
+    />
+  )
+}
+
+function roundRect(
+  ctx: CanvasRenderingContext2D,
+  x: number,
+  y: number,
+  w: number,
+  h: number,
+  r: number,
+) {
+  const rr = Math.min(r, w / 2, h / 2)
+  ctx.beginPath()
+  ctx.moveTo(x + rr, y)
+  ctx.lineTo(x + w - rr, y)
+  ctx.quadraticCurveTo(x + w, y, x + w, y + rr)
+  ctx.lineTo(x + w, y + h - rr)
+  ctx.quadraticCurveTo(x + w, y + h, x + w - rr, y + h)
+  ctx.lineTo(x + rr, y + h)
+  ctx.quadraticCurveTo(x, y + h, x, y + h - rr)
+  ctx.lineTo(x, y + rr)
+  ctx.quadraticCurveTo(x, y, x + rr, y)
+  ctx.closePath()
+}
+
+export default TitleMiniPreview

apps/web/src/pages/generate/components/title/TitleStylePanel.css

@@ -190,3 +190,255 @@
   border-color: var(--primary-color);
   color: #fff;
 }
+
+/* ============================================================
+   #2001 爆款标题样式面板升级 — 新增样式（ts- 前缀）
+   ============================================================ */
+
+.ts-panel {
+  position: relative;
+}
+
+/* 预览 */
+.ts-preview-wrap {
+  margin-bottom: 14px;
+  display: flex;
+  justify-content: center;
+  padding: 10px;
+  background: #0f172a;
+  border-radius: 8px;
+}
+
+/* 表单字段 */
+.ts-form-field {
+  margin-bottom: 12px;
+}
+.ts-form-field label {
+  display: block;
+  font-weight: 600;
+  margin-bottom: 6px;
+  font-size: 12px;
+  color: var(--text-primary, #1f2937);
+}
+.ts-field-label-row {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin-bottom: 6px;
+}
+.ts-field-value {
+  font-size: 12px;
+  font-weight: 600;
+  color: var(--primary-color, #7c3aed);
+}
+.ts-row-2 {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 10px;
+}
+.ts-half {
+  margin-bottom: 0;
+}
+
+.ts-select {
+  width: 100%;
+  height: 34px;
+  border: 1px solid var(--border-color, #e5e7eb);
+  border-radius: 6px;
+  background: var(--bg-primary, #fff);
+  padding: 0 10px;
+  font-size: 13px;
+  outline: 0;
+  color: var(--text-primary, #1f2937);
+}
+.ts-select:focus {
+  border-color: var(--primary-color, #7c3aed);
+  box-shadow: 0 0 0 2px rgba(124, 58, 237, 0.1);
+}
+.ts-input {
+  width: 100%;
+  height: 34px;
+  border: 1px solid var(--border-color, #e5e7eb);
+  border-radius: 6px;
+  padding: 0 10px;
+  font-size: 13px;
+  outline: 0;
+}
+
+.ts-slider {
+  width: 100%;
+  height: 4px;
+  -webkit-appearance: none;
+  appearance: none;
+  background: #e5e7eb;
+  border-radius: 2px;
+  outline: none;
+}
+.ts-slider::-webkit-slider-thumb {
+  -webkit-appearance: none;
+  appearance: none;
+  width: 16px;
+  height: 16px;
+  border-radius: 50%;
+  background: #7c3aed;
+  cursor: pointer;
+  border: 2px solid #fff;
+  box-shadow: 0 1px 3px rgba(0, 0, 0, 0.2);
+}
+.ts-slider::-moz-range-thumb {
+  width: 16px;
+  height: 16px;
+  border-radius: 50%;
+  background: #7c3aed;
+  cursor: pointer;
+  border: 2px solid #fff;
+}
+
+/* 样式按钮 B/I/S/☁ */
+.ts-style-btns {
+  display: flex;
+  gap: 6px;
+}
+.ts-style-btn {
+  width: 34px;
+  height: 34px;
+  border-radius: 6px;
+  border: 1px solid #e5e7eb;
+  background: #fff;
+  cursor: pointer;
+  font-size: 14px;
+  transition: 0.15s;
+  color: #374151;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+}
+.ts-style-btn:hover {
+  border-color: #7c3aed;
+  color: #7c3aed;
+}
+.ts-style-btn.active {
+  background: #faf5ff;
+  color: #6d28d9;
+  border-color: #7c3aed;
+  font-weight: 700;
+}
+
+/* 色板 */
+.ts-color-row {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 6px;
+  align-items: center;
+}
+.ts-color-swatch {
+  width: 24px;
+  height: 24px;
+  border-radius: 4px;
+  border: 2px solid #fff;
+  box-shadow: 0 0 0 1px #e5e7eb;
+  cursor: pointer;
+  padding: 0;
+  transition: 0.15s;
+}
+.ts-color-swatch:hover {
+  transform: scale(1.1);
+}
+.ts-color-swatch.active {
+  box-shadow: 0 0 0 2px #7c3aed;
+  transform: scale(1.1);
+}
+.ts-color-custom {
+  background: repeating-conic-gradient(#ccc 0% 25%, #fff 0% 50%) 50%/8px 8px;
+  color: #666;
+  font-size: 14px;
+  line-height: 20px;
+}
+.ts-color-native {
+  width: 0;
+  height: 0;
+  border: 0;
+  padding: 0;
+}
+
+/* 预设网格 10个 - 5列 */
+.ts-presets-grid {
+  display: grid;
+  grid-template-columns: repeat(5, 1fr);
+  gap: 6px;
+}
+.ts-preset-card {
+  border: 1px solid #e5e7eb;
+  border-radius: 6px;
+  background: #fff;
+  padding: 4px;
+  cursor: pointer;
+  transition: 0.15s;
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+}
+.ts-preset-card:hover {
+  border-color: #7c3aed;
+}
+.ts-preset-card.active {
+  border-color: #7c3aed;
+  background: #faf5ff;
+  box-shadow: 0 0 0 1px #7c3aed;
+}
+.ts-preset-preview {
+  height: 34px;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  overflow: hidden;
+  border-radius: 4px;
+  background: #0f172a;
+}
+.ts-preset-preview canvas {
+  max-width: 100%;
+  max-height: 100%;
+}
+.ts-preset-meta {
+  display: flex;
+  align-items: center;
+  gap: 2px;
+  font-size: 10px;
+  color: #4b5563;
+  justify-content: center;
+  white-space: nowrap;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  padding: 0 2px 2px;
+}
+.ts-preset-emoji {
+  font-size: 11px;
+}
+.ts-preset-label {
+  overflow: hidden;
+  text-overflow: ellipsis;
+}
+
+.ts-toggle-row label {
+  display: inline-flex;
+  align-items: center;
+  gap: 6px;
+  font-size: 13px;
+  font-weight: 500;
+  cursor: pointer;
+  margin-bottom: 10px;
+}
+.ts-toggle-row input[type="checkbox"] {
+  width: 16px;
+  height: 16px;
+  accent-color: #7c3aed;
+}
+
+/* Tabs 紧凑样式 */
+.xx-title-style-section .ant-tabs-nav {
+  margin-bottom: 10px;
+}
+.xx-title-style-section .ant-tabs-tab {
+  font-size: 12px !important;
+  padding: 6px 8px !important;
+}

apps/web/src/pages/generate/components/title/TitleStylePanel.tsx

@@ -1,12 +1,26 @@
 /**
- * 标题样式设置区
- * 位置/字体/字号/样式按钮/预设
+ * 标题样式设置面板（#2001 升级）
+ *
+ * P0：描边宽度滑块 / 描边颜色选择器 / 每行最大字符数 / 行距+顶部边距 /
+ *     4款爆款字体 / 抖音爆款黄预设
+ * P1：阴影参数化 / 背景色块 / Canvas 实时迷你预览 /
+ *     封面独立标题配置入口
+ *
+ * 向后兼容：旧的 onToggleBold/Italic/Stroke/Shadow/onUpdatePosition/onUpdateFont/
+ * onUpdateSize/onApplyPreset props 全部保留；新增字段通过 onUpdateStyle 统一回写。
  */
-import React from "react"
+import React, { useState } from "react"
+import { Tabs } from "antd"
 import type { TitleSettings } from "../../types"
-import TitlePresetsGrid from "./TitlePresetsGrid"
-// 标题样式面板共用样式（#1809 ⑦）：智能剪辑与 AI数字人复用同一组件，
-// 由组件自带样式，避免 AI数字人页面重复引入整个 generate.css
+import {
+  FONT_OPTIONS as NEW_FONT_OPTIONS,
+  TITLE_PRESETS,
+  TITLE_COLOR_PALETTE,
+  STROKE_COLOR_PALETTE,
+  BG_COLOR_PALETTE,
+} from "@/components/title/constants"
+
+import TitleMiniPreview from "./TitleMiniPreview"
 import "./TitleStylePanel.css"
 
 interface PositionOption {
@@ -14,14 +28,17 @@ interface PositionOption {
   label: string
 }
 
-interface TitlePresetItem {
+interface LegacyPreset {
   key: string
   label: string
-  previewStyle: React.CSSProperties
+  emoji?: string
+  style?: Record<string, unknown>
+  previewStyle?: React.CSSProperties
 }
 
 interface TitleStylePanelProps {
   settings: TitleSettings
+  /* 旧 props（兼容） */
   onUpdatePosition: (position: string) => void
   onUpdateFont: (font: string) => void
   onUpdateSize: (size: number) => void
@@ -31,9 +48,130 @@ interface TitleStylePanelProps {
   onToggleShadow: () => void
   onApplyPreset: (presetKey: string) => void
   activePreset: string | null
-  titlePresets: TitlePresetItem[]
+  titlePresets: LegacyPreset[]
   POSITION_OPTIONS: PositionOption[]
-  FONT_OPTIONS: string[]
+  FONT_OPTIONS?: Array<{ value: string; label: string; family?: string; tag?: string }>
+  /* 新增：统一字段更新 */
+  onUpdateStyle?: (patch: Partial<TitleSettings>) => void
+  /* 是否显示封面独立标题切换 */
+  showCoverToggle?: boolean
+  /** 画布预览宽度（默认 200） */
+  previewWidth?: number
+}
+
+/* ── 通用 Slider + Label 行 ── */
+const SliderRow: React.FC<{
+  label: string
+  value: number
+  min: number
+  max: number
+  step?: number
+  unit?: string
+  onChange: (v: number) => void
+}> = ({ label, value, min, max, step = 1, unit = "px", onChange }) => (
+  <div className="ts-form-field">
+    <div className="ts-field-label-row">
+      <label>{label}</label>
+      <span className="ts-field-value">
+        {value}
+        {unit}
+      </span>
+    </div>
+    <input
+      type="range"
+      className="ts-slider"
+      min={min}
+      max={max}
+      step={step}
+      value={value}
+      onChange={(e) => onChange(Number(e.target.value))}
+    />
+  </div>
+)
+
+/* ── 色板 + 自定义颜色选择 ── */
+const ColorPicker: React.FC<{
+  label?: string
+  value: string
+  palette: string[]
+  onChange: (c: string) => void
+}> = ({ label, value, palette, onChange }) => {
+  const [customOpen, setCustomOpen] = useState(false)
+  return (
+    <div className="ts-form-field">
+      {label && <label>{label}</label>}
+      <div className="ts-color-row">
+        {palette.map((c) => (
+          <button
+            key={c}
+            type="button"
+            className={`ts-color-swatch${value.toLowerCase() === c.toLowerCase() ? " active" : ""}`}
+            style={{ background: c }}
+            onClick={() => onChange(c)}
+            title={c}
+          />
+        ))}
+        <button
+          type="button"
+          className="ts-color-swatch ts-color-custom"
+          onClick={() => setCustomOpen((v) => !v)}
+          title="自定义颜色"
+        >
+          +
+        </button>
+        <input
+          type="color"
+          className="ts-color-native"
+          value={value.startsWith("rgba") ? "#000000" : value}
+          onChange={(e) => onChange(e.target.value)}
+          style={{
+            opacity: customOpen ? 1 : 0,
+            position: customOpen ? "static" : "absolute",
+            pointerEvents: customOpen ? "auto" : "none",
+            width: 0,
+            height: 0,
+          }}
+        />
+      </div>
+      <div style={{ fontSize: 11, color: "#9ca3af", marginTop: 2 }}>
+        当前：<code style={{ fontSize: 11 }}>{value}</code>
+      </div>
+    </div>
+  )
+}
+
+/* ── 预设网格（含爆款黄，10 个 + 迷你 Canvas 缩略） ── */
+const PresetGrid: React.FC<{
+  activePreset: string | null
+  onApply: (key: string) => void
+  settings: TitleSettings
+}> = ({ activePreset, onApply, settings }) => {
+  return (
+    <div className="ts-presets-grid">
+      {TITLE_PRESETS.map((p) => {
+        const isActive = activePreset === p.key
+        // 合并当前 style 与 preset.style 用于预览（仅预览时覆盖）
+        const previewStyle: TitleSettings = { ...settings, ...(p.style as Partial<TitleSettings>) }
+        return (
+          <button
+            key={p.key}
+            type="button"
+            className={`ts-preset-card${isActive ? " active" : ""}`}
+            onClick={() => onApply(p.key)}
+            title={p.label}
+          >
+            <div className="ts-preset-preview">
+              <TitleMiniPreview settings={previewStyle} width={100} sampleText="标题" />
+            </div>
+            <div className="ts-preset-meta">
+              <span className="ts-preset-emoji">{p.emoji}</span>
+              <span className="ts-preset-label">{p.label}</span>
+            </div>
+          </button>
+        )
+      })}
+    </div>
+  )
 }
 
 const TitleStylePanel: React.FC<TitleStylePanelProps> = ({
@@ -47,107 +185,352 @@ const TitleStylePanel: React.FC<TitleStylePanelProps> = ({
   onToggleShadow,
   onApplyPreset,
   activePreset,
-  titlePresets,
+  titlePresets: _titlePresets,
   POSITION_OPTIONS,
-  FONT_OPTIONS,
+  showCoverToggle = false,
+  previewWidth = 220,
+  onUpdateStyle,
 }) => {
+  const upd = (patch: Partial<TitleSettings>) => {
+    onUpdateStyle?.(patch)
+  }
+
+  /* 封面独立标题切换 */
+  const [coverOpen, setCoverOpen] = useState(!!settings.coverTitle)
+
   return (
-    <div className="xx-title-style-section">
-      <h4 className="xx-section-subtitle">标题样式</h4>
-
-      {/* 位置 + 字体 一行 */}
-      <div className="xx-title-style-row">
-        <div className="xx-form-field xx-half-field">
-          <label>位置</label>
-          <select
-            className="xx-form-select"
-            value={settings.position}
-            onChange={(e) => onUpdatePosition(e.target.value)}
-          >
-            {POSITION_OPTIONS.map((opt) => (
-              <option key={opt.value} value={opt.value}>
-                {opt.label}
-              </option>
-            ))}
-          </select>
-        </div>
-        <div className="xx-form-field xx-half-field">
-          <label>字体</label>
-          <select
-            className="xx-form-select"
-            value={settings.font}
-            onChange={(e) => onUpdateFont(e.target.value)}
-          >
-            {FONT_OPTIONS.map((f) => (
-              <option key={f} value={f}>
-                {f}
-              </option>
-            ))}
-          </select>
-        </div>
-      </div>
-
-      {/* 字号滑块 */}
-      <div className="xx-form-field">
-        <div className="xx-field-label-row">
-          <label>字号</label>
-          <span className="xx-field-value">{settings.size}px</span>
-        </div>
-        <input
-          className="xx-slider"
-          type="range"
-          min={12}
-          max={128}
-          value={settings.size}
-          onChange={(e) => onUpdateSize(Number(e.target.value))}
+    <div className="xx-title-style-section ts-panel">
+      {/* 实时迷你预览 */}
+      <div className="ts-preview-wrap">
+        <TitleMiniPreview
+          settings={settings}
+          width={previewWidth}
+          sampleText={settings.title || "预览标题文字"}
         />
       </div>
 
-      {/* 预设样式 */}
-      <div className="xx-form-field">
-        <label>预设样式</label>
-        <TitlePresetsGrid
-          presets={titlePresets}
-          activePreset={activePreset}
-          onApply={onApplyPreset}
-          fontFamily={settings.font}
-        />
+      {/* 预设样式（10个，含抖音爆款黄） */}
+      <div className="ts-form-field">
+        <label>爆款预设</label>
+        <PresetGrid activePreset={activePreset} onApply={onApplyPreset} settings={settings} />
       </div>
 
-      {/* 样式按钮：粗体/斜体/描边/阴影 */}
-      <div className="xx-form-field">
-        <label>样式</label>
-        <div className="xx-style-btns">
-          <button
-            className={`xx-style-btn ${settings.bold ? "active" : ""}`}
-            onClick={onToggleBold}
-            title="粗体"
-          >
-            <b>B</b>
-          </button>
-          <button
-            className={`xx-style-btn ${settings.italic ? "active" : ""}`}
-            onClick={onToggleItalic}
-            title="斜体"
-          >
-            <i>I</i>
-          </button>
-          <button
-            className={`xx-style-btn ${settings.stroke ? "active" : ""}`}
-            onClick={onToggleStroke}
-            title="描边"
-          >
-            S
-          </button>
-          <button
-            className={`xx-style-btn ${settings.shadow ? "active" : ""}`}
-            onClick={onToggleShadow}
-            title="阴影"
-          >
-            ☁
-          </button>
-        </div>
-      </div>
+      <Tabs
+        size="small"
+        defaultActiveKey="basic"
+        items={[
+          {
+            key: "basic",
+            label: "基础",
+            children: (
+              <>
+                {/* 位置 + 字体 */}
+                <div className="ts-row-2">
+                  <div className="ts-form-field ts-half">
+                    <label>位置</label>
+                    <select
+                      className="ts-select"
+                      value={settings.position}
+                      onChange={(e) => onUpdatePosition(e.target.value)}
+                    >
+                      {POSITION_OPTIONS.map((o) => (
+                        <option key={o.value} value={o.value}>
+                          {o.label}
+                        </option>
+                      ))}
+                    </select>
+                  </div>
+                  <div className="ts-form-field ts-half">
+                    <label>字体</label>
+                    <select
+                      className="ts-select"
+                      value={settings.font}
+                      onChange={(e) => onUpdateFont(e.target.value)}
+                    >
+                      {NEW_FONT_OPTIONS.map((f) => (
+                        <option key={f.value} value={f.value}>
+                          {f.tag === "hot" ? "🔥 " : f.tag === "new" ? "🆕 " : ""}
+                          {f.label}
+                        </option>
+                      ))}
+                    </select>
+                  </div>
+                </div>
+
+                <SliderRow
+                  label="字号"
+                  value={settings.size}
+                  min={16}
+                  max={120}
+                  onChange={onUpdateSize}
+                />
+
+                {/* 样式按钮 */}
+                <div className="ts-form-field">
+                  <label>样式</label>
+                  <div className="ts-style-btns">
+                    <button
+                      type="button"
+                      className={`ts-style-btn${settings.bold ? " active" : ""}`}
+                      onClick={onToggleBold}
+                    >
+                      <b>B</b>
+                    </button>
+                    <button
+                      type="button"
+                      className={`ts-style-btn${settings.italic ? " active" : ""}`}
+                      onClick={onToggleItalic}
+                    >
+                      <i>I</i>
+                    </button>
+                    <button
+                      type="button"
+                      className={`ts-style-btn${settings.stroke ? " active" : ""}`}
+                      onClick={() => {
+                        onToggleStroke()
+                        // 如果之前 strokeWidth 为 0，启用时给个默认值
+                        if (!settings.stroke && (settings.strokeWidth ?? 0) < 2) {
+                          upd({ strokeWidth: 4 })
+                        }
+                      }}
+                      title="描边"
+                    >
+                      S
+                    </button>
+                    <button
+                      type="button"
+                      className={`ts-style-btn${settings.shadow ? " active" : ""}`}
+                      onClick={() => {
+                        onToggleShadow()
+                        if (!settings.shadow) {
+                          upd({
+                            shadowOffsetX: 2,
+                            shadowOffsetY: 2,
+                            shadowBlur: 4,
+                            shadowColor: "rgba(0,0,0,0.8)",
+                          })
+                        }
+                      }}
+                      title="阴影"
+                    >
+                      ☁
+                    </button>
+                  </div>
+                </div>
+
+                {/* 字色 */}
+                <ColorPicker
+                  label="字色"
+                  value={settings.color}
+                  palette={TITLE_COLOR_PALETTE}
+                  onChange={(c) => upd({ color: c })}
+                />
+              </>
+            ),
+          },
+          {
+            key: "stroke",
+            label: "描边",
+            children: (
+              <>
+                <div className="ts-toggle-row">
+                  <label>
+                    <input type="checkbox" checked={settings.stroke} onChange={onToggleStroke} />
+                    启用描边
+                  </label>
+                </div>
+                {settings.stroke && (
+                  <>
+                    <SliderRow
+                      label="描边宽度"
+                      value={settings.strokeWidth ?? 4}
+                      min={0}
+                      max={20}
+                      onChange={(v) => upd({ strokeWidth: v })}
+                    />
+                    <ColorPicker
+                      label="描边颜色"
+                      value={settings.strokeColor ?? "#000000"}
+                      palette={STROKE_COLOR_PALETTE}
+                      onChange={(c) => upd({ strokeColor: c })}
+                    />
+                  </>
+                )}
+              </>
+            ),
+          },
+          {
+            key: "shadow",
+            label: "阴影",
+            children: (
+              <>
+                <div className="ts-toggle-row">
+                  <label>
+                    <input type="checkbox" checked={settings.shadow} onChange={onToggleShadow} />
+                    启用阴影
+                  </label>
+                </div>
+                {settings.shadow && (
+                  <>
+                    <SliderRow
+                      label="X偏移"
+                      value={settings.shadowOffsetX ?? 2}
+                      min={-20}
+                      max={20}
+                      onChange={(v) => upd({ shadowOffsetX: v })}
+                    />
+                    <SliderRow
+                      label="Y偏移"
+                      value={settings.shadowOffsetY ?? 2}
+                      min={-20}
+                      max={20}
+                      onChange={(v) => upd({ shadowOffsetY: v })}
+                    />
+                    <SliderRow
+                      label="模糊半径"
+                      value={settings.shadowBlur ?? 4}
+                      min={0}
+                      max={30}
+                      onChange={(v) => upd({ shadowBlur: v })}
+                    />
+                    <div className="ts-form-field">
+                      <label>阴影颜色</label>
+                      <input
+                        type="text"
+                        className="ts-input"
+                        value={settings.shadowColor ?? "rgba(0,0,0,0.8)"}
+                        onChange={(e) => upd({ shadowColor: e.target.value })}
+                        placeholder="rgba(0,0,0,0.8)"
+                      />
+                    </div>
+                  </>
+                )}
+              </>
+            ),
+          },
+          {
+            key: "bg",
+            label: "背景",
+            children: (
+              <>
+                <div className="ts-toggle-row">
+                  <label>
+                    <input
+                      type="checkbox"
+                      checked={settings.bgEnabled}
+                      onChange={() => upd({ bgEnabled: !settings.bgEnabled })}
+                    />
+                    启用背景色块
+                  </label>
+                </div>
+                {settings.bgEnabled && (
+                  <>
+                    <ColorPicker
+                      label="背景颜色（含透明度）"
+                      value={settings.bgColor}
+                      palette={BG_COLOR_PALETTE}
+                      onChange={(c) => upd({ bgColor: c })}
+                    />
+                    <SliderRow
+                      label="内边距"
+                      value={settings.bgPadding}
+                      min={0}
+                      max={40}
+                      onChange={(v) => upd({ bgPadding: v })}
+                    />
+                    <SliderRow
+                      label="圆角"
+                      value={settings.bgRadius}
+                      min={0}
+                      max={30}
+                      onChange={(v) => upd({ bgRadius: v })}
+                    />
+                  </>
+                )}
+              </>
+            ),
+          },
+          {
+            key: "layout",
+            label: "排版",
+            children: (
+              <>
+                <SliderRow
+                  label="每行最大字符数"
+                  value={settings.maxCharsPerLine ?? 0}
+                  min={0}
+                  max={20}
+                  unit=""
+                  onChange={(v) => upd({ maxCharsPerLine: v })}
+                />
+                <div
+                  className="ts-form-field"
+                  style={{ fontSize: 11, color: "#9ca3af", marginTop: -4 }}
+                >
+                  0 = 不自动换行（按 / 手动分行）
+                </div>
+                <SliderRow
+                  label="行距倍数"
+                  value={Math.round((settings.lineHeight ?? 1.2) * 100) / 100}
+                  min={1}
+                  max={2}
+                  step={0.05}
+                  unit=""
+                  onChange={(v) => upd({ lineHeight: Number(v.toFixed(2)) })}
+                />
+                <SliderRow
+                  label="顶部边距"
+                  value={settings.marginTop ?? 24}
+                  min={0}
+                  max={200}
+                  onChange={(v) => upd({ marginTop: v })}
+                />
+              </>
+            ),
+          },
+          ...(showCoverToggle
+            ? [
+                {
+                  key: "cover",
+                  label: "封面",
+                  children: (
+                    <>
+                      <div className="ts-toggle-row">
+                        <label>
+                          <input
+                            type="checkbox"
+                            checked={coverOpen}
+                            onChange={(e) => {
+                              setCoverOpen(e.target.checked)
+                              if (!e.target.checked) {
+                                upd({ coverTitle: null })
+                              } else {
+                                upd({
+                                  coverTitle: {
+                                    font: settings.font,
+                                    size: Math.round(settings.size * 0.9),
+                                    color: settings.color,
+                                    bold: settings.bold,
+                                  },
+                                })
+                              }
+                            }}
+                          />
+                          封面使用独立标题样式
+                        </label>
+                      </div>
+                      {coverOpen && settings.coverTitle && (
+                        <div style={{ fontSize: 12, color: "#6b7280", lineHeight: 1.6 }}>
+                          封面样式已开启。可在「封面设置」面板单独调整封面标题的字体/字号/颜色。
+                        </div>
+                      )}
+                    </>
+                  ),
+                },
+              ]
+            : []),
+        ]}
+      />
     </div>
   )
 }

apps/web/src/pages/generate/constants.ts

@@ -54,11 +54,28 @@ export const POSITION_OPTIONS = [
   { value: "custom", label: "自定义" },
 ]
 
-/* ── 标题字体选项 ── */
-export const FONT_OPTIONS = ["思源黑体", "思源宋体", "苹方", "微软雅黑", "楷体"]
+/* ── 标题字体选项（#2001：新增 4 款爆款字体） ── */
+export const FONT_OPTIONS = [
+  "优设标题黑",
+  "阿里普惠体Bold",
+  "抖音美好体",
+  "思源黑体Heavy",
+  "思源黑体",
+  "思源宋体",
+  "苹方",
+  "微软雅黑",
+  "楷体",
+]
 
 /* ── 标题字体 CSS font-family 映射（中文显示名 → 浏览器可识别的字体栈） ── */
 export const FONT_FAMILY_MAP: Record<string, string> = {
+  优设标题黑:
+    '"YouSheBiaoTiHei","YouShe Title Black","Source Han Sans SC Heavy","Noto Sans SC","PingFang SC",sans-serif',
+  阿里普惠体Bold:
+    '"Alibaba PuHuiTi Bold","Alibaba PuHuiTi","Source Han Sans SC","PingFang SC",sans-serif',
+  抖音美好体: '"Douyin Sans","DouyinSansBold","Source Han Sans SC Heavy","PingFang SC",sans-serif',
+  思源黑体Heavy:
+    '"Source Han Sans SC Heavy","Noto Sans SC Heavy","Source Han Sans CN Heavy","PingFang SC",sans-serif',
   思源黑体: '"Source Han Sans SC", "Noto Sans SC", "PingFang SC", "Microsoft YaHei", sans-serif',
   思源宋体: '"Source Han Serif SC", "Noto Serif SC", "Songti SC", "SimSun", serif',
   苹方: '"PingFang SC", -apple-system, "Helvetica Neue", sans-serif',

apps/web/src/pages/generate/hooks/useBatchCovers.ts

@@ -30,8 +30,23 @@ interface UseBatchCoversOptions {
     color: string
     position: string
     bold: boolean
+    italic?: boolean
     stroke: boolean
+    strokeWidth?: number
+    strokeColor?: string
     shadow: boolean
+    shadowOffsetX?: number
+    shadowOffsetY?: number
+    shadowBlur?: number
+    shadowColor?: string
+    lineHeight?: number
+    marginTop?: number
+    maxCharsPerLine?: number
+    bgEnabled?: boolean
+    bgColor?: string
+    bgPadding?: number
+    bgRadius?: number
+    lineOverrides?: unknown[]
   }
   covers: string[]
   onCoversChange: CoversChangeFn
@@ -100,8 +115,37 @@ export function useBatchCovers({
                   font_color: titleStyle.color,
                   position: titleStyle.position,
                   bold: titleStyle.bold,
-                  stroke: titleStyle.stroke,
-                  shadow: titleStyle.shadow,
+                  italic: titleStyle.italic,
+                  stroke: titleStyle.stroke
+                    ? {
+                        enabled: true,
+                        width: titleStyle.strokeWidth ?? 4,
+                        color: titleStyle.strokeColor ?? "#000000",
+                      }
+                    : { enabled: false },
+                  shadow: titleStyle.shadow
+                    ? {
+                        enabled: true,
+                        offset_x: titleStyle.shadowOffsetX ?? 2,
+                        offset_y: titleStyle.shadowOffsetY ?? 2,
+                        blur: titleStyle.shadowBlur ?? 4,
+                        color: titleStyle.shadowColor ?? "rgba(0,0,0,0.8)",
+                      }
+                    : { enabled: false },
+                  line_height: titleStyle.lineHeight ?? 1.2,
+                  margin_top: titleStyle.marginTop ?? 24,
+                  max_chars_per_line: titleStyle.maxCharsPerLine ?? 0,
+                  background: titleStyle.bgEnabled
+                    ? {
+                        enabled: true,
+                        color: titleStyle.bgColor,
+                        padding: titleStyle.bgPadding,
+                        radius: titleStyle.bgRadius,
+                      }
+                    : { enabled: false },
+                  line_overrides: (titleStyle.lineOverrides ?? []) as Array<
+                    Record<string, unknown>
+                  >,
                 },
               }
             : {}),

apps/web/src/pages/generate/hooks/useGenerateFormState/index.ts

@@ -34,6 +34,21 @@ const DEFAULT_TITLE_SETTINGS: TitleSettings = {
   color: "#ffffff",
   posX: null,
   posY: null,
+  lineHeight: 1.2,
+  marginTop: 24,
+  maxCharsPerLine: 0,
+  strokeWidth: 4,
+  strokeColor: "#000000",
+  shadowOffsetX: 2,
+  shadowOffsetY: 2,
+  shadowBlur: 4,
+  shadowColor: "rgba(0,0,0,0.8)",
+  bgEnabled: false,
+  bgColor: "rgba(0,0,0,0.5)",
+  bgPadding: 12,
+  bgRadius: 8,
+  lineOverrides: [],
+  coverTitle: null,
 }
 
 export interface GenerateFormState {

apps/web/src/pages/generate/hooks/useGenerateFormState/usePlanConfigLoader.ts

@@ -1,6 +1,7 @@
 import { useEffect } from "react"
 import type { CoverConfig } from "../../types/cover"
 import type { TitleSettings } from "../../types"
+import type { TitleLineOverride } from "@/components/title/types"
 import type { TitleConfig } from "@/api/template-editor"
 import { getEditPlan } from "@/api/template-editor"
 
@@ -12,6 +13,156 @@ interface UsePlanConfigLoaderOptions {
   setSelectedMaterials: (ids: string[]) => void
 }
 
+/** #2001：统一归一化 title_config snake_case -> camelCase TitleSettings */
+function mapTitleCfgToSettings(
+  prev: TitleSettings,
+  tc: TitleConfig & Record<string, unknown>,
+): TitleSettings {
+  const stroke = tc.stroke as
+    boolean | { enabled?: boolean; width?: number; color?: string } | undefined
+  const strokeEnabled: boolean | undefined =
+    typeof stroke === "object" && stroke ? stroke.enabled !== false : !!stroke || undefined
+  const strokeW: number | undefined =
+    typeof stroke === "object" && stroke
+      ? (stroke.width ?? (tc.stroke_width as number | undefined))
+      : (tc.stroke_width as number | undefined)
+  const strokeC: string | undefined =
+    typeof stroke === "object" && stroke
+      ? (stroke.color ?? (tc.stroke_color as string | undefined))
+      : (tc.stroke_color as string | undefined)
+
+  const shadow = tc.shadow as
+    | boolean
+    | { enabled?: boolean; offset_x?: number; offset_y?: number; blur?: number; color?: string }
+    | undefined
+  const shadowEnabled: boolean | undefined =
+    typeof shadow === "object" && shadow ? shadow.enabled !== false : !!shadow || undefined
+  const shOffX: number | undefined =
+    typeof shadow === "object" && shadow
+      ? (shadow.offset_x ?? (tc.shadow_offset_x as number | undefined))
+      : (tc.shadow_offset_x as number | undefined)
+  const shOffY: number | undefined =
+    typeof shadow === "object" && shadow
+      ? (shadow.offset_y ?? (tc.shadow_offset_y as number | undefined))
+      : (tc.shadow_offset_y as number | undefined)
+  const shBlur: number | undefined =
+    typeof shadow === "object" && shadow
+      ? (shadow.blur ?? (tc.shadow_blur as number | undefined))
+      : (tc.shadow_blur as number | undefined)
+  const shColor: string | undefined =
+    typeof shadow === "object" && shadow
+      ? (shadow.color ?? (tc.shadow_color as string | undefined))
+      : (tc.shadow_color as string | undefined)
+
+  const bg = tc.background as
+    { enabled?: boolean; color?: string; padding?: number; radius?: number } | undefined
+  const bgEnabled: boolean | undefined =
+    (bg && typeof bg === "object" ? bg.enabled : undefined) ??
+    (tc.bg_enabled as boolean | undefined)
+  const bgColor: string | undefined =
+    (bg && typeof bg === "object" ? bg.color : undefined) ?? (tc.bg_color as string | undefined)
+  const bgPadding: number | undefined =
+    (bg && typeof bg === "object" ? bg.padding : undefined) ?? (tc.bg_padding as number | undefined)
+  const bgRadius: number | undefined =
+    (bg && typeof bg === "object" ? bg.radius : undefined) ?? (tc.bg_radius as number | undefined)
+
+  const ct = (tc.cover_title_config ?? null) as null | Record<string, unknown>
+  let coverTitle: TitleSettings["coverTitle"] = prev.coverTitle
+  if (ct) {
+    const ctStroke = ct.stroke as
+      boolean | { enabled?: boolean; width?: number; color?: string } | undefined
+    const ctShadow = ct.shadow as
+      | boolean
+      | { enabled?: boolean; offset_x?: number; offset_y?: number; blur?: number; color?: string }
+      | undefined
+    const ctBg = ct.background as
+      { enabled?: boolean; color?: string; padding?: number; radius?: number } | undefined
+    coverTitle = {
+      title: (ct.title as string | undefined) ?? prev.coverTitle?.title ?? "",
+      font: (ct.font as string | undefined) ?? prev.coverTitle?.font,
+      size:
+        (ct.font_size as number | undefined) ??
+        (ct.size as number | undefined) ??
+        prev.coverTitle?.size,
+      color:
+        (ct.font_color as string | undefined) ??
+        (ct.color as string | undefined) ??
+        prev.coverTitle?.color,
+      bold: (ct.bold as boolean | undefined) ?? prev.coverTitle?.bold,
+      italic: (ct.italic as boolean | undefined) ?? prev.coverTitle?.italic,
+      position: (ct.position as string | undefined) ?? prev.coverTitle?.position,
+      stroke:
+        typeof ctStroke === "object" && ctStroke
+          ? ctStroke.enabled !== false
+          : ((ctStroke as boolean | undefined) ?? prev.coverTitle?.stroke),
+      strokeWidth:
+        (typeof ctStroke === "object" && ctStroke ? ctStroke.width : undefined) ??
+        (ct.stroke_width as number | undefined) ??
+        prev.coverTitle?.strokeWidth,
+      strokeColor:
+        (typeof ctStroke === "object" && ctStroke ? ctStroke.color : undefined) ??
+        (ct.stroke_color as string | undefined) ??
+        prev.coverTitle?.strokeColor,
+      shadow:
+        typeof ctShadow === "object" && ctShadow
+          ? ctShadow.enabled !== false
+          : ((ctShadow as boolean | undefined) ?? prev.coverTitle?.shadow),
+      shadowOffsetX:
+        (typeof ctShadow === "object" && ctShadow ? ctShadow.offset_x : undefined) ??
+        (ct.shadow_offset_x as number | undefined) ??
+        prev.coverTitle?.shadowOffsetX,
+      shadowOffsetY:
+        (typeof ctShadow === "object" && ctShadow ? ctShadow.offset_y : undefined) ??
+        (ct.shadow_offset_y as number | undefined) ??
+        prev.coverTitle?.shadowOffsetY,
+      shadowBlur:
+        (typeof ctShadow === "object" && ctShadow ? ctShadow.blur : undefined) ??
+        (ct.shadow_blur as number | undefined) ??
+        prev.coverTitle?.shadowBlur,
+      shadowColor:
+        (typeof ctShadow === "object" && ctShadow ? ctShadow.color : undefined) ??
+        (ct.shadow_color as string | undefined) ??
+        prev.coverTitle?.shadowColor,
+      bgEnabled:
+        ctBg?.enabled ?? (ct.bg_enabled as boolean | undefined) ?? prev.coverTitle?.bgEnabled,
+      bgColor: ctBg?.color ?? (ct.bg_color as string | undefined) ?? prev.coverTitle?.bgColor,
+      bgPadding:
+        ctBg?.padding ?? (ct.bg_padding as number | undefined) ?? prev.coverTitle?.bgPadding,
+      bgRadius: ctBg?.radius ?? (ct.bg_radius as number | undefined) ?? prev.coverTitle?.bgRadius,
+    }
+  }
+
+  const result: TitleSettings = {
+    ...prev,
+    title: (tc.content as string | undefined) || prev.title,
+    aiAutoSelect: (tc.ai_auto_select as boolean | undefined) || false,
+    position: prev.position,
+    font: (tc.font_preset as string | undefined) || prev.font,
+    size: (tc.font_size as number | undefined) || prev.size,
+    color: (tc.font_color as string | undefined) || prev.color,
+    bold: (tc.bold as boolean | undefined) ?? prev.bold,
+    italic: (tc.italic as boolean | undefined) ?? prev.italic,
+    stroke: strokeEnabled ?? prev.stroke,
+    strokeWidth: strokeW ?? prev.strokeWidth,
+    strokeColor: strokeC ?? prev.strokeColor,
+    shadow: shadowEnabled ?? prev.shadow,
+    shadowOffsetX: shOffX ?? prev.shadowOffsetX,
+    shadowOffsetY: shOffY ?? prev.shadowOffsetY,
+    shadowBlur: shBlur ?? prev.shadowBlur,
+    shadowColor: shColor ?? prev.shadowColor,
+    lineHeight: (tc.line_height as number | undefined) ?? prev.lineHeight,
+    marginTop: (tc.margin_top as number | undefined) ?? prev.marginTop,
+    maxCharsPerLine: (tc.max_chars_per_line as number | undefined) ?? prev.maxCharsPerLine,
+    bgEnabled: bgEnabled ?? prev.bgEnabled,
+    bgColor: bgColor ?? prev.bgColor,
+    bgPadding: bgPadding ?? prev.bgPadding,
+    bgRadius: bgRadius ?? prev.bgRadius,
+    lineOverrides: ((tc.line_overrides as unknown[] | undefined) ?? []) as TitleLineOverride[],
+    coverTitle,
+  }
+  return result
+}
+
 /**
  * 从 URL 参数或编辑计划 ID 加载表单配置
  */
@@ -27,14 +178,7 @@ export function usePlanConfigLoader({
     if (!planConfigStr) return
     try {
       const config = JSON.parse(planConfigStr) as {
-        title_config?: {
-          content?: string
-          ai_auto_select?: boolean
-          position?: string
-          font_preset?: string
-          font_size?: number
-          font_color?: string
-        }
+        title_config?: Record<string, unknown>
         subtitle_config?: { enabled?: boolean }
         bgm_config?: { enabled?: boolean; music_id?: string }
         mode?: string
@@ -43,16 +187,8 @@ export function usePlanConfigLoader({
       }
 
       if (config.title_config) {
-        const tc = config.title_config as TitleConfig
-        setTitleSettings((prev: TitleSettings) => ({
-          ...prev,
-          title: tc.content || "",
-          aiAutoSelect: tc.ai_auto_select || false,
-          position: prev.position, // 强制保留默认/用户选择，不从草稿配置同步位置
-          font: tc.font_preset || prev.font,
-          size: tc.font_size || prev.size,
-          color: tc.font_color || prev.color,
-        }))
+        const tc = config.title_config as TitleConfig & Record<string, unknown>
+        setTitleSettings((prev: TitleSettings) => mapTitleCfgToSettings(prev, tc))
       }
       if (config.segments && config.segments.length > 0) {
         const assetIds = config.segments
@@ -76,15 +212,8 @@ export function usePlanConfigLoader({
         if (plan.name) setTitleSettings((prev: TitleSettings) => ({ ...prev, title: plan.name }))
         const cfg = plan.config
         if (cfg?.title_config) {
-          setTitleSettings((prev: TitleSettings) => ({
-            ...prev,
-            aiAutoSelect: cfg.title_config!.ai_auto_select,
-            title: cfg.title_config!.content || prev.title,
-            position: prev.position, // 强制保留默认/用户选择，不从远程草稿同步位置
-            font: cfg.title_config!.font_preset || prev.font,
-            size: cfg.title_config!.font_size || prev.size,
-            color: cfg.title_config!.font_color || prev.color,
-          }))
+          const tc2 = cfg.title_config as unknown as TitleConfig & Record<string, unknown>
+          setTitleSettings((prev: TitleSettings) => mapTitleCfgToSettings(prev, tc2))
         }
         if (cfg?.cover_config) {
           const cc = cfg.cover_config as CoverConfig

apps/web/src/pages/generate/hooks/useGenerateVideo.ts

@@ -241,8 +241,91 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
                       }
                     : {}),
                   bold: props.titleSettings.bold,
-                  stroke: props.titleSettings.stroke,
-                  shadow: props.titleSettings.shadow,
+                  italic: props.titleSettings.italic,
+                  stroke: props.titleSettings.stroke
+                    ? {
+                        enabled: true,
+                        width: props.titleSettings.strokeWidth ?? 4,
+                        color: props.titleSettings.strokeColor ?? "#000000",
+                      }
+                    : { enabled: false },
+                  shadow: props.titleSettings.shadow
+                    ? {
+                        enabled: true,
+                        offset_x: props.titleSettings.shadowOffsetX ?? 2,
+                        offset_y: props.titleSettings.shadowOffsetY ?? 2,
+                        blur: props.titleSettings.shadowBlur ?? 4,
+                        color: props.titleSettings.shadowColor ?? "rgba(0,0,0,0.8)",
+                      }
+                    : { enabled: false },
+                  line_height: props.titleSettings.lineHeight ?? 1.2,
+                  margin_top: props.titleSettings.marginTop ?? 24,
+                  max_chars_per_line: props.titleSettings.maxCharsPerLine ?? 0,
+                  ...(props.titleSettings.bgEnabled
+                    ? {
+                        background: {
+                          enabled: true,
+                          color: props.titleSettings.bgColor,
+                          padding: props.titleSettings.bgPadding,
+                          radius: props.titleSettings.bgRadius,
+                        },
+                      }
+                    : { background: { enabled: false } }),
+                  line_overrides: (props.titleSettings.lineOverrides ?? []).map((lo) => ({
+                    line_index: lo.line_index,
+                    text: lo.text,
+                    size: lo.size,
+                    color: lo.color,
+                    bold: lo.bold,
+                    italic: lo.italic,
+                    stroke: lo.stroke,
+                    highlights: lo.highlights?.map((h) => ({
+                      word: h.word,
+                      color: h.color,
+                      bold: h.bold,
+                      scale: h.scale,
+                    })),
+                  })),
+                  ...(props.titleSettings.coverTitle
+                    ? {
+                        cover_title_config: {
+                          title: props.titleSettings.coverTitle.title,
+                          font: props.titleSettings.coverTitle.font,
+                          font_size: props.titleSettings.coverTitle.size,
+                          font_color: props.titleSettings.coverTitle.color,
+                          bold: props.titleSettings.coverTitle.bold,
+                          italic: props.titleSettings.coverTitle.italic,
+                          position: props.titleSettings.coverTitle.position,
+                          stroke: props.titleSettings.coverTitle.stroke
+                            ? {
+                                enabled: true,
+                                width: props.titleSettings.coverTitle.strokeWidth ?? 4,
+                                color: props.titleSettings.coverTitle.strokeColor ?? "#000000",
+                              }
+                            : { enabled: false },
+                          shadow: props.titleSettings.coverTitle.shadow
+                            ? {
+                                enabled: true,
+                                offset_x: props.titleSettings.coverTitle.shadowOffsetX ?? 2,
+                                offset_y: props.titleSettings.coverTitle.shadowOffsetY ?? 2,
+                                blur: props.titleSettings.coverTitle.shadowBlur ?? 4,
+                                color:
+                                  props.titleSettings.coverTitle.shadowColor ?? "rgba(0,0,0,0.8)",
+                              }
+                            : { enabled: false },
+                          ...(props.titleSettings.coverTitle.bgEnabled
+                            ? {
+                                background: {
+                                  enabled: true,
+                                  color: props.titleSettings.coverTitle.bgColor,
+                                  padding: props.titleSettings.coverTitle.bgPadding,
+                                  radius: props.titleSettings.coverTitle.bgRadius,
+                                },
+                              }
+                            : { background: { enabled: false } }),
+                        },
+                      }
+                    : {}),
                 },
               }
             : {}),

apps/web/src/pages/generate/hooks/useStep4Title/useTitleStyleUpdaters.ts

@@ -1,5 +1,6 @@
 import { useCallback, useMemo } from "react"
 import { TITLE_PRESETS } from "../../constants"
+import { TITLE_PRESETS as NEW_TITLE_PRESETS } from "@/components/title/constants"
 import type { TitleSettings } from "../../types"
 
 interface UseTitleStyleUpdatersOptions {
@@ -97,26 +98,45 @@ export function useTitleStyleUpdaters({
     onTitleSettingsChange({ ...titleSettings, shadow: !titleSettings.shadow })
   }, [titleSettings, onTitleSettingsChange])
 
-  /** 应用预设：只覆盖 color/bold/italic/stroke/shadow，不改变字号 */
+  /** 应用预设（支持新预设细粒度字段） */
   const applyPreset = useCallback(
     (presetKey: string) => {
-      const preset = TITLE_PRESETS.find((p) => p.key === presetKey)
-      if (!preset) return
+      // 优先匹配新预设（10个爆款预设），fallback 旧预设
+      const newPreset = NEW_TITLE_PRESETS.find((p) => p.key === presetKey)
+      const oldPreset = TITLE_PRESETS.find((p) => p.key === presetKey)
+      if (newPreset) {
+        onTitleSettingsChange({
+          ...titleSettings,
+          ...(newPreset.style as Partial<TitleSettings>),
+          // 清除逐行覆盖
+          lineOverrides: [],
+        })
+        return
+      }
+      if (!oldPreset) return
       onTitleSettingsChange({
         ...titleSettings,
-        color: preset.style.color,
-        bold: preset.style.bold,
-        italic: preset.style.italic,
-        stroke: preset.style.stroke,
-        shadow: preset.style.shadow,
+        color: oldPreset.style.color,
+        bold: oldPreset.style.bold,
+        italic: oldPreset.style.italic,
+        stroke: oldPreset.style.stroke,
+        shadow: oldPreset.style.shadow,
       })
     },
     [titleSettings, onTitleSettingsChange],
   )
 
+  /** 通用字段更新（patch） */
+  const updateStyle = useCallback(
+    (patch: Partial<TitleSettings>) => {
+      onTitleSettingsChange({ ...titleSettings, ...patch })
+    },
+    [titleSettings, onTitleSettingsChange],
+  )
+
   return {
     activePreset,
-    titlePresets: TITLE_PRESETS,
+    titlePresets: NEW_TITLE_PRESETS,
     updateTitle,
     toggleAiAutoSelect,
     updatePosition,
@@ -129,5 +149,6 @@ export function useTitleStyleUpdaters({
     toggleStroke,
     toggleShadow,
     applyPreset,
+    updateStyle,
   }
 }

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

@@ -3,8 +3,9 @@
  */
 
 import type { AssetItem } from "@/api/assets"
+import type { TitleLineOverride } from "@/components/title/types"
 
-/* ── 标题设置 ── */
+/* ── 标题设置（#2001 升级：新增描边/阴影/背景/逐行/封面独立样式/排版字段） ── */
 export interface TitleSettings {
   aiAutoSelect: boolean
   title: string
@@ -19,6 +20,58 @@ export interface TitleSettings {
   /** 自由位置坐标（PlayRes 像素），仅当 position="custom" 时有效 */
   posX: number | null
   posY: number | null
+
+  /* ── 排版（P0） ── */
+  /** 行距倍数，默认 1.2 */
+  lineHeight: number
+  /** 顶部边距（position=top，px @720p） */
+  marginTop: number
+  /** 每行最大字符数（4-20），0=不自动换行 */
+  maxCharsPerLine: number
+
+  /* ── 描边参数化（P0） ── */
+  strokeWidth: number
+  strokeColor: string
+
+  /* ── 阴影参数化（P1） ── */
+  shadowOffsetX: number
+  shadowOffsetY: number
+  shadowBlur: number
+  shadowColor: string
+
+  /* ── 背景色块（P1） ── */
+  bgEnabled: boolean
+  bgColor: string
+  bgPadding: number
+  bgRadius: number
+
+  /* ── 逐行独立样式（P1） ── */
+  lineOverrides: TitleLineOverride[]
+
+  /* ── 封面独立标题（P1）：null=沿用主标题 ── */
+  coverTitle: null | {
+    title?: string
+    font?: string
+    size?: number
+    color?: string
+    bold?: boolean
+    italic?: boolean
+    position?: string
+    stroke?: boolean
+    strokeWidth?: number
+    strokeColor?: string
+    shadow?: boolean
+    shadowOffsetX?: number
+    shadowOffsetY?: number
+    shadowBlur?: number
+    shadowColor?: string
+    bgEnabled?: boolean
+    bgColor?: string
+    bgPadding?: number
+    bgRadius?: number
+    lineHeight?: number
+    maxCharsPerLine?: number
+  }
 }
 
 /* ── 智能匹配结果 ── */
@@ -49,28 +102,33 @@ export interface StepDef {
   label: string
 }
 
-/* ── 标题预设样式 ── */
-export interface TitlePresetStyle {
-  size: number
-  color: string
-  bold: boolean
-  italic: boolean
-  stroke: boolean
-  shadow: boolean
-}
-
-export interface TitlePreset {
-  key: string
-  label: string
-  style: TitlePresetStyle
-  previewStyle: Record<string, string | number>
-}
-
-/* ── 生成结果视频 ── */
-export interface GeneratedVideoResult {
-  id: string
-  url: string
-  thumbnail: string
-  duration: number
-  title: string
+/** 旧版 TitleSettings 的默认值字段（P0/P1 新字段补齐默认值） */
+export const DEFAULT_TITLE_SETTINGS_FULL: TitleSettings = {
+  aiAutoSelect: false,
+  title: "",
+  position: "top",
+  font: "思源黑体",
+  size: 28,
+  bold: true,
+  italic: false,
+  stroke: true,
+  shadow: false,
+  color: "#ffffff",
+  posX: null,
+  posY: null,
+  lineHeight: 1.2,
+  marginTop: 24,
+  maxCharsPerLine: 0,
+  strokeWidth: 4,
+  strokeColor: "#000000",
+  shadowOffsetX: 2,
+  shadowOffsetY: 2,
+  shadowBlur: 4,
+  shadowColor: "rgba(0,0,0,0.8)",
+  bgEnabled: false,
+  bgColor: "rgba(0,0,0,0.5)",
+  bgPadding: 12,
+  bgRadius: 8,
+  lineOverrides: [],
+  coverTitle: null,
 }

deploy/gpu_worker/MUSELIP_ANALYSIS.md

@@ -1,218 +0,0 @@
-# MuseTalk 口型精度问题分析与修复方案
-
-## 问题总结
-
-运维逐帧分析确认：
-- ✅ 音频对齐正常（0ms偏移，相关系数0.99997）
-- ❌ **口型只跟音频能量张合，没有音素级精度**（发o/u没有圆唇，发m/b/p没有闭唇，静音段嘴巴还张着）
-- ❌ 5s循环边界有视觉跳变（帧间差异是正常的3-4倍）
-
-## 根因分析
-
-### 根因 1：`musetalk_server.py` 的 `_run_inference()` 是 STUB 代码（最严重）
-
-`deploy/gpu_worker/musetalk_server.py` 第 343-466 行：
-
-```python
-# 2. 模拟 MuseTalk 推理：输入原视频帧 + 全量音频，输出音频时长的无声画面。
-# TODO: 替换为 MuseTalk 真实推理逻辑。
-logger.warning("使用示例推理逻辑，未实际调用 MuseTalk 模型")
-```
-
-**当前代码从未调用 MuseTalk 模型**。整个推理流程是：
-1. 从视频提取帧（ffmpeg）
-2. 生成一段循环原视频的无声画面（ffmpeg `-stream_loop`）
-3. 用 ffmpeg `-c:v copy` 封装音频
-
-**根本没有 MuseTalk 推理！** 所以"口型只跟能量张合"可能是因为 MuseTalk 的实际推理代码没被调用。
-
-### 根因 2：音频格式未预处理
-
-`gpu_worker.py` 把音频下载为 `input_audio.bin`（第 360 行），直接透传给 `musetalk_server.py`。
-`musetalk_server.py` 保存为 `input_audio.wav`（第 527 行）但**没有做任何格式转换**。
-
-CosyVoice TTS 输出的音频格式：
-- **采样率**: 22050Hz（`cosyvoice_sample_rate = 22050`）
-- **格式**: MP3（`cosyvoice_format = "mp3"`）
-
-MuseTalk 期望的输入：
-- **采样率**: 16000Hz
-- **声道**: mono
-- **格式**: WAV（16bit PCM）
-
-**22050Hz MP3 → 16kHz mono 16bit WAV 的重采样转换完全没有做**。
-
-这会导致 MuseTalk 的 whisper feature extractor 拿到错误采样率的音频，提取的 mel 频谱特征频率错位，音素识别完全错误。
-
-### 根因 3：循环边界无过渡帧
-
-`_mux_video_with_audio()` 的兜底路径（第 241-286 行）用 `ffmpeg -stream_loop -1` 直接循环视频，循环边界处：
-- 最后一帧 → 第一帧：没有过渡，硬切
-- 帧间差异是正常帧的 3-4 倍
-
-## 修复方案
-
-### 修复 1：实现真正的 MuseTalk 推理调用
-
-`_run_inference()` 必须调用真实的 MuseTalk 模型。参考 MuseTalk 官方推理流程：
-
-```python
-def _run_inference_real(video_path, audio_path, output_path):
-    """真正的 MuseTalk 推理 — 替换 stub。"""
-    # 1. 音频预处理：转换为 16kHz mono 16bit WAV
-    preprocessed_audio = audio_path.parent / "audio_16k_mono.wav"
-    _preprocess_audio(audio_path, preprocessed_audio)
-    
-    # 2. 调用 MuseTalk 推理
-    # MuseTalk 代码在 ~/projects/MuseTalk/ 下
-    import sys
-    muse_dir = Path.home() / "projects" / "MuseTalk"
-    if str(muse_dir) not in sys.path:
-        sys.path.insert(0, str(muse_dir))
-    
-    from musetalk.utils.utils import get_file_type
-    from musetalk.whisper.audio2feature import load_audio
-    
-    # MuseTalk 内部推理 API（具体取决于部署的 MuseTalk 版本）
-    # 典型调用：
-    # model = MuseTalkModel(...)
-    # result = model.infer(video=video_path, audio=preprocessed_audio)
-```
-
-### 修复 2：音频预处理（必须）
-
-在 `musetalk_server.py` 中添加 `_preprocess_audio()` 函数：
-
-```python
-def _preprocess_audio(input_path: Path, output_path: Path, target_sr: int = 16000) -> None:
-    """将输入音频转换为 MuseTalk 要求的格式：16kHz mono 16bit WAV。
-    
-    MuseTalk 的 whisper audio2feature 要求 16kHz 采样率，
-    当前 TTS 输出 22050Hz MP3，不转换会导致 mel 频谱错位、
-    音素特征提取错误，口型只跟能量不跟音素。
-    """
-    cmd = [
-        "ffmpeg", "-y",
-        "-i", str(input_path),
-        "-ar", str(target_sr),    # 重采样到 16kHz
-        "-ac", "1",                # 单声道
-        "-sample_fmt", "s16",      # 16bit PCM
-        "-f", "wav",               # WAV 格式
-        str(output_path),
-    ]
-    _run_ffmpeg(cmd, timeout=60)
-    
-    if not output_path.exists() or output_path.stat().st_size < 100:
-        raise RuntimeError(f"音频预处理失败: {output_path}")
-    
-    # 验证输出格式
-    try:
-        import wave
-        with wave.open(str(output_path), 'rb') as wf:
-            sr = wf.getframerate()
-            ch = wf.getnchannels()
-            sw = wf.getsampwidth()
-            if sr != target_sr or ch != 1 or sw != 2:
-                logger.warning("音频格式异常: sr=%d ch=%d sw=%d", sr, ch, sw)
-    except Exception as e:
-        logger.warning("无法验证 WAV 格式: %s", e)
-    
-    logger.info("音频预处理完成: %s → %s (%dHz mono 16bit WAV)", 
-                input_path.name, output_path.name, target_sr)
-```
-
-### 修复 3：循环边界 crossfade
-
-在兜底循环路径中，使用 ffmpeg `xfade` 滤镜在循环边界处混合：
-
-```python
-def _mux_with_loop_crossfade(video_path, audio_path, output_path, crossfade_frames=3):
-    """循环视频 + crossfade 过渡，减少循环边界跳变。"""
-    fps = _get_video_fps(video_path)
-    crossfade_duration = crossfade_frames / fps  # 通常 0.12s (3帧@25fps)
-    video_duration = _get_media_duration(video_path)
-    
-    # 使用 tpad 滤镜在视频末尾添加 crossfade
-    # 方案：用 concat 把 [视频 + 视频前N帧crossfade] 拼起来
-    cmd = [
-        "ffmpeg", "-y",
-        "-stream_loop", "-1",
-        "-i", str(video_path),
-        "-i", str(audio_path),
-        "-filter_complex",
-        f"[0:v]split=2[v1][v2];"
-        f"[v2]trim=0:{crossfade_duration},setpts=PTS-STARTPTS+{video_duration - crossfade_duration}/TB[xfade_in];"
-        f"[v1][xfade_in]xfade=transition=fade:duration={crossfade_duration}:offset={video_duration - crossfade_duration}[looped];"
-        f"[looped]format=yuv420p[out]",
-        "-map", "[out]",
-        "-map", "1:a:0",
-        "-c:v", encoder,
-        "-preset", preset,
-        "-c:a", "aac", "-b:a", "128k",
-        "-t", f"{audio_duration:.3f}",
-        str(output_path),
-    ]
-```
-
-**但更简单的方案**：如果 MuseTalk 正确处理了全量音频（输出时长=音频时长），就不需要兜底循环了。crossfade 只是兜底路径的优化。
-
-### 修复 4：推理参数
-
-MuseTalk 的关键推理参数：
-- `bbox_shift`: 口型区域偏移量，默认 0，范围 [-5, 5]。影响口型位置
-- `batch_size`: 推理批次大小，RTX2060 6G 显存建议 4-8
-- 视频帧率：MuseTalk 期望 25fps，当前已正确获取 fps
-
-## 完整修复优先级
-
-| 优先级 | 修复 | 影响 |
-|---|---|---|
-| **P0** | `_run_inference()` 调用真实 MuseTalk 模型 | **没有这个其他都没意义** |
-| **P0** | 音频预处理：22050Hz MP3 → 16kHz mono 16bit WAV | 音素特征正确提取 |
-| **P1** | 循环边界 crossfade | 减少视觉跳变 |
-| **P2** | bbox_shift 参数暴露为可配置 | 微调口型位置 |
-
-## 在 RTX2060 上调试需要什么
-
-1. **确认 MuseTalk 推理代码**：
-   ```bash
-   # 在 RTX2060 上
-   ls ~/projects/MuseTalk/
-   cat ~/projects/MuseTalk/musetalk_server.py  # 看实际部署版本
-   ```
-
-2. **验证音频格式**：
-   ```bash
-   # 在 musetalk_server.py 的推理前添加日志
-   ffprobe -v error -show_entries stream=sample_rate,channels,codec_name,bits_per_sample \
-     -of default /tmp/musetalk_*/input_audio.wav
-   ```
-
-3. **提取中间结果对比**：
-   ```bash
-   # 对比 TTS 原始音频 vs 预处理后的 16kHz WAV
-   # 用 whisper 提取 mel 特征并可视化
-   python -c "
-   import whisper
-   model = whisper.load_model('tiny')
-   audio = whisper.load_audio('/tmp/test_audio.wav')  # 预处理后的
-   audio = whisper.pad_or_trim(audio)
-   mel = whisper.log_mel_spectrogram(audio).to(model.device)
-   # 保存 mel 特征供可视化
-   import numpy as np
-   np.save('/tmp/mel_features.npy', mel.cpu().numpy())
-   "
-   ```
-
-4. **MuseTalk 推理中间帧**：
-   ```bash
-   # 在 MuseTalk 推理代码中保存中间帧到 /tmp/debug_frames/
-   # 对比输入帧 vs 推理后帧 vs 原视频帧
-   ```
-
-## 建议的下一步
-
-1. 先在 RTX2060 上 `cat ~/projects/MuseTalk/musetalk_server.py` 确认实际部署的代码
-2. 确认 MuseTalk 是否真的被调用（看日志有没有 "使用示例推理逻辑" 的 warning）
-3. 如果确认是 stub，需要把真正的 MuseTalk 推理逻辑集成到 `_run_inference()`
-4. 同时修复音频预处理（P0），这个无论推理代码如何都是必须的

deploy/gpu_worker/gpu_worker.py

@@ -108,15 +108,23 @@ def _check_musetalk_health() -> tuple[bool, dict]:
         return False, {"error": str(exc)}
 
 
-def _register(task_id: Optional[str] = None) -> bool:
+def _register(task_id: Optional[str] = None) -> tuple[bool, bool]:
     """向服务端注册 / 心跳，附带 GPU 信息。
 
     推理期间的心跳线程传 task_id：服务端会同步刷新该 processing 任务的
-    last_heartbeat_at，防止长推理被误判超时回收。
+    last_heartbeat_at，防止长推理被误判超时回收。同时服务端会检查该任务
+    是否已被用户取消，若是则返回 cancel_task=True。
+
+    返回 (ok, cancel_task)。
     """
     ok, info = _check_musetalk_health()
-    free_vram = int(info.get("free_vram_mb", 0) or 0) if isinstance(info, dict) else 0
-    gpu_name = info.get("gpu_name", "") if isinstance(info, dict) else ""
+    if isinstance(info, dict):
+        gpu_info = info.get("gpu", info)
+        free_vram = int(gpu_info.get("free_vram_mb", gpu_info.get("memory_free_mb", 0)) or 0)
+        gpu_name = gpu_info.get("gpu_name", info.get("gpu_name", ""))
+    else:
+        free_vram = 0
+        gpu_name = ""
     if not gpu_name:
         # 尝试在 Windows 上读 nvidia-smi
         gpu_name = _probe_gpu_name()
@@ -137,12 +145,14 @@ def _register(task_id: Optional[str] = None) -> bool:
             timeout=15,
         )
         if r.status_code == 200:
-            return True
+            resp_body = r.json()
+            cancel_task = resp_body.get("cancel_task", False)
+            return True, cancel_task
         logger.error("注册/心跳失败: HTTP %d body=%s", r.status_code, r.text[:300])
-        return False
+        return False, False
     except Exception as exc:
         logger.error("注册/心跳异常: %s", exc)
-        return False
+        return False, False
 
 
 def _probe_gpu_name() -> str:
@@ -324,6 +334,9 @@ class TaskHeartbeat(threading.Thread):
     期间无法发送，服务端会因任务 last_heartbeat_at 停滞而误判超时回退 pending。
     本线程每 task_heartbeat_interval 秒（默认 30s）POST /gpu/register 并
     携带当前 task_id，让服务端持续续期任务心跳；任务处理结束 stop()。
+
+    同时检测服务端返回的 cancel_task 信号：若为 True，说明用户已取消任务，
+    立即调用 _cancel_musetalk() 终止本地推理，并设置 cancelled 标志供主流程检查。
     """
 
     def __init__(self, task_id: str, interval: float):
@@ -331,13 +344,21 @@ class TaskHeartbeat(threading.Thread):
         self.task_id = task_id
         self.interval = max(5.0, interval)
         self._stop_event = threading.Event()
+        self.cancelled = False  # 外部可读的取消标志
 
     def run(self) -> None:
         # 先立即发一次，再按间隔循环（首次心跳失败不影响主流程）
         while not self._stop_event.is_set():
             try:
-                if _register(self.task_id):
+                ok, cancel_task = _register(self.task_id)
+                if ok:
                     logger.debug("任务 %s 心跳已发送", self.task_id)
+                if cancel_task:
+                    logger.warning("任务 %s 已被用户取消，正在终止本地推理...", self.task_id)
+                    self.cancelled = True
+                    _cancel_musetalk()
+                    self._stop_event.set()
+                    return
             except Exception as exc:  # noqa: BLE001
                 logger.warning("任务 %s 心跳异常（忽略）: %s", self.task_id, exc)
             self._stop_event.wait(self.interval)
@@ -364,9 +385,17 @@ def _handle_task(task: dict) -> None:
             if not _download(task["video_url"], video_path):
                 _report_result(task_id, False, 0.0, "下载人物视频失败")
                 return
+            if hb.cancelled:
+                logger.info("任务 %s 在下载阶段被用户取消", task_id)
+                _report_result(task_id, False, 0.0, "用户取消任务")
+                return
             if not _download(task["audio_url"], audio_path):
                 _report_result(task_id, False, 0.0, "下载驱动音频失败")
                 return
+            if hb.cancelled:
+                logger.info("任务 %s 在下载阶段被用户取消", task_id)
+                _report_result(task_id, False, 0.0, "用户取消任务")
+                return
 
             # 2. 输入时长前置校验：短视频 MuseTalk 会 division by zero，
             #    直接上报 failed，不浪费 GPU 时间。ffprobe 不可用/读失败（0.0）
@@ -387,12 +416,20 @@ def _handle_task(task: dict) -> None:
             err = ""
             retryable = False
             for attempt in range(Config.task_max_retry + 1):
+                if hb.cancelled:
+                    logger.info("任务 %s 在推理前被用户取消", task_id)
+                    _report_result(task_id, False, 0.0, "用户取消任务")
+                    return
                 if attempt > 0:
                     logger.info("任务 %s 第 %d 次重试（瞬时错误）...", task_id, attempt + 1)
                     time.sleep(2)
                 success, duration, err, retryable = _call_musetalk(video_path, audio_path, out_path)
                 if success or not retryable:
                     break
+            if hb.cancelled:
+                logger.info("任务 %s 被用户取消（推理已终止）", task_id)
+                _report_result(task_id, False, 0.0, "用户取消任务")
+                return
             if not success:
                 logger.error("任务 %s 推理失败: %s", task_id, err)
                 _report_result(task_id, False, 0.0, err)
@@ -462,7 +499,8 @@ def main() -> int:
             # 心跳
             now = time.time()
             if now - last_heartbeat >= Config.heartbeat_interval:
-                if _register():
+                ok, _ = _register()
+                if ok:
                     last_heartbeat = now
 
             # 轮询任务

deploy/gpu_worker/musetalk_server.py

@@ -2,6 +2,15 @@
 
 部署在 RTX2060 本地，接收 gpu_worker.py 的推理请求，调用 MuseTalk 生成口型同步视频。
 
+Bug 修复（2026-09-21）：
+  - Bug1: 60fps降帧逻辑 — 输入视频 >30fps 时先降帧至 25fps 推理，直接输出 25fps 结果
+    （MCI 运动补偿插帧已移除，CPU密集且口型场景 25fps 足够）
+  - Bug2: 超时终止机制 — 推理线程改为 daemon + abort_event 机制，超时时 set event
+    让推理循环检测退出，同时 kill 所有活跃 ffmpeg 子进程，等线程退出后再释放锁和清理目录
+  - Bug3: /health 接口增加 gfpgan_loaded 和 gfpgan_load_error 字段
+  - Bonus: 动态超时计算（基础120s + 帧数*0.15s，上限1800s）
+  - Bonus: GFPGAN 增强异常时 log warning 而非静默跳过
+
 #2000 关键修复：
   - 集成真实 MuseTalk 推理（替换原有 stub 代码）
   - 音频预处理：22050Hz MP3 → 16kHz mono 16bit WAV（MuseTalk 要求）
@@ -87,12 +96,23 @@ class Config:
     use_float16: bool = _env("MUSE_USE_FLOAT16", "1") == "1"
     # 推理批次大小（RTX2060 6G 显存建议 4-8）
     batch_size: int = int(_env("MUSE_BATCH_SIZE", "8"))
+    # GFPGAN 人脸超分增强（提升生成人脸清晰度，+~170MB VRAM, +60ms/帧）
+    use_gfpgan: bool = _env("MUSE_USE_GFPGAN", "1") == "1"
 
 
 # ── 全局状态 ──────────────────────────────────────────────────────────
 inference_lock = threading.Lock()
 current_task: dict = {"task_id": None, "process": None, "start_time": 0.0}
 shutdown_event = threading.Event()
+# 当前推理线程的终止信号和线程引用
+current_abort_event: Optional[threading.Event] = None
+current_thread: Optional[threading.Thread] = None
+# 全局追踪正在运行的 ffmpeg 子进程（用于超时终止）
+_active_ffmpeg_procs: list[subprocess.Popen] = []
+_active_ffmpeg_lock = threading.Lock()
+# GFPGAN 加载状态（供 /health 接口查询）
+_gfpgan_loaded = False
+_gfpgan_load_error: Optional[str] = None
 
 # ── MuseTalk 模型懒加载 ─────────────────────────────────────────────
 _muse_models = None
@@ -372,21 +392,33 @@ def _check_file_size(file, max_mb: int, label: str) -> Optional[str]:
 
 
 def _run_ffmpeg(cmd: list, timeout: float = 120) -> subprocess.CompletedProcess:
-    """运行 ffmpeg 命令，检查返回码和超时."""
+    """运行 ffmpeg 命令，检查返回码和超时。进程会被注册到全局列表以便外部终止."""
+    proc = None
     try:
-        result = subprocess.run(
+        proc = subprocess.Popen(
             cmd,
             stdout=subprocess.PIPE,
             stderr=subprocess.PIPE,
-            timeout=timeout,
-            check=True,
         )
-        return result
-    except subprocess.CalledProcessError as exc:
-        stderr = exc.stderr.decode(errors="ignore") if exc.stderr else ""
-        raise RuntimeError(f"ffmpeg 失败 (code={exc.returncode}): {stderr[:500]}") from exc
-    except subprocess.TimeoutExpired as exc:
-        raise RuntimeError(f"ffmpeg 超时（>{timeout}s）") from exc
+        with _active_ffmpeg_lock:
+            _active_ffmpeg_procs.append(proc)
+        try:
+            stdout, stderr = proc.communicate(timeout=timeout)
+        except subprocess.TimeoutExpired:
+            proc.kill()
+            proc.wait(timeout=5)
+            raise RuntimeError(f"ffmpeg 超时（>{timeout}s）")
+        if proc.returncode != 0:
+            stderr_text = stderr.decode(errors="ignore") if stderr else ""
+            raise RuntimeError(f"ffmpeg 失败 (code={proc.returncode}): {stderr_text[:500]}")
+        return subprocess.CompletedProcess(cmd, proc.returncode, stdout, stderr)
+    finally:
+        if proc is not None:
+            with _active_ffmpeg_lock:
+                try:
+                    _active_ffmpeg_procs.remove(proc)
+                except ValueError:
+                    pass
 
 
 # ── MuseTalk 模型加载 ────────────────────────────────────────────────
@@ -399,7 +431,7 @@ def _load_musetalk_models():
     加载到 GPU 后转为 FP16（如果配置开启）以节省显存。
     RTX2060 6G 显存，FP16 大约需要 3-4GB。
     """
-    global _muse_models, _muse_models_loaded, _muse_load_error
+    global _muse_models, _muse_models_loaded, _muse_load_error, _gfpgan_loaded, _gfpgan_load_error
 
     if _muse_models_loaded:
         return _muse_models
@@ -477,6 +509,43 @@ def _load_musetalk_models():
             audio_processor = AudioProcessor()
             face_parsing = FaceParsing()
 
+            # 加载 GFPGAN 人脸超分模型（FP16，仅 ~170MB VRAM）
+            gfpgan_model = None
+            if Config.use_gfpgan:
+                try:
+                    from gfpgan.archs.gfpganv1_clean_arch import GFPGANv1Clean
+                    gfpgan_path = muse_dir / "models" / "GFPGAN" / "GFPGANv1.4.pth"
+                    if gfpgan_path.exists():
+                        logger.info("加载 GFPGANv1.4 人脸超分模型: %s", gfpgan_path)
+                        gfpgan_ckpt = torch.load(str(gfpgan_path), map_location="cpu")
+                        gfpgan_model = GFPGANv1Clean(
+                            out_size=512, num_style_feat=512, channel_multiplier=2,
+                            decoder_load_path=None, fix_decoder=False, num_mlp=8,
+                            input_is_latent=True, different_w=True, narrow=1, sft_half=True,
+                        )
+                        gfpgan_key = "params_ema" if "params_ema" in gfpgan_ckpt else "params"
+                        gfpgan_model.load_state_dict(gfpgan_ckpt[gfpgan_key], strict=True)
+                        gfpgan_model.eval()
+                        # GFPGAN 始终使用 FP32 推理，避免 FP16 色偏导致紫/灰色块
+                        gfpgan_model = gfpgan_model.to(device)
+                        del gfpgan_ckpt
+                        _gfpgan_loaded = True
+                        _gfpgan_load_error = None
+                        logger.info("GFPGAN 加载完成 (FP32，避免色偏)")
+                    else:
+                        _gfpgan_loaded = False
+                        _gfpgan_load_error = f"模型文件不存在: {gfpgan_path}"
+                        logger.warning("GFPGAN 模型不存在: %s，跳过人脸增强", gfpgan_path)
+                except Exception as e:
+                    _gfpgan_loaded = False
+                    _gfpgan_load_error = str(e)
+                    logger.warning("GFPGAN 加载失败，跳过人脸增强: %s", e)
+                    gfpgan_model = None
+            else:
+                _gfpgan_loaded = False
+                _gfpgan_load_error = "已通过环境变量禁用 (MUSE_USE_GFPGAN=0)"
+                logger.info("GFPGAN 已禁用 (MUSE_USE_GFPGAN=0)")
+
             _muse_models = {
                 "vae": vae,
                 "unet": unet,
@@ -484,6 +553,7 @@ def _load_musetalk_models():
                 "timesteps": timesteps,
                 "audio_processor": audio_processor,
                 "face_parsing": face_parsing,
+                "gfpgan": gfpgan_model,
                 "device": device,
                 "model_version": model_version,
             }
@@ -529,6 +599,7 @@ def _run_inference(
     audio_path: Path,
     output_path: Path,
     bbox_shift: int = 0,
+    abort_event: threading.Event = None,
 ) -> None:
     """执行 MuseTalk 真实推理.
 
@@ -552,9 +623,38 @@ def _run_inference(
     import torch
     from tqdm import tqdm
 
-    from musetalk.utils.preprocessing import get_landmark_and_bbox, read_imgs
-    from musetalk.utils.utils import datagen
+    from musetalk.utils.preprocessing import get_landmark_and_bbox as _orig_get_landmark_and_bbox
     from musetalk.utils.blending import get_image
+    import tempfile as _tempfile, math as _math, shutil as _shutil
+    from einops import rearrange as _rearrange
+
+    # read_imgs: 读取视频帧（支持视频文件路径），返回 numpy BGR 帧列表
+    def read_imgs(path):
+        import cv2 as _cv2
+        cap = _cv2.VideoCapture(str(path))
+        frames = []
+        while True:
+            ret, frame = cap.read()
+            if not ret:
+                break
+            frames.append(frame)
+        cap.release()
+        return frames
+
+    # get_landmark_and_bbox 适配：旧版签名(img_list, upperbondrange=0)，且 img_list 是文件路径列表
+    def get_landmark_and_bbox(frames, vid_pts=0, bbox_shift=0):
+        import cv2 as _cv2
+        _tmpdir = _tempfile.mkdtemp(prefix="muse_frames_")
+        frame_paths = []
+        for _i, _frm in enumerate(frames):
+            _fp = f"{_tmpdir}/{_i:08d}.png"
+            _cv2.imwrite(_fp, _frm)
+            frame_paths.append(_fp)
+        coords_list, _ = _orig_get_landmark_and_bbox(frame_paths, upperbondrange=bbox_shift)
+        _shutil.rmtree(_tmpdir, ignore_errors=True)
+        _sentinel = object()
+        coords_list = [c if c is not None else _sentinel for c in coords_list]
+        return coords_list, _sentinel
 
     # 加载模型（首次调用时加载，后续复用）
     models = _load_musetalk_models()
@@ -566,14 +666,89 @@ def _run_inference(
     device = models["device"]
     model_version = models["model_version"]
 
+    # 给旧版 AudioProcessor 动态添加 feature2chunks 方法
+    import types as _types
+    def _feature2chunks(self, feature_array, fps=25, weight_dtype=None,
+                        batch_size=8, audio_padding_length_left=2,
+                        audio_padding_length_right=2):
+        import torch
+        sr = 16000
+        audio_fps = 50
+        chunk_len = 2 * (audio_padding_length_left + audio_padding_length_right + 1)
+        whisper_idx_multiplier = audio_fps / fps
+        num_frames = int(_math.floor((len(feature_array) / sr) * fps))
+        actual_length = int(_math.floor((len(feature_array) / sr) * audio_fps))
+        inputs = self.feature_extractor(
+            feature_array, return_tensors="pt", sampling_rate=sr
+        ).input_features.to(device)
+        if weight_dtype is not None:
+            inputs = inputs.to(dtype=weight_dtype)
+        global _whisper_enc_model
+        if "_whisper_enc_model" not in globals() or _whisper_enc_model is None:
+            from transformers import WhisperModel
+            _wp = str(Path(Config.muse_dir) / "models" / "whisper")
+            _whisper_enc_model = WhisperModel.from_pretrained(_wp).to(device)
+            _whisper_enc_model.eval()
+            if Config.use_float16:
+                _whisper_enc_model = _whisper_enc_model.half()
+        with torch.no_grad():
+            _af = _whisper_enc_model.encoder(inputs, output_hidden_states=True).hidden_states
+            _af = torch.stack(_af, dim=2)
+        _af = _af[0, :actual_length, ...]
+        _pn = int(_math.ceil(whisper_idx_multiplier))
+        _af = torch.cat([
+            torch.zeros_like(_af[:_pn * audio_padding_length_left]),
+            _af,
+            torch.zeros_like(_af[:_pn * 3 * audio_padding_length_right]),
+        ], dim=0)
+        _all = []
+        for _fi in range(num_frames):
+            _ai = int(_math.floor(_fi * whisper_idx_multiplier))
+            _clip = _af[_ai:_ai + chunk_len]
+            if _clip.shape[0] < chunk_len:
+                _pad = torch.zeros(chunk_len - _clip.shape[0], *_clip.shape[1:],
+                                    device=device, dtype=_clip.dtype)
+                _clip = torch.cat([_clip, _pad], dim=0)
+            _all.append(_clip)
+        _prompts = torch.stack(_all, dim=0)
+        _prompts = _rearrange(_prompts, "b c h w -> b (c h) w")
+        return _prompts
+    audio_processor.feature2chunks = _types.MethodType(_feature2chunks, audio_processor)
+
     fps = _get_video_fps(video_path)
+    original_fps = fps
     audio_duration = _get_media_duration(audio_path)
     video_duration = _get_media_duration(video_path)
     logger.info(
-        "MuseTalk 推理开始: video=%.2fs, audio=%.2fs, fps=%.1f, bbox_shift=%d",
+        "MuseTalk 推理开始: video=%.2fs, audio=%.2fs, input_fps=%.1f, bbox_shift=%d",
         video_duration, audio_duration, fps, bbox_shift,
     )
 
+    # ── Step 0: 高帧率视频降帧（>30fps → 25fps 推理，推理后再插帧回原帧率）──
+    inference_fps = fps
+    video_downsampled = False
+    if fps > 30:
+        inference_fps = 25.0
+        downsampled_video = video_path.parent / "input_25fps.mp4"
+        logger.info("检测到高帧率视频 %.1f fps，降帧至 %.1f fps 进行推理", fps, inference_fps)
+        try:
+            _run_ffmpeg([
+                "ffmpeg", "-y", "-v", "warning",
+                "-i", str(video_path),
+                "-r", str(inference_fps),
+                "-c:v", "libx264", "-preset", "veryfast",
+                "-crf", "18", "-pix_fmt", "yuv420p",
+                str(downsampled_video),
+            ], timeout=120)
+            video_path = downsampled_video
+            video_downsampled = True
+            # 更新帧数和时长
+            total_input_frames_approx = int(audio_duration * inference_fps)
+            logger.info("降帧完成: input_fps=%.1f → inference_fps=%.1f", original_fps, inference_fps)
+        except Exception as e:
+            logger.warning("视频降帧失败，使用原始帧率 %.1f fps 推理: %s", fps, e)
+            inference_fps = fps
+
     # ── Step 1: 音频预处理（关键修复：22050Hz MP3 → 16kHz mono WAV）──
     audio_wav_path = video_path.parent / "audio_16k_mono.wav"
     _preprocess_audio(audio_path, audio_wav_path, target_sr=16000)
@@ -592,7 +767,7 @@ def _run_inference(
     logger.info("人脸检测完成，有效 bbox: %d/%d", sum(1 for c in coord_list if c is not coord_placeholder), total_frames)
 
     # 使用 mirror indexing 循环帧和坐标（避免硬切跳变）
-    num_output_frames = int(audio_duration * fps)
+    num_output_frames = int(audio_duration * inference_fps)
     if num_output_frames <= 0:
         num_output_frames = total_frames
 
@@ -602,138 +777,238 @@ def _run_inference(
     audio_array, _ = librosa.load(str(audio_wav_path), sr=16000, mono=True)
     whisper_features = audio_processor.feature2chunks(
         feature_array=audio_array,
-        fps=fps,
+        fps=inference_fps,
         weight_dtype=(torch.float16 if Config.use_float16 else torch.float32),
         batch_size=Config.batch_size,
     )
+    if isinstance(whisper_features, torch.Tensor):
+        whisper_features = whisper_features.detach().cpu()
+    torch.cuda.empty_cache()
     logger.info("音频特征提取完成: %d 个 chunk", len(whisper_features))
 
-    # ── Step 5: 视频帧 VAE 编码为 latent ──
+    # ── Step 5: 逐帧裁剪人脸并编码为 8ch latent (masked+ref) ──
+    face_parsing = models.get("face_parsing", None)
     input_latent_list = []
+    valid_frame_indices = []  # 记录成功编码的帧索引（跳过无脸帧）
+
     with torch.no_grad():
-        for frame in input_frames:
-            frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
-            frame_resized = cv2.resize(frame_rgb, (256, 256))
-            frame_tensor = torch.from_numpy(frame_resized).float() / 127.5 - 1.0
-            frame_tensor = frame_tensor.permute(2, 0, 1).unsqueeze(0).to(device)
-            if Config.use_float16:
-                frame_tensor = frame_tensor.half()
-            latent = vae.encode_latents(frame_tensor)
-            input_latent_list.append(latent)
-    logger.info("视频帧 VAE 编码完成: %d 个 latent", len(input_latent_list))
+        for idx, (frame, bbox) in enumerate(zip(input_frames, coord_list)):
+            if bbox is coord_placeholder:
+                continue
+            x1, y1, x2, y2 = bbox
+            # v1.5 额外扩展下边界（下巴区域），与 Step 7 保持一致
+            extra_y2 = 10 if model_version == "v15" else 0
+            y2_eff = min(y2 + extra_y2, frame.shape[0])
+            if y2_eff <= y1 or x2 <= x1:
+                continue
+            # 裁剪人脸区域 → resize 256×256
+            crop = frame[y1:y2_eff, x1:x2]
+            if crop.size == 0:
+                continue
+            crop_rgb = cv2.cvtColor(crop, cv2.COLOR_BGR2RGB)
+            crop_resized = cv2.resize(crop_rgb, (256, 256), interpolation=cv2.INTER_LANCZOS4)
+            # 使用 VAE 的 get_latents_for_unet 得到 8 通道输入
+            # get_latents_for_unet 内部: preprocess(half_mask=True) encode + preprocess(half_mask=False) encode → cat → [1,8,32,32]
+            latents = vae.get_latents_for_unet(crop_resized).detach().cpu()
+            input_latent_list.append(latents)
+            # 保存此帧的实际bbox(含extra_y2)和原帧索引供 Step 7 使用
+            valid_frame_indices.append((idx, x1, y1, x2, y2_eff))
 
-    # ── Step 6: 批量推理 ──
-    result_frames = []
-    total_batches = (num_output_frames + Config.batch_size - 1) // Config.batch_size
+    # 构建循环列表：正序+倒序，实现旧版的平滑首尾帧循环
+    frame_list_cycle = input_frames + input_frames[::-1]
+    coord_cycle = []
+    for _i, _x1, _y1, _x2, _y2 in valid_frame_indices:
+        coord_cycle.append((_x1, _y1, _x2, _y2))
+    coord_cycle = coord_cycle + coord_cycle[::-1]
+    latent_cycle = input_latent_list + input_latent_list[::-1]
+    valid_cycle = valid_frame_indices + [(i, x1, y1, x2, y2) for (i, x1, y1, x2, y2) in reversed(valid_frame_indices)]
 
-    for batch_idx in tqdm(range(total_batches), desc="MuseTalk 推理"):
-        # 构建 whisper batch
-        whisper_batch = whisper_features[
-            batch_idx * Config.batch_size : (batch_idx + 1) * Config.batch_size
-        ]
+    torch.cuda.empty_cache()
+    logger.info("人脸裁剪+VAE 编码完成: %d 个有效latent", len(input_latent_list))
+
+    # ── Step 6: 批量推理（仿旧版 datagen 循环）──
+    res_frame_list = []
+    video_num = len(whisper_features)
+    bs = min(Config.batch_size, 8)  # RTX3060 12G 显存，FP16+GFPGAN batch=8 约用 7-8GB，留足余量
+    total_batches = (video_num + bs - 1) // bs
+
+    for bi in tqdm(range(total_batches), desc="MuseTalk 推理"):
+        # 检查是否收到终止信号
+        if abort_event is not None and abort_event.is_set():
+            logger.warning("收到终止信号，中止推理 (batch %d/%d)", bi, total_batches)
+            raise InterruptedError("推理被外部终止")
+        whisper_batch = whisper_features[bi*bs:(bi+1)*bs]
         if len(whisper_batch) == 0:
             break
+        # 对应 latent 索引（循环取 latent_cycle）
+        latent_batch_parts = []
+        for j in range(len(whisper_batch)):
+            global_idx = bi*bs + j
+            lat_idx = global_idx % len(latent_cycle)
+            latent_batch_parts.append(latent_cycle[lat_idx])
 
-        # 构建 latent batch（使用 mirror indexing 循环）
-        latent_indices = []
-        for i in range(len(whisper_batch)):
-            global_idx = batch_idx * Config.batch_size + i
-            frame_idx = _mirror_index(total_frames, global_idx)
-            latent_indices.append(frame_idx)
-
-        latent_batch = torch.cat(
-            [input_latent_list[idx] for idx in latent_indices], dim=0
-        )
-        latent_batch = latent_batch.to(device)
+        # whisper_batch 是 [bs,50,384] tensor slice (feature2chunks 已返回 stacked tensor)
+        if isinstance(whisper_batch, list):
+            whisper_batch_t = torch.stack(whisper_batch).to(device)
+        else:
+            whisper_batch_t = whisper_batch.to(device)
+        latent_batch_t = torch.cat(latent_batch_parts, dim=0).to(device)
         if Config.use_float16:
-            latent_batch = latent_batch.half()
+            latent_batch_t = latent_batch_t.to(dtype=unet.model.dtype)
+            whisper_batch_t = whisper_batch_t.to(dtype=unet.model.dtype)
 
-        # 位置编码
-        audio_feature_batch = pe(whisper_batch)
+        audio_feature_batch = pe(whisper_batch_t)
 
-        # UNet 推理
         with torch.no_grad():
-            latent_batch = latent_batch.to(dtype=unet.model.dtype)
             pred_latents = unet.model(
-                latent_batch,
+                latent_batch_t,
                 timesteps,
                 encoder_hidden_states=audio_feature_batch,
             ).sample
 
-        # VAE 解码
         recon_frames = vae.decode_latents(pred_latents)
-        result_frames.extend(recon_frames)
+        for rf in recon_frames:
+            res_frame_list.append(rf)
+        del pred_latents, recon_frames, latent_batch_t, whisper_batch_t
+        if "audio_feature_batch" in dir():
+            try: del audio_feature_batch
+            except: pass
+        torch.cuda.empty_cache()
 
-    logger.info("推理完成，生成 %d 帧", len(result_frames))
+    logger.info("推理完成，生成 %d 帧", len(res_frame_list))
 
-    # ── Step 7: 合成最终帧并写视频 ──
-    output_frames_dir = video_path.parent / "output_frames"
-    output_frames_dir.mkdir(parents=True, exist_ok=True)
+    # ── Step 7: 合成最终帧 → ffmpeg pipe 编码（零磁盘IO） ──
+    gfpgan_enhancer = models.get("gfpgan")
+    silent_video_path = video_path.parent / "silent_output.mp4"
+    frame_h, frame_w = frame_list_cycle[0].shape[:2]
 
-    for i, res_frame in enumerate(tqdm(result_frames[:num_output_frames], desc="合成帧")):
-        orig_idx = _mirror_index(total_frames, i)
-        ori_frame = copy.deepcopy(input_frames[orig_idx])
-        bbox = coord_list[orig_idx] if orig_idx < len(coord_list) else coord_placeholder
+    # 启动 ffmpeg：stdin 接收 raw BGR24 帧，直接编码 H.264（省去PNG落盘+回读）
+    _ff_cmd = [
+        "ffmpeg", "-y", "-v", "warning",
+        "-f", "rawvideo", "-pix_fmt", "bgr24",
+        "-s", f"{frame_w}x{frame_h}", "-r", str(inference_fps),
+        "-i", "-",
+        "-vcodec", "libx264", "-preset", "veryfast",
+        "-vf", "format=yuv420p", "-crf", "18",
+        str(silent_video_path),
+    ]
+    import subprocess as _sp
+    _ff_proc = _sp.Popen(_ff_cmd, stdin=_sp.PIPE, stdout=_sp.DEVNULL, stderr=_sp.PIPE)
+    # 注册 ffmpeg 进程到全局列表，以便超时终止
+    with _active_ffmpeg_lock:
+        _active_ffmpeg_procs.append(_ff_proc)
+
+    n_out = min(len(res_frame_list), num_output_frames)
+    try:
+        for i in tqdm(range(n_out), desc="合成帧"):
+            # 检查终止信号
+            if abort_event is not None and abort_event.is_set():
+                logger.warning("收到终止信号，中止合成 (frame %d/%d)", i, n_out)
+                raise InterruptedError("推理被外部终止")
+            cyc_i = i % len(coord_cycle)
+            x1, y1, x2, y2 = coord_cycle[cyc_i]
+            ori_frame = copy.deepcopy(frame_list_cycle[cyc_i])
+            res_frame = res_frame_list[i]
 
-        if bbox is coord_placeholder:
-            # 没有检测到人脸的帧，保持原样
-            combined = ori_frame
-        else:
-            x1, y1, x2, y2 = bbox
-            if model_version == "v15":
-                # v1.5 额外扩展下边界（下巴区域）
-                y2 = min(y2 + 10, ori_frame.shape[0])
             try:
                 res_frame_resized = cv2.resize(
-                    res_frame.astype(np.uint8), (x2 - x1, y2 - y1)
+                    res_frame.astype(np.uint8), (x2-x1, y2-y1),
+                    interpolation=cv2.INTER_LANCZOS4
                 )
             except Exception:
-                combined = ori_frame
-                cv2.imwrite(
-                    str(output_frames_dir / f"{i:08d}.png"), combined
-                )
+                _ff_proc.stdin.write(ori_frame.tobytes())
                 continue
 
-            # 使用 face parsing 做边缘融合
-            combined = get_image(
-                ori_frame,
-                res_frame_resized,
-                [x1, y1, x2, y2],
-            )
+            # GFPGAN 人脸超分增强（FP32 + 色彩校正，避免 FP16 色偏）
+            if gfpgan_enhancer is not None:
+                try:
+                    _fh, _fw = res_frame_resized.shape[:2]
+                    _face_up = cv2.resize(res_frame_resized, (512, 512),
+                                          interpolation=cv2.INTER_LANCZOS4)
+                    # 保存原始人脸区域用于色彩校正
+                    _face_original_bgr = _face_up.copy()
+                    _face_rgb = cv2.cvtColor(_face_up, cv2.COLOR_BGR2RGB).astype(np.float32) / 255.0
+                    _face_t = torch.from_numpy(_face_rgb.transpose(2,0,1)).unsqueeze(0)
+                    # GFPGAN 始终 FP32，避免 FP16 精度导致色偏；归一化到 [-1, 1]
+                    _face_t = ((_face_t - 0.5) / 0.5).to(device)
+                    with torch.no_grad():
+                        _out = gfpgan_enhancer(_face_t, return_rgb=False, weight=0.35)[0]
+                    _out = _out.squeeze(0).float().cpu().clamp_(-1,1)
+                    _out = ((_out + 1)/2*255).numpy().transpose(1,2,0)
+                    _out_bgr = cv2.cvtColor(_out.astype(np.uint8), cv2.COLOR_RGB2BGR)
 
-        cv2.imwrite(str(output_frames_dir / f"{i:08d}.png"), combined)
+                    # 色彩校正：将增强结果的均值/标准差对齐到原始人脸，消除色调偏移
+                    _orig_mean = _face_original_bgr.mean(axis=(0, 1))
+                    _orig_std = _face_original_bgr.std(axis=(0, 1)) + 1e-6
+                    _enh_mean = _out_bgr.mean(axis=(0, 1))
+                    _enh_std = _out_bgr.std(axis=(0, 1)) + 1e-6
+                    _out_bgr = ((_out_bgr.astype(np.float32) - _enh_mean) * (_orig_std / _enh_std) + _orig_mean)
+                    _out_bgr = np.clip(_out_bgr, 0, 255).astype(np.uint8)
 
-    # 帧序列 → 无声视频
-    silent_video_path = video_path.parent / "silent_output.mp4"
-    _run_ffmpeg(
-        [
-            "ffmpeg", "-y", "-v", "warning",
-            "-r", str(fps),
-            "-f", "image2",
-            "-i", str(output_frames_dir / "%08d.png"),
-            "-vcodec", "libx264",
-            "-vf", "format=yuv420p",
-            "-crf", "18",
-            str(silent_video_path),
-        ],
-        timeout=300,
-    )
+                    res_frame_resized = cv2.resize(_out_bgr, (_fw, _fh),
+                                                    interpolation=cv2.INTER_LANCZOS4)
+                    del _face_t, _out, _out_bgr, _face_original_bgr
+                except Exception as _gfpgan_err:
+                    logger.warning("GFPGAN 增强失败（帧 %d），使用原图: %s", i, _gfpgan_err)
 
-    # 将无声视频复制到输出路径
-    shutil.copy2(str(silent_video_path), str(output_path))
+            # face parsing 融合
+            try:
+                if face_parsing is not None:
+                    combined = get_image(ori_frame, res_frame_resized,
+                                         [x1,y1,x2,y2], fp=face_parsing)
+                else:
+                    combined = get_image(ori_frame, res_frame_resized, [x1,y1,x2,y2])
+            except Exception:
+                combined = ori_frame.copy()
+                try: combined[y1:y2, x1:x2] = res_frame_resized
+                except Exception: combined = ori_frame
+
+            _ff_proc.stdin.write(combined.tobytes())
+
+        _ff_proc.stdin.close()
+        _ff_ret = _ff_proc.wait(timeout=120)
+        # 从全局列表中移除已完成的 ffmpeg 进程
+        with _active_ffmpeg_lock:
+            try:
+                _active_ffmpeg_procs.remove(_ff_proc)
+            except ValueError:
+                pass
+        if _ff_ret != 0:
+            _ff_err = _ff_proc.stderr.read().decode(errors="ignore") if _ff_proc.stderr else ""
+            raise RuntimeError(f"ffmpeg编码失败(exit={_ff_ret}): {_ff_err[-300:]}")
+    except Exception:
+        try: _ff_proc.kill()
+        except Exception: pass
+        with _active_ffmpeg_lock:
+            try:
+                _active_ffmpeg_procs.remove(_ff_proc)
+            except ValueError:
+                pass
+        raise
+
+    # ── Step 8: 高帧率视频处理 ──
+    # 对口型数字人视频，25fps 完全够用。
+    # MCI 运动补偿插帧极其耗时（343帧>6分钟），已移除。
+    # 直接输出 25fps 推理结果，后续封装音频后播放器自动适配帧率。
+    final_video = silent_video_path
+    if video_downsampled:
+        logger.info("输入视频 %.1f fps，降帧至 %.1f fps 推理后直接输出（不做插帧还原）", original_fps, inference_fps)
+
+    _mux_video_with_audio(final_video, audio_path, output_path)
 
-    # 清理中间文件
     try:
-        shutil.rmtree(str(output_frames_dir))
+        torch.cuda.empty_cache()
         if audio_wav_path.exists():
             audio_wav_path.unlink()
-        if silent_video_path.exists() and str(silent_video_path) != str(output_path):
-            silent_video_path.unlink()
+        # 清理中间文件
+        for _tmp in [silent_video_path] + ([video_path] if video_downsampled else []):
+            if _tmp.exists() and str(_tmp) != str(output_path):
+                _tmp.unlink()
     except Exception as e:
         logger.warning("清理中间文件失败: %s", e)
 
-    logger.info("MuseTalk 推理完成: output=%s, duration=%.2fs",
-                output_path.name, _get_media_duration(output_path))
+    logger.info("MuseTalk 推理完成: output=%s, duration=%.2fs, input_fps=%.1f, inference_fps=%.1f",
+                output_path.name, _get_media_duration(output_path), original_fps, inference_fps)
 
 
 # ── 路由 ──────────────────────────────────────────────────────────────
@@ -755,6 +1030,8 @@ def health():
             "current_task": task_info,
             "musetalk_loaded": _muse_models_loaded,
             "musetalk_load_error": str(_muse_load_error) if _muse_load_error else None,
+            "gfpgan_loaded": _gfpgan_loaded,
+            "gfpgan_load_error": _gfpgan_load_error,
             "timestamp": time.time(),
         }
     )
@@ -813,24 +1090,52 @@ def inference():
         current_task["start_time"] = time.time()
         current_task["process"] = "inference_thread"  # 标记为运行中
 
-        # 在线程中运行推理（支持超时）
+        # 在线程中运行推理（支持超时 + 终止信号）
         result_container = {"error": None}
+        abort_event = threading.Event()
+        global current_abort_event, current_thread
+        current_abort_event = abort_event
 
         def inference_thread():
             try:
-                _run_inference(video_path, audio_path, output_path, bbox_shift=bbox_shift)
+                _run_inference(video_path, audio_path, output_path, bbox_shift=bbox_shift, abort_event=abort_event)
             except Exception as exc:
                 logger.exception("推理异常: %s", exc)
                 result_container["error"] = str(exc)
 
-        thread = threading.Thread(target=inference_thread)
+        thread = threading.Thread(target=inference_thread, daemon=True)
+        current_thread = thread
         thread.start()
-        thread.join(timeout=Config.inference_timeout)
+
+        # 动态超时：基础 120s + 预估帧数×0.15s，上限 1800s
+        video_fps = _get_video_fps(video_path)
+        video_duration = _get_media_duration(video_path)
+        estimated_frames = int(video_duration * video_fps) if video_duration > 0 else 500
+        dynamic_timeout = min(max(Config.inference_timeout, 120 + estimated_frames * 0.15), 1800)
+
+        thread.join(timeout=dynamic_timeout)
 
         if thread.is_alive():
-            # 超时，终止
-            logger.error("推理超时 (>%ds)，终止任务 %s", Config.inference_timeout, task_id)
-            return jsonify({"error": f"推理超时（>{Config.inference_timeout}s）", "task_id": task_id}), 504
+            # 超时处理：发终止信号 + 杀 ffmpeg 子进程 + 等待线程退出
+            logger.error("推理超时 (>%ds)，终止任务 %s", dynamic_timeout, task_id)
+            abort_event.set()
+            # 终止所有正在运行的 ffmpeg 子进程
+            with _active_ffmpeg_lock:
+                for proc in _active_ffmpeg_procs[:]:
+                    try:
+                        proc.kill()
+                        proc.wait(timeout=5)
+                    except Exception:
+                        pass
+                _active_ffmpeg_procs.clear()
+            # 等待线程退出（daemon 线程会在主进程退出时自动终止，但这里给一定时间让它清理）
+            thread.join(timeout=10)
+            # 标记超时错误，由 finally 统一清理
+            result_container["error"] = f"推理超时（>{dynamic_timeout:.0f}s）"
+            result_container["timeout"] = True
+
+        if result_container.get("timeout"):
+            return jsonify({"error": result_container["error"], "task_id": task_id}), 504
 
         if result_container["error"]:
             logger.error("推理失败 task_id=%s: %s", task_id, result_container["error"])
@@ -850,6 +1155,8 @@ def inference():
         current_task["task_id"] = None
         current_task["process"] = None
         current_task["start_time"] = 0.0
+        current_abort_event = None
+        current_thread = None
 
         # 清理临时文件
         if video_path and video_path.parent.exists():
@@ -863,20 +1170,35 @@ def inference():
 @app.route("/cancel", methods=["POST"])
 def cancel():
     """终止当前正在进行的推理任务."""
+    global current_abort_event, current_thread
+
     if current_task["task_id"] is None:
         return jsonify({"message": "当前无正在运行的任务"})
 
     task_id = current_task["task_id"]
     logger.info("收到取消请求，终止任务 %s", task_id)
 
-    # 终止推理进程（如果是 subprocess）
-    if current_task["process"] and current_task["process"] != "inference_thread":
-        try:
-            current_task["process"].terminate()
-            current_task["process"].wait(timeout=5)
-            logger.info("已终止推理进程")
-        except Exception as exc:
-            logger.warning("终止进程失败: %s", exc)
+    # 发送终止信号给推理线程
+    if current_abort_event is not None:
+        current_abort_event.set()
+        logger.info("已发送终止信号给推理线程")
+
+    # 终止所有正在运行的 ffmpeg 子进程
+    with _active_ffmpeg_lock:
+        for proc in _active_ffmpeg_procs[:]:
+            try:
+                proc.kill()
+                proc.wait(timeout=5)
+                logger.info("已终止 ffmpeg 子进程")
+            except Exception as exc:
+                logger.warning("终止 ffmpeg 子进程失败: %s", exc)
+        _active_ffmpeg_procs.clear()
+
+    # 等待推理线程退出（daemon 线程会在主进程退出时自动终止）
+    if current_thread is not None and current_thread.is_alive():
+        current_thread.join(timeout=10)
+        if current_thread.is_alive():
+            logger.warning("推理线程未能在 10s 内退出，将作为 daemon 线程自动终止")
 
     # 清理临时文件
     task_dir = Path(Config.temp_dir) / task_id
@@ -891,6 +1213,8 @@ def cancel():
     current_task["task_id"] = None
     current_task["process"] = None
     current_task["start_time"] = 0.0
+    current_abort_event = None
+    current_thread = None
 
     return jsonify({"message": f"已取消任务 {task_id}"})
 
@@ -899,6 +1223,8 @@ def cancel():
 
 
 def main():
+    import os as _os
+    _os.environ.setdefault("PYTORCH_CUDA_ALLOC_CONF", "max_split_size_mb:128")
     """启动 Flask 服务."""
     # 创建临时目录
     Path(Config.temp_dir).mkdir(parents=True, exist_ok=True)

infra/docker/api-base.Dockerfile

@@ -31,6 +31,12 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 COPY infra/fonts/NotoSansSC-VF.ttf /usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf
 COPY infra/fonts/NotoSerifCJKsc-VF.otf /usr/share/fonts/opentype/noto/NotoSerifCJKsc-VF.otf
 COPY infra/fonts/LXGWWenKai-Regular.ttf /usr/share/fonts/truetype/lxgw/LXGWWenKai-Regular.ttf
+# #2001 爆款标题字体：优设标题黑 / 阿里普惠体 Bold / 抖音美好体 / 思源黑体 Heavy
+RUN mkdir -p /usr/share/fonts/truetype/xiaoxia
+COPY infra/fonts/xiaoxia/YouSheBiaoTiHei.ttf /usr/share/fonts/truetype/xiaoxia/YouSheBiaoTiHei.ttf
+COPY infra/fonts/xiaoxia/AlibabaPuHuiTi-Bold.ttf /usr/share/fonts/truetype/xiaoxia/AlibabaPuHuiTi-Bold.ttf
+COPY infra/fonts/xiaoxia/DouyinSansBold.otf /usr/share/fonts/truetype/xiaoxia/DouyinSansBold.otf
+COPY infra/fonts/xiaoxia/NotoSansSC-Black.otf /usr/share/fonts/truetype/xiaoxia/NotoSansSC-Black.otf
 RUN fc-cache -fv
 
 # 创建虚拟环境

infra/docker/worker-base.Dockerfile

@@ -35,6 +35,12 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 COPY infra/fonts/NotoSansSC-VF.ttf /usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf
 COPY infra/fonts/NotoSerifCJKsc-VF.otf /usr/share/fonts/opentype/noto/NotoSerifCJKsc-VF.otf
 COPY infra/fonts/LXGWWenKai-Regular.ttf /usr/share/fonts/truetype/lxgw/LXGWWenKai-Regular.ttf
+# #2001 爆款标题字体：优设标题黑 / 阿里普惠体 Bold / 抖音美好体 / 思源黑体 Heavy
+RUN mkdir -p /usr/share/fonts/truetype/xiaoxia
+COPY infra/fonts/xiaoxia/YouSheBiaoTiHei.ttf /usr/share/fonts/truetype/xiaoxia/YouSheBiaoTiHei.ttf
+COPY infra/fonts/xiaoxia/AlibabaPuHuiTi-Bold.ttf /usr/share/fonts/truetype/xiaoxia/AlibabaPuHuiTi-Bold.ttf
+COPY infra/fonts/xiaoxia/DouyinSansBold.otf /usr/share/fonts/truetype/xiaoxia/DouyinSansBold.otf
+COPY infra/fonts/xiaoxia/NotoSansSC-Black.otf /usr/share/fonts/truetype/xiaoxia/NotoSansSC-Black.otf
 RUN fc-cache -fv
 
 # 创建虚拟环境

packages/adapters/sqlalchemy_impl/models.py

@@ -711,6 +711,7 @@ class LipsyncJobModel(Base):
     voice_id = Column(String(200), nullable=False, default="")
     script_text = Column(Text, nullable=False, default="")
     speed = Column(Float, nullable=False, default=1.0)
+    style = Column(String(32), nullable=False, default="")
     emotion = Column(String(20), nullable=False, default="")
 
     # MediaKit 任务状态
@@ -750,6 +751,8 @@ class AiAvatarRenderJob(Base):
     # b_roll_segments 格式: [{"script_segment_index": 0, "asset_url": "...", "mode": "fullscreen|pip", "start_time": 5.0, "end_time": 10.0}, ...]
     title_config = Column(JSON, nullable=False, default=dict)
     cover_config = Column(JSON, nullable=False, default=dict)
+    # #2001 封面独立标题配置（结构同 title_config；为空时封面不叠标题）
+    cover_title_config = Column(JSON, nullable=False, default=dict)
 
     # 任务状态
     status = Column(String(20), nullable=False, default="pending", index=True)

packages/adapters/sqlalchemy_impl/session.py

@@ -18,6 +18,10 @@ def build_engine(
     pool_timeout: int = 30,
     pool_recycle: int = 3600,
 ):
+    # SQLite 不支持 QueuePool 的 pool_size/max_overflow/pool_timeout，
+    # 传了会在 create_engine 阶段直接 TypeError，这里只对非 SQLite 传连接池参数。
+    if _is_sqlite(database_url):
+        return create_engine(database_url, pool_recycle=pool_recycle)
     return create_engine(
         database_url,
         pool_size=pool_size,

packages/application/cosyvoice_service.py

@@ -26,12 +26,43 @@ from packages.shared.config import get_shared_settings
 logger = logging.getLogger(__name__)
 
 
-# CosyVoice v3 情绪通过 input.instruction 中文自然语言指令控制（不再使用枚举 emotion 字段）。
-# 官方文档：instruction 格式严格为 "你说话的情感是<情感值>。"，结尾中文句号不可省略；
-# 情感值必须是 7 种英文枚举之一：neutral/happy/sad/angry/surprised/fearful/disgusted。
+# ── Style（语气风格）→ CosyVoice instruct 自然语言指令 ──
+# 前端 PR#2002 传 6 种 style：natural/sweet/excited/professional/news/livestream。
+# style 是新的统一参数；emotion 为 deprecated 兼容别名，内部映射为 style。
 # 参考：https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list
-# 前端可传英文枚举或中文标签（中立/开心/难过/生气/惊讶/恐惧/厌恶），统一归一化为英文枚举。
-# 映射表 key（不区分大小写）: 英文枚举/旧英文/中文标签 → 7 种标准英文枚举
+STYLE_INSTRUCTION_MAP: dict[str, str] = {
+    "natural": "用自然、平和的语气说话。",
+    "sweet": "用温柔甜美、亲切柔和的语气说话。",
+    "excited": "用兴奋、激动的语气说话。",
+    "professional": "用专业、正式的语气说话。",
+    "news": "用新闻播报的语气说话。",
+    "livestream": "用直播解说的语气说话。",
+}
+
+# 有效 style 值集合（供 schema / 校验使用）
+VALID_STYLES: frozenset[str] = frozenset(STYLE_INSTRUCTION_MAP.keys())
+
+# ── 旧 emotion → 新 style 兼容映射（方案 B：统一 style，emotion deprecated）──
+_EMOTION_TO_STYLE: dict[str, str] = {
+    "neutral": "natural",
+    "happy": "excited",
+    "sad": "sweet",
+    "angry": "excited",
+    "surprised": "excited",
+    "fearful": "sweet",
+    "disgusted": "natural",
+}
+
+# 严格格式系统音色：style → emotion 回退（用于无法使用自由文本指令的音色）
+_STYLE_TO_EMOTION: dict[str, str] = {
+    "natural": "neutral",
+    "sweet": "sad",
+    "excited": "happy",
+    "professional": "neutral",
+    # news / livestream 无直接对应 emotion，特殊处理
+}
+
+# ── 旧 emotion 映射表（deprecated，保留以兼容历史数据）──
 EMOTION_MAP: dict[str, str] = {
     # ── 7 种标准英文枚举（CosyVoice v3 官方支持的情感值）──
     "neutral": "neutral",
@@ -71,28 +102,6 @@ EMOTION_MAP: dict[str, str] = {
     "friendly": "happy",
 }
 
-# ── Style（语气风格）→ Instruct 指令映射（CosyVoice v3 Instruct 模式）──
-# 前端传 6 种 style（natural/excited/professional/gentle/news/livestream），
-# 映射为中文自然语言指令，写入 CosyVoice input.instruction 字段。
-# 克隆音色直接使用中文自然语言；系统音色（严格格式）需转换为兼容的 instruct。
-STYLE_INSTRUCTION_MAP: dict[str, str] = {
-    "natural": "用自然、平和的语气说话。",
-    "excited": "用兴奋、激动的语气说话。",
-    "professional": "用专业、正式的语气说话。",
-    "gentle": "用温柔、柔和的语气说话。",
-    "news": "用新闻播报的语气说话。",
-    "livestream": "用直播解说的语气说话。",
-}
-
-# 严格格式系统音色：style → emotion 映射（用于无法使用自由文本指令的音色）
-_STYLE_TO_EMOTION: dict[str, str] = {
-    "natural": "neutral",
-    "excited": "happy",
-    "professional": "neutral",
-    "gentle": "neutral",
-    # news / livestream 无直接对应 emotion，由 build_style_instruction 特殊处理
-}
-
 
 # ── 支持 emotion Instruct 的 v3-flash 系统音色白名单（官方音色列表标注"Instruct：支持"且支持情感值）──
 # 这些音色的 instruction 必须使用中文固定格式 "你说话的情感是<emotion>。"；
@@ -144,6 +153,70 @@ def build_emotion_instruction(voice_id: str, emotion_enum: str) -> str:
     return ""
 
 
+def resolve_style(style: str = "", emotion: str = "") -> str:
+    """统一解析 style 参数（方案 B）.
+
+    - style 有值且合法：直接使用（style 优先级最高）。
+    - style 为空但 emotion 有值：将旧 emotion 归一化后映射为 style。
+    - 两者皆空：返回空串（调用方不传 instruction）。
+
+    Args:
+        style: 新的语气风格（natural/sweet/excited/professional/news/livestream）
+        emotion: 旧的情绪参数（deprecated，内部映射为 style）
+
+    Returns:
+        解析后的 style 字符串；无需 instruct 时返回空串
+    """
+    s = (style or "").strip().lower()
+    if s:
+        if s in VALID_STYLES:
+            return s
+        logger.warning("未知的 style 值 %r，忽略 style 参数", style)
+    # 回退：emotion → style
+    norm = normalize_emotion(emotion)
+    if not norm:
+        return ""
+    mapped = _EMOTION_TO_STYLE.get(norm)
+    if not mapped:
+        logger.warning("emotion %r 无法映射到 style，跳过 instruct", norm)
+    return mapped or ""
+
+
+def build_style_instruction(voice_id: str, style: str) -> str:
+    """根据 voice 类型构造 style instruction.
+
+    - natural：返回空串（不额外加 instruct，使用 CosyVoice 默认自然语气）。
+    - 克隆/设计音色：使用中文自然语言指令（DashScope 允许任意自然语言）。
+    - 系统音色中支持 emotion instruct 的白名单音色：
+      若 style 可映射到 emotion，用严格格式 "你说话的情感是<emotion>。"；
+      news/livestream 尝试直接用中文 instruct（部分音色支持自由文本）。
+    - 其他系统音色：返回空串。
+
+    Args:
+        voice_id: CosyVoice voice 参数
+        style: 已通过 resolve_style() 解析的 style 值
+
+    Returns:
+        拼接好的 instruction 字符串；无需 instruct 时返回空串
+    """
+    if not style or style == "natural":
+        return ""
+    # 克隆音色：直接使用中文自然语言指令
+    if _is_cloned_voice(voice_id):
+        return STYLE_INSTRUCTION_MAP.get(style, "")
+    # 系统音色白名单：优先映射到严格 emotion 格式
+    if voice_id in _SYSTEM_VOICES_WITH_EMOTION_INSTRUCT:
+        emotion_val = _STYLE_TO_EMOTION.get(style)
+        if emotion_val:
+            return f"你说话的情感是{emotion_val}。"
+        # news/livestream 无 emotion 对应，尝试自由中文 instruct
+        desc = STYLE_INSTRUCTION_MAP.get(style, "")
+        if desc:
+            logger.info("音色 %s 使用自由文本 style instruct: %s", voice_id, desc)
+            return desc
+    return ""
+
+
 def normalize_emotion(emotion: str) -> str:
     """将前端情绪值归一化为 CosyVoice v3 官方英文枚举，用于拼入 instruction.
 
@@ -170,45 +243,6 @@ def normalize_emotion(emotion: str) -> str:
     return "neutral"
 
 
-def build_style_instruction(voice_id: str, style: str) -> str:
-    """根据 voice 类型构造 style instruction.
-
-    - 克隆音色：直接使用中文自然语言指令（CosyVoice 对克隆音色支持任意自然语言）
-    - 系统音色（emotion-instruct 白名单内）：转换为兼容的 instruct 格式
-      * excited → "你说话的情感是happy。"
-      * news → "你正在进行新闻播报，你说话的情感是neutral。"
-      * 其他无直接对应 → 返回空串（不传 instruction）
-    - 其他系统音色（不支持 Instruct）：返回空串
-
-    Args:
-        voice_id: CosyVoice voice 参数
-        style: 风格值（natural/excited/professional/gentle/news/livestream）
-
-    Returns:
-        instruction 字符串；不支持时返回空串
-    """
-    if not style or style == "natural":
-        # natural 是默认风格，不额外添加 instruct
-        return ""
-    description = STYLE_INSTRUCTION_MAP.get(style, "")
-    if not description:
-        logger.warning("未知的 style 值 %r，跳过 style instruction", style)
-        return ""
-    # 克隆音色：直接使用中文自然语言
-    if _is_cloned_voice(voice_id):
-        return description
-    # 严格格式系统音色（white-listed）：转换为兼容格式
-    if voice_id in _SYSTEM_VOICES_WITH_EMOTION_INSTRUCT:
-        if style == "news":
-            return "你正在进行新闻播报，你说话的情感是neutral。"
-        emotion_val = _STYLE_TO_EMOTION.get(style)
-        if emotion_val:
-            return f"你说话的情感是{emotion_val}。"
-        return ""
-    # 其他系统音色：不支持 Instruct，返回空串
-    return ""
-
-
 # 系统音色仅支持中文/英文（language_hints 取值）
 _SYSTEM_VOICE_LANGS = {"zh", "en"}
 
@@ -649,14 +683,11 @@ class CosyVoiceService:
             format: 输出格式（mp3/wav/pcm），空表示使用配置默认值
             speed: 语速（0.5-2.0），1.0 为正常速度
             volume: 音量（0-100），默认 50
-            emotion: 情绪，英文枚举 neutral/happy/sad/angry/surprised/fearful/disgusted
-                     或前端中文标签（中立/开心/难过/生气/惊讶/恐惧/厌恶），兼容旧值
-                     natural/excited/calm/friendly；空串不传，未知值默认 neutral
-            language: 语言代码（zh/en 等，默认 zh；系统音色仅 zh/en 传 language_hints）
-            style: 语气风格（natural/excited/professional/gentle/news/livestream），
-                   可选参数，用于 CosyVoice v3 Instruct 模式；
-                   style 与 emotion 互斥，style 优先级高于 emotion
+            style: 语气风格（natural/sweet/excited/professional/news/livestream），
+                   新的统一参数；优先级高于 emotion
             pitch: 音调（0.5-2.0），1.0 为默认值
+            emotion: 【deprecated】旧情绪参数，内部通过 resolve_style() 映射为 style
+            language: 语言代码（zh/en 等，默认 zh；系统音色仅 zh/en 传 language_hints）
 
         Returns:
             dict: {"audio_url": str, "request_id": str,
@@ -684,16 +715,16 @@ class CosyVoiceService:
             "rate": speed,
             "volume": volume,
         }
-        # pitch: CosyVoice API 支持 [0.5, 2.0]，1.0 为默认值，非默认时才传
-        if pitch != 1.0:
+        # pitch: CosyVoice API 支持 [0.5, 2.0]，非默认值时才传
+        if pitch and pitch != 1.0:
             input_payload["pitch"] = pitch
-        # style 优先级高于 emotion：style 是新的 Instruct 模式参数，
-        # 前端传 style 时不再走旧的 emotion 逻辑
+        # style 优先：显式传 style 时走 build_style_instruction()；
+        # 无 style 时回退到旧的 emotion → build_emotion_instruction() 逻辑（向后兼容）
+        s = (style or "").strip().lower()
         instruction = ""
-        if style:
-            instruction = build_style_instruction(voice_id, style)
+        if s:
+            instruction = build_style_instruction(voice_id, s)
         if not instruction:
-            # 回退到旧的 emotion → instruction 逻辑
             norm_emotion = normalize_emotion(emotion)
             instruction = build_emotion_instruction(voice_id, norm_emotion)
         if instruction:
@@ -757,6 +788,7 @@ class CosyVoiceService:
         volume: int = 50,
         emotion: str = "",
         language: str = "zh",
+        timeout: float = 120.0,
         style: str = "",
         pitch: float = 1.0,
     ) -> SynthesizeResult:
@@ -772,7 +804,7 @@ class CosyVoiceService:
             format: 输出格式（mp3/wav/pcm），空表示使用配置默认值
             speed: 语速（0.5-2.0），1.0 为正常速度
             volume: 音量（0-100），默认 50
-            style: 语气风格（natural/excited/professional/gentle/news/livestream）
+            style: 语气风格（natural/sweet/excited/professional/news/livestream）
             pitch: 音调（0.5-2.0），1.0 为默认值
             timeout: 超时时间（秒），保留参数兼容
 

packages/application/tts_job/workflow.py

@@ -145,22 +145,22 @@ class TTSWorkflowService:
         try:
             _meta = dict(job.metadata)
             _speed = float(_meta.get("speed", 1.0) or 1.0)
-            _volume = int(_meta.get("volume", 50) or 50)
             _emotion = str(_meta.get("emotion", "") or "")
             _style = str(_meta.get("style", "") or "")
-            _language = str(_meta.get("language", "zh-CN") or "zh-CN")
+            _volume = int(_meta.get("volume", 50) or 50)
             _pitch = float(_meta.get("pitch", 1.0) or 1.0)
+            _language = str(_meta.get("language", "zh-CN") or "zh-CN")
             submit_result = self.cosyvoice_service.submit_synthesize_task(
                 text=job.input_text,
                 voice_id=job.voice_id,
                 sample_rate=job.sample_rate,
                 format=job.format,
                 speed=_speed,
+                style=_style,
                 volume=_volume,
+                pitch=_pitch,
                 emotion=_emotion,
                 language=_language,
-                style=_style,
-                pitch=_pitch,
             )
 
             # 保存 task_id / request_id 到 metadata
@@ -296,10 +296,10 @@ class TTSWorkflowService:
             job_metadata = job.metadata or {}
             speed = float(job_metadata.get("speed", 1.0))
             volume = int(job_metadata.get("volume", 50))
-            emotion = str(job_metadata.get("emotion", "") or "")
             style = str(job_metadata.get("style", "") or "")
+            pitch = float(job_metadata.get("pitch", 1.0))
+            emotion = str(job_metadata.get("emotion", "") or "")
             language = str(job_metadata.get("language", "zh-CN") or "zh-CN")
-            pitch = float(job_metadata.get("pitch", 1.0) or 1.0)
 
             result = self.cosyvoice_service.submit_synthesize_task(
                 text=job.input_text,
@@ -307,11 +307,11 @@ class TTSWorkflowService:
                 sample_rate=job.sample_rate,
                 format=job.format,
                 speed=speed,
+                style=style,
                 volume=volume,
+                pitch=pitch,
                 emotion=emotion,
                 language=language,
-                style=style,
-                pitch=pitch,
             )
             audio_url = result.get("audio_url", "")
             if not audio_url:
@@ -424,9 +424,9 @@ class TTSWorkflowService:
         results: list[dict | None] = [None] * len(segments)
         _seg_meta = job.metadata or {}
         _seg_speed = float(_seg_meta.get("speed", 1.0) or 1.0)
-        _seg_volume = int(_seg_meta.get("volume", 50) or 50)
         _seg_emotion = str(_seg_meta.get("emotion", "") or "")
         _seg_style = str(_seg_meta.get("style", "") or "")
+        _seg_volume = int(_seg_meta.get("volume", 50) or 50)
         _seg_pitch = float(_seg_meta.get("pitch", 1.0) or 1.0)
 
         with ThreadPoolExecutor(max_workers=max_workers) as executor:
@@ -439,11 +439,11 @@ class TTSWorkflowService:
                     sample_rate=job.sample_rate,
                     format=job.format,
                     speed=_seg_speed,
+                    style=_seg_style,
                     volume=_seg_volume,
+                    pitch=_seg_pitch,
                     emotion=_seg_emotion,
                     language=_seg_meta.get("language", "zh-CN") or "zh-CN",
-                    style=_seg_style,
-                    pitch=_seg_pitch,
                 )
                 future_to_idx[future] = idx
 
@@ -533,6 +533,8 @@ class TTSWorkflowService:
         job_metadata = job.metadata or {}
         speed = float(job_metadata.get("speed", 1.0))
         volume = int(job_metadata.get("volume", 50))
+        style = str(job_metadata.get("style", "") or "")
+        pitch = float(job_metadata.get("pitch", 1.0))
         emotion = str(job_metadata.get("emotion", "") or "")
 
         # 分段文本（用于缺失段重新合成）
@@ -567,7 +569,9 @@ class TTSWorkflowService:
                         sample_rate=job.sample_rate,
                         format=job.format,
                         speed=speed,
+                        style=style,
                         volume=volume,
+                        pitch=pitch,
                         emotion=emotion,
                         language=job.metadata.get("language", "zh-CN") if hasattr(job, "metadata") else "zh-CN",
                     )

packages/domain/ass_subtitle_builder.py

@@ -30,6 +30,8 @@ TITLE_MARGIN_SIDE = 40
 # - 楷体 → LXGW WenKai（霞鹜文楷，#1896 新增 SIL OFL 开源楷体）
 # - 苹方/PingFang/微软雅黑：服务器 Linux 无对应字体，fallback 思源黑体
 # - 华康俪金黑：商业字体有版权风险，前端已移除，后端保留映射 fallback 思源黑体（兼容老数据）
+# #2001 爆款标题字体（部署到 /usr/share/fonts/truetype/xiaoxia/）：
+# - 优设标题黑 / 阿里普惠体 Bold / 抖音美好体 / 思源黑体 Heavy（独立 Black 字重）
 FONT_NAME_MAP: dict[str, str] = {
     "思源黑体": "Noto Sans SC",
     "思源宋体": "Noto Serif CJK SC",
@@ -40,6 +42,15 @@ FONT_NAME_MAP: dict[str, str] = {
     "楷体": "LXGW WenKai",
     "霞鹜文楷": "LXGW WenKai",
     "华康俪金黑": "Noto Sans SC",
+    # #2001 爆款标题字体
+    "优设标题黑": "YouSheBiaoTiHei",
+    "阿里普惠体": "Alibaba PuHuiTi",
+    "阿里普惠体 Bold": "Alibaba PuHuiTi",
+    "阿里巴巴普惠体": "Alibaba PuHuiTi",
+    "抖音美好体": "Douyin Sans",
+    "抖音体": "Douyin Sans",
+    "思源黑体 Heavy": "Noto Sans SC",
+    "思源黑体 Black": "Noto Sans SC",
 }
 
 # ASS Fontsize 是字体 em-square 高度（含 Latin 升降部留白），
@@ -421,6 +432,14 @@ def build_ass_content(
         safe_title_text_raw = escape_ass_text(title_text)
         safe_title_text = _wrap_title_text(safe_title_text_raw, video_width, title_font_size)
 
+        # #2001 逐行样式覆盖：按 line_overrides 在每行前注入 ASS inline override 标签
+        # line_overrides 透传自前端爆款标题面板，SubtitleStyle.from_dict 已做安全过滤
+        if title_config.get("line_overrides"):
+            from packages.domain.subtitle_style import SubtitleStyle
+
+            _title_style_for_overrides = SubtitleStyle.from_dict(title_config)
+            safe_title_text = _title_style_for_overrides.apply_line_overrides(safe_title_text)
+
         # 自由位置：在文本前注入 \pos override tag（锚点为文本块中心，配合 \an5）
         if title_pos is not None:
             safe_title_text = f"{{\\pos({title_pos[0]},{title_pos[1]})}}{safe_title_text}"
@@ -455,6 +474,13 @@ def build_ass_content(
 
         safe_subtitle_text = escape_ass_text(subtitle_text)
 
+        # #2001 逐行样式覆盖（字幕路径同样支持）
+        if subtitle_config.get("line_overrides"):
+            from packages.domain.subtitle_style import SubtitleStyle
+
+            _sub_style_for_overrides = SubtitleStyle.from_dict(subtitle_config)
+            safe_subtitle_text = _sub_style_for_overrides.apply_line_overrides(safe_subtitle_text)
+
         events.append(
             "Dialogue: 0,0:00:00.00,"
             f"{format_ass_time(video_duration)},"

packages/domain/subtitle_style.py

@@ -6,7 +6,7 @@
 
 from __future__ import annotations
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Any
 
 # ── 常量 ──────────────────────────────────────────────────────────────────────
@@ -163,6 +163,12 @@ class SubtitleStyle:
     fade_out: float = 0.0
     animation_type: str = "none"
 
+    # 逐行独立样式覆盖（#2001 爆款标题样式面板）
+    # list[dict]，每项可选字段: line_index(0-based，支持负数从末尾倒数),
+    # color/font/size/bold/italic/stroke_color/stroke_width/shadow_color/shadow_offset_x/shadow_offset_y
+    # 渲染时按行索引匹配，用 ASS 内联 override 标签包裹该行。缺省字段继承主样式。
+    line_overrides: list[dict[str, Any]] = field(default_factory=list)
+
     @classmethod
     def from_dict(cls, config: dict[str, Any] | None) -> "SubtitleStyle":
         """从字典创建样式配置，带安全类型转换."""
@@ -193,6 +199,14 @@ class SubtitleStyle:
         if position not in POSITION_ALIGNMENT:
             position = DEFAULT_POSITION
 
+        # 逐行覆盖：仅保留 dict 类型项；非 dict 项过滤掉避免渲染崩溃
+        raw_overrides = config.get("line_overrides") or []
+        line_overrides: list[dict[str, Any]] = []
+        if isinstance(raw_overrides, list):
+            for item in raw_overrides:
+                if isinstance(item, dict):
+                    line_overrides.append(dict(item))
+
         return cls(
             font_name=safe_str("font", DEFAULT_FONT),
             font_size=safe_int("size", DEFAULT_FONT_SIZE),
@@ -221,6 +235,7 @@ class SubtitleStyle:
             fade_in=max(0.0, safe_float("fade_in", 0.0)),
             fade_out=max(0.0, safe_float("fade_out", 0.0)),
             animation_type=safe_str("animation_type", "none"),
+            line_overrides=line_overrides,
         )
 
     @property
@@ -248,6 +263,130 @@ class SubtitleStyle:
         color_bgr = hex_to_ass_bgr(self.background_color)
         return f"&H{alpha_hex}{color_bgr}"
 
+    def build_line_override_tag(self, line_index: int, total_lines: int) -> str:
+        r"""按 line_overrides 配置为指定行构造 ASS 内联 override 标签 {\c&HBBGGRR&...}.
+
+        仅返回大括号包裹的 override 标签串；调用方拼到该行文本前即可。
+        未配置该覆盖项时返回空串。缺省字段继承主样式，不生成对应 tag。
+
+        Args:
+            line_index: 行号（0-based）；支持负数（-1 为最后一行）。
+            total_lines: 总行数（用于解析负数索引）。
+        """
+        if not self.line_overrides or total_lines <= 0:
+            return ""
+
+        # 解析负数索引
+        resolved = line_index if line_index >= 0 else total_lines + line_index
+        if resolved < 0 or resolved >= total_lines:
+            return ""
+
+        override: dict[str, Any] | None = None
+        for item in self.line_overrides:
+            if not isinstance(item, dict):
+                continue
+            idx = item.get("line_index")
+            try:
+                idx_int = int(idx) if idx is not None else None
+            except (TypeError, ValueError):
+                continue
+            if idx_int is None:
+                continue
+            if idx_int < 0:
+                idx_int = total_lines + idx_int
+            if idx_int == resolved:
+                override = item
+                break
+
+        if not override:
+            return ""
+
+        tags: list[str] = []
+
+        # 主色（字体颜色）：\c&HBBGGRR&
+        color_val = override.get("color")
+        if isinstance(color_val, str) and color_val:
+            tags.append(f"\\c{hex_to_ass_color(color_val)}")
+
+        # 字号：\fsN
+        size_val = override.get("size") or override.get("font_size")
+        try:
+            size_int = int(size_val) if size_val is not None else None
+            if size_int and size_int > 0:
+                tags.append(f"\\fs{size_int}")
+        except (TypeError, ValueError):
+            pass
+
+        # 字体：\fnFontName
+        font_val = override.get("font") or override.get("font_name")
+        if isinstance(font_val, str) and font_val:
+            tags.append(f"\\fn{font_val}")
+
+        # 粗体：\b1 / \b0
+        bold_val = override.get("bold")
+        if isinstance(bold_val, bool):
+            tags.append("\\b1" if bold_val else "\\b0")
+
+        # 斜体：\i1 / \i0
+        italic_val = override.get("italic")
+        if isinstance(italic_val, bool):
+            tags.append("\\i1" if italic_val else "\\i0")
+
+        # 描边色：\3c&HBBGGRR&
+        stroke_c = override.get("stroke_color")
+        if isinstance(stroke_c, str) and stroke_c:
+            tags.append(f"\\3c{hex_to_ass_color(stroke_c)}")
+
+        # 描边宽：\bordN
+        stroke_w = override.get("stroke_width")
+        try:
+            sw = float(stroke_w) if stroke_w is not None else None
+            if sw is not None and sw >= 0:
+                tags.append(f"\\bord{sw:g}")
+        except (TypeError, ValueError):
+            pass
+
+        # 阴影色：\4c&HBBGGRR&
+        shadow_c = override.get("shadow_color")
+        if isinstance(shadow_c, str) and shadow_c:
+            tags.append(f"\\4c{hex_to_ass_color(shadow_c)}")
+
+        # 阴影偏移：\shadN（单值，同时设置 x/y；精细控制用 \xshad/\yshad）
+        sx = override.get("shadow_offset_x")
+        sy = override.get("shadow_offset_y")
+        try:
+            sx_i = int(sx) if sx is not None else None
+            sy_i = int(sy) if sy is not None else None
+            if sx_i is not None:
+                tags.append(f"\\xshad{sx_i}")
+            if sy_i is not None:
+                tags.append(f"\\yshad{sy_i}")
+        except (TypeError, ValueError):
+            pass
+
+        if not tags:
+            return ""
+        return "{" + "".join(tags) + "}"
+
+    def apply_line_overrides(self, text: str) -> str:
+        """按 line_overrides 对 ASS 文本（已 escape、换行用 \\N 分隔）逐行套 override 标签.
+
+        仅对换行后的每行首加对应 override 标签；无 override 的行保持原样。
+        """
+        if not self.line_overrides or not text:
+            return text
+        if "\\N" not in text:
+            # 单行
+            tag = self.build_line_override_tag(0, 1)
+            return tag + text if tag else text
+        lines = text.split("\\N")
+        total = len(lines)
+        out: list[str] = []
+        for i, ln in enumerate(lines):
+            tag = self.build_line_override_tag(i, total)
+            out.append(tag + ln if tag else ln)
+        return "\\N".join(out)
+
 
 # ── 字幕片段 ──────────────────────────────────────────────────────────────────
 

packages/domain/video_filter_builder.py

@@ -388,6 +388,11 @@ DRAWTEXT_FONT_SEARCH_PATHS: list[str] = [
     "/usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf",
     "/usr/share/fonts/opentype/noto/NotoSerifCJKsc-VF.otf",
     "/usr/share/fonts/truetype/lxgw/LXGWWenKai-Regular.ttf",
+    # #2001 爆款标题字体（优设标题黑 / 阿里普惠体 Bold / 抖音美好体 / 思源黑体 Heavy）
+    "/usr/share/fonts/truetype/xiaoxia/YouSheBiaoTiHei.ttf",
+    "/usr/share/fonts/truetype/xiaoxia/AlibabaPuHuiTi-Bold.ttf",
+    "/usr/share/fonts/truetype/xiaoxia/DouyinSansBold.otf",
+    "/usr/share/fonts/truetype/xiaoxia/NotoSansSC-Black.otf",
     "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
     "/usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc",
     "/usr/share/fonts/noto-cjk/NotoSansCJK-Regular.ttc",
@@ -400,6 +405,7 @@ DRAWTEXT_FONT_SEARCH_PATHS: list[str] = [
 # #1896 字体映射修复：每个字体映射到独立的关键字，而非全部回退到 NotoSansSC
 # - 苹方（macOS）/ 微软雅黑（Windows）/ PingFang：服务器 Linux 无对应文件，fallback 思源黑体
 # - 华康俪金黑：商业字体有版权风险，前端已按 #1896 要求移除，后端保留映射但 fallback 思源黑体（兼容老数据）
+# #2001 新增爆款标题字体映射
 DRAWTEXT_FONT_MAP: dict[str, str] = {
     "思源黑体": "NotoSansSC",
     "思源宋体": "NotoSerifCJKsc",
@@ -410,6 +416,15 @@ DRAWTEXT_FONT_MAP: dict[str, str] = {
     "楷体": "LXGWWenKai",
     "霞鹜文楷": "LXGWWenKai",
     "华康俪金黑": "NotoSansSC",
+    # #2001 爆款标题字体
+    "优设标题黑": "YouSheBiaoTiHei",
+    "阿里普惠体": "AlibabaPuHuiTi-Bold",
+    "阿里普惠体 Bold": "AlibabaPuHuiTi-Bold",
+    "阿里巴巴普惠体": "AlibabaPuHuiTi-Bold",
+    "抖音美好体": "DouyinSansBold",
+    "抖音体": "DouyinSansBold",
+    "思源黑体 Heavy": "NotoSansSC-Black",
+    "思源黑体 Black": "NotoSansSC-Black",
 }
 
 

tests/unit/test_1970_gpu_worker_heartbeat.py

@@ -68,7 +68,9 @@ def test_register_payload_includes_task_id_only_when_provided(worker, monkeypatc
 
     class _Resp:
         status_code = 200
-        text = ""
+
+        def json(self):
+            return {"worker_id": captured[-1]["worker_id"], "cancel_task": False}
 
     def _fake_post(url, json=None, headers=None, timeout=None):
         captured.append(json)
@@ -77,7 +79,9 @@ def test_register_payload_includes_task_id_only_when_provided(worker, monkeypatc
     monkeypatch.setattr(worker.requests, "post", _fake_post)
     monkeypatch.setattr(worker, "_check_musetalk_health", lambda: (True, {}))
 
-    assert worker._register("task-abc") is True
+    _ok, _cancel = worker._register("task-abc")
+    assert _ok is True
+    assert _cancel is False
     assert captured[-1]["task_id"] == "task-abc"
     assert captured[-1]["worker_id"]
 
@@ -93,7 +97,7 @@ def test_task_heartbeat_thread_sends_and_stops(worker, monkeypatch):
 
     def _fake_register(task_id=None):
         calls.append(task_id)
-        return True
+        return True, False
 
     monkeypatch.setattr(worker, "_register", _fake_register)
     hb = worker.TaskHeartbeat("task-hb1", interval=5)
@@ -105,6 +109,50 @@ def test_task_heartbeat_thread_sends_and_stops(worker, monkeypatch):
     assert calls and all(c == "task-hb1" for c in calls)
 
 
+def test_task_heartbeat_cancel_calls_musetalk_cancel(worker, monkeypatch):
+    """心跳响应 cancel_task=True → 调 _cancel_musetalk 并设置 cancelled 标志。"""
+    cancel_calls = []
+
+    monkeypatch.setattr(worker, "_register", lambda *a, **k: (True, True))
+    monkeypatch.setattr(worker, "_cancel_musetalk", lambda: cancel_calls.append(1))
+
+    hb = worker.TaskHeartbeat("task-cancel-1", interval=5)
+    hb.start()
+    hb.join(timeout=2)  # 检测到取消后线程自行 return
+    assert not hb.is_alive()
+    assert hb.cancelled is True
+    assert cancel_calls == [1]
+
+
+def test_handle_task_reports_cancelled_after_musetalk_abort(worker, monkeypatch):
+    """推理被 /cancel 终止后，hb.cancelled=True → 上报失败而非重试。"""
+    reports = []
+
+    monkeypatch.setattr(worker, "_register", lambda *a, **k: (True, False))
+    monkeypatch.setattr(worker, "_download", lambda url, path: True)
+    monkeypatch.setattr(worker, "_probe_duration", lambda path: 12.0)
+    # 模拟推理被终止（/inference 返回错误）
+    monkeypatch.setattr(worker, "_call_musetalk", lambda v, a, o: (False, 0.0, "推理被取消", False))
+    monkeypatch.setattr(
+        worker,
+        "_report_result",
+        lambda task_id, success, duration=0.0, error_msg="": reports.append(error_msg) or True,
+    )
+
+    # 让 TaskHeartbeat 在主线程检查时报告已取消
+    orig_hb_init = worker.TaskHeartbeat
+
+    def _hb(task_id, interval):
+        h = orig_hb_init(task_id, interval)
+        h.cancelled = True
+        return h
+
+    monkeypatch.setattr(worker, "TaskHeartbeat", _hb)
+
+    worker._handle_task({"task_id": "t-canceled", "video_url": "u", "audio_url": "u"})
+    assert reports == ["用户取消任务"]
+
+
 # ── 短视频前置拦截 ─────────────────────────────────────────────────
 
 
@@ -115,7 +163,7 @@ def test_handle_task_short_video_reports_failed_without_inference(worker, monkey
     audio.write_bytes(b"fake-audio")
     reports = []
 
-    monkeypatch.setattr(worker, "_register", lambda *a, **k: True)
+    monkeypatch.setattr(worker, "_register", lambda *a, **k: (True, False))
     monkeypatch.setattr(worker, "_download", lambda url, path: True)
     # ffprobe 读出 1.2s → 低于 3s 阈值
     monkeypatch.setattr(worker, "_probe_duration", lambda path: 1.2)
@@ -148,7 +196,7 @@ def test_handle_task_short_video_reports_failed_without_inference(worker, monkey
 def test_handle_task_probe_failure_does_not_block(worker, monkeypatch):
     """ffprobe 不可用（duration=0.0）时不能误杀，应继续推理."""
     reports = []
-    monkeypatch.setattr(worker, "_register", lambda *a, **k: True)
+    monkeypatch.setattr(worker, "_register", lambda *a, **k: (True, False))
     monkeypatch.setattr(worker, "_download", lambda url, path: True)
     monkeypatch.setattr(worker, "_probe_duration", lambda path: 0.0)
     monkeypatch.setattr(
@@ -214,7 +262,7 @@ def test_handle_task_retries_once_for_transient_then_succeeds(worker, monkeypatc
             return False, 0.0, "MuseTalk HTTP 503: busy", True
         return True, 6.5, "", False
 
-    monkeypatch.setattr(worker, "_register", lambda *a, **k: True)
+    monkeypatch.setattr(worker, "_register", lambda *a, **k: (True, False))
     monkeypatch.setattr(worker, "_download", lambda url, path: True)
     monkeypatch.setattr(worker, "_probe_duration", lambda path: 12.0)
     monkeypatch.setattr(worker, "_call_musetalk", _fake_call)
@@ -239,7 +287,7 @@ def test_handle_task_no_retry_for_deterministic_failure(worker, monkeypatch):
         return False, 0.0, "MuseTalk HTTP 400: bad input", False
 
     reports = []
-    monkeypatch.setattr(worker, "_register", lambda *a, **k: True)
+    monkeypatch.setattr(worker, "_register", lambda *a, **k: (True, False))
     monkeypatch.setattr(worker, "_download", lambda url, path: True)
     monkeypatch.setattr(worker, "_probe_duration", lambda path: 12.0)
     monkeypatch.setattr(worker, "_call_musetalk", _fake_call)

tests/unit/test_ai_avatar_emotion_tts_lipsync.py

@@ -6,6 +6,7 @@ CI 增量映射:
   ai_avatar_cover_service 智能选帧
 """
 
+import importlib
 import os
 from unittest.mock import MagicMock, patch
 
@@ -166,6 +167,116 @@ def test_submit_synthesize_payload_cloned_voice_english_emotion():
     assert inp["instruction"] == "Speak in a sad tone."
 
 
+# ── style（语气风格，#2002）────────────────────────────────────────────
+
+
+def test_style_natural_omits_instruction():
+    """style=natural 不加 instruct，使用 CosyVoice 默认自然语气。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="myclone_voice", style="natural")
+
+    inp = captured["json"]["input"]
+    assert "instruction" not in inp
+
+
+def test_style_sweet_cloned_voice_uses_chinese_instruction():
+    """克隆音色 + style=sweet → 中文自然语言指令。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="myclone_voice", style="sweet")
+
+    inp = captured["json"]["input"]
+    assert inp["instruction"] == "用温柔甜美、亲切柔和的语气说话。"
+
+
+@pytest.mark.parametrize(
+    ("style", "expected_fragment"),
+    [
+        ("excited", "兴奋"),
+        ("professional", "专业"),
+        ("news", "新闻"),
+        ("livestream", "直播"),
+    ],
+)
+def test_style_values_cloned_voice(style, expected_fragment):
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="myclone_voice", style=style)
+
+    inp = captured["json"]["input"]
+    assert expected_fragment in inp["instruction"]
+
+
+def test_style_system_voice_uses_emotion_mapping():
+    """系统白名单音色 + style=sweet → 严格中文 emotion 格式（映射到 sad）。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="longanyang", style="sweet")
+
+    inp = captured["json"]["input"]
+    assert inp["instruction"] == "你说话的情感是sad。"
+
+
+def test_style_takes_priority_over_emotion():
+    """同时传 style 和 emotion 时以 style 为准。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="myclone_voice", style="excited", emotion="sad")
+
+    inp = captured["json"]["input"]
+    assert "兴奋" in inp["instruction"]
+
+
+def test_unknown_style_ignored_falls_back_to_emotion():
+    """未知 style 值被忽略，回退到 emotion 逻辑。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="hi", voice_id="myclone_voice", style="nonexistent", emotion="sad")
+
+    inp = captured["json"]["input"]
+    assert inp["instruction"] == "Speak in a sad tone."
+
+
+def test_pitch_passed_only_when_non_default():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="hi", voice_id="myclone_voice", pitch=1.5)
+
+    inp = captured["json"]["input"]
+    assert inp["pitch"] == 1.5
+
+
+def test_pitch_omitted_at_default():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="hi", voice_id="myclone_voice", pitch=1.0)
+
+    inp = captured["json"]["input"]
+    assert "pitch" not in inp
+
+
+def test_resolve_style_maps_emotion_to_style():
+    """旧 emotion 值通过 resolve_style 映射为 style。"""
+    mod = importlib.import_module("packages.application.cosyvoice_service")
+
+    assert mod.resolve_style(emotion="happy") == "excited"
+    assert mod.resolve_style(emotion="sad") == "sweet"
+    assert mod.resolve_style(emotion="neutral") == "natural"
+    assert mod.resolve_style(style="news") == "news"
+    # style 优先
+    assert mod.resolve_style(style="news", emotion="happy") == "news"
+    assert mod.resolve_style() == ""
+
+
 # ── 对口型 TTS 直生分支 ─────────────────────────────────────────────────
 
 

tests/unit/test_ai_avatar_render_routes.py

@@ -50,6 +50,7 @@ def _make_mock_render_job(
     m.b_roll_segments = []
     m.title_config = {}
     m.cover_config = {}
+    m.cover_title_config = {}
     m.status = status
     m.progress = progress
     m.output_video_url = output_video_url
@@ -93,6 +94,7 @@ class TestRenderRoutes:
         body.b_roll_segments = []
         body.title_config = {}
         body.cover_config = {}
+        body.cover_title_config = {}
         body.project_id = ""
 
         result = create_render_job(
@@ -118,6 +120,7 @@ class TestRenderRoutes:
         body.b_roll_segments = []
         body.title_config = {}
         body.cover_config = {}
+        body.cover_title_config = {}
         body.project_id = ""
 
         with pytest.raises(HTTPException) as exc_info:
@@ -141,6 +144,7 @@ class TestRenderRoutes:
         body.b_roll_segments = []
         body.title_config = {}
         body.cover_config = {}
+        body.cover_title_config = {}
         body.project_id = ""
 
         with pytest.raises(HTTPException) as exc_info:

tests/unit/test_gpu_lipsync_routes.py

@@ -0,0 +1,89 @@
+"""GPU Worker 路由单测 — #2009 取消链路.
+
+直接调用路由函数（不经 HTTP 栈），显式注入 svc / _token 以跳过 Depends。
+CI 增量映射: gpu_lipsync.py (route) → test_gpu_lipsync_routes.py
+"""
+
+from __future__ import annotations
+
+import json
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+
+def _payload(**overrides):
+    from app.schemas.gpu_lipsync import GpuWorkerRegisterRequest
+
+    data = {
+        "worker_id": "w-1",
+        "hostname": "gpu-host",
+        "gpu_name": "RTX3060",
+        "free_vram_mb": 10000,
+        "capabilities": json.dumps({"musetalk": True}),
+    }
+    data.update(overrides)
+    return GpuWorkerRegisterRequest(**data)
+
+
+def test_register_returns_cancel_task_true_when_cancelled():
+    """心跳接口在任务已取消时必须把 cancel_task=True 透传给 Worker."""
+    fake_worker = MagicMock()
+    fake_worker.worker_id = "w-1"
+    fake_worker.hostname = "gpu-host"
+    fake_worker.gpu_name = "RTX3060"
+    fake_worker.free_vram_mb = 10000
+    fake_worker.capabilities = "musetalk"
+    fake_svc = MagicMock()
+    fake_svc.register_worker.return_value = (fake_worker, True)
+
+    from app.api.routes.gpu_lipsync import register_worker as route
+
+    resp = route(_payload(task_id="task-cancelled"), svc=fake_svc, _token="t")
+
+    assert resp.cancel_task is True
+    assert resp.ok is True
+    fake_svc.register_worker.assert_called_once()
+    kwargs = fake_svc.register_worker.call_args.kwargs
+    assert kwargs["task_id"] == "task-cancelled"
+
+
+def test_register_returns_cancel_task_false_normal():
+    """正常心跳 cancel_task=False."""
+    fake_worker = MagicMock()
+    fake_worker.worker_id = "w-1"
+    fake_worker.hostname = "gpu-host"
+    fake_worker.gpu_name = "RTX3060"
+    fake_worker.free_vram_mb = 10000
+    fake_worker.capabilities = "musetalk"
+    fake_svc = MagicMock()
+    fake_svc.register_worker.return_value = (fake_worker, False)
+
+    from app.api.routes.gpu_lipsync import register_worker as route
+
+    resp = route(_payload(), svc=fake_svc, _token="t")
+
+    assert resp.cancel_task is False
+
+
+def test_cancel_route_accepts_processing_status():
+    """cancel 路由允许 processing 状态（GPU 推理中），不再 400。"""
+    fake_job = MagicMock()
+    fake_job.status = "cancelled"
+
+    svc = MagicMock()
+    svc.cancel_job.return_value = fake_job
+
+    current_user = MagicMock()
+    current_user.user.id = "u1"
+
+    from app.api.routes.lipsync import cancel_lipsync_job as route
+
+    result = route("job-1", current_user, svc)
+
+    svc.cancel_job.assert_called_once_with("job-1", "u1")
+    assert result.status == "cancelled"
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-q"])

tests/unit/test_gpu_lipsync_service.py

@@ -173,10 +173,10 @@ def test_timed_out_task_is_redispatched(svc):
 
 
 def test_register_worker_creates_then_updates(svc):
-    w = svc.register_worker("w-1", hostname="pc1", gpu_name="RTX2060", free_vram_mb=3500)
+    w, _cancel = svc.register_worker("w-1", hostname="pc1", gpu_name="RTX2060", free_vram_mb=3500)
     assert w.worker_id == "w-1"
     assert w.gpu_name == "RTX2060"
-    w2 = svc.register_worker("w-1", hostname="pc1", gpu_name="RTX2060", free_vram_mb=2000)
+    w2, _cancel2 = svc.register_worker("w-1", hostname="pc1", gpu_name="RTX2060", free_vram_mb=2000)
     assert w2.free_vram_mb == 2000  # 更新
     assert w2.created_at == w.created_at  # 没新建
 
@@ -195,7 +195,7 @@ def test_register_with_task_id_refreshes_task_heartbeat(svc):
         {"last_heartbeat_at": old_hb - timedelta(seconds=300)}
     )
     svc.db.commit()
-    svc.register_worker("w-1", task_id=t.id)
+    _w, _c = svc.register_worker("w-1", task_id=t.id)
     svc.db.refresh(t)
     assert t.last_heartbeat_at > old_hb
     assert t.status == "processing"  # 心跳不改变状态
@@ -213,7 +213,7 @@ def test_register_task_heartbeat_ignores_finished_or_foreign_task(svc):
     svc.poll_task("w-1")
     done = svc.report_result(t.id, "w-1", success=True, duration_seconds=10.0)
     hb_when_done = done.last_heartbeat_at
-    svc.register_worker("w-1", task_id=t.id)
+    _w, _c = svc.register_worker("w-1", task_id=t.id)
     svc.db.refresh(t)
     assert t.status == "done"
     assert t.last_heartbeat_at == hb_when_done  # 没被改写
@@ -232,17 +232,61 @@ def test_register_task_heartbeat_ignores_finished_or_foreign_task(svc):
         {"last_heartbeat_at": owner_hb - timedelta(seconds=600)}
     )
     svc.db.commit()
-    svc.register_worker("w-1", task_id=t2.id)  # 旧 worker 迟到心跳
+    _w2, _c2 = svc.register_worker("w-1", task_id=t2.id)  # 旧 worker 迟到心跳
     svc.db.refresh(t2)
     assert t2.worker_id == "w-2"
     assert t2.status == "processing"
     assert t2.last_heartbeat_at == owner_hb
 
     # 场景 3：不存在的 task_id 不报错
-    svc.register_worker("w-1", task_id="nonexistent-id")
+    _wn, _cn = svc.register_worker("w-1", task_id="nonexistent-id")
     assert svc.db.get(GpuLipsyncTaskModel, "nonexistent-id") is None
 
 
+def test_register_task_heartbeat_detects_cancelled(svc):
+    """取消链路：任务已 cancelled 时，register 心跳必须返回 cancel_task=True."""
+    t = svc.create_task(video_url="v", audio_url="a")
+    svc.poll_task("w-1")
+    svc.db.refresh(t)
+    # 用户取消：直接把任务置为 cancelled
+    t.status = "cancelled"
+    t.finished_at = datetime.now(UTC)
+    svc.db.commit()
+
+    _w, cancel_task = svc.register_worker("w-1", task_id=t.id)
+    assert cancel_task is True
+    svc.db.refresh(t)
+    assert t.status == "cancelled"  # 心跳不改写已取消状态
+
+
+def test_report_result_cancelled_stays_cancelled(svc):
+    """Worker 终止取消任务后上报失败，report_result 必须保持 cancelled 不回退 pending."""
+    t = svc.create_task(video_url="v", audio_url="a")
+    svc.poll_task("w-1")
+    svc.db.refresh(t)
+    t.status = "cancelled"
+    svc.db.commit()
+
+    result = svc.report_result(t.id, "w-1", success=False, error_msg="推理被终止")
+    assert result.status == "cancelled"
+    assert result.finished_at is not None
+    assert "推理被终止" in (result.error_msg or "")
+
+
+def test_wait_for_result_returns_when_cancelled(svc):
+    """wait_for_result 将 cancelled 视为终态，立即返回，Celery 不回退 MediaKit."""
+    t = svc.create_task(video_url="v", audio_url="a")
+    svc.poll_task("w-1")
+    svc.db.refresh(t)
+    t.status = "cancelled"
+    t.finished_at = datetime.now(UTC)
+    svc.db.commit()
+
+    result = svc.wait_for_result(t.id, timeout_seconds=5, poll_interval=0.1)
+    assert result is not None
+    assert result.status == "cancelled"
+
+
 def test_default_gpu_task_timeout_is_900(svc):
     """#1970 默认超时 300→900，覆盖 RTX2060 长视频推理."""
     assert svc.settings.gpu_task_timeout_seconds == 900

tests/unit/test_lipsync_gpu_async_task.py

@@ -248,3 +248,41 @@ class TestSignMediaUrl:
         with patch.object(task_mod, "get_shared_storage_service", side_effect=RuntimeError("x")):
             url = "https://own-bucket.oss-cn-beijing.aliyuncs.com/a.wav"
             assert task_mod._sign_media_url(url) == url
+
+
+def test_cancelled_gpu_task_does_not_fallback_mediakit(monkeypatch):
+    """GPU 任务被用户取消 → Celery 任务直接标记 cancelled，不回退 MediaKit。"""
+    job = MagicMock()
+    job.id = "job-1"
+    job.status = "processing"
+    job.mediakit_task_id = "gpu:gpu-task-1"
+
+    gpu_task = MagicMock()
+    gpu_task.status = "cancelled"
+    gpu_task.error_msg = "用户取消"
+
+    fake_gpu_svc = MagicMock()
+    fake_gpu_svc.wait_for_result.return_value = gpu_task
+    gpu_service_cls = MagicMock(return_value=fake_gpu_svc)
+
+    fake_db = MagicMock()
+    fake_db.query.return_value.filter_by.return_value.first.return_value = job
+
+    # 直接替换 sys.modules 里的 gpu_lipsync_service 模块（全量跑时它可能已被
+    # 其他测试换成 MagicMock），保证任务函数内 from...import 一定拿到我们的类；
+    # 并替换 _get_db_session 绕开 worker_app / app.db 两条 import 分支。
+    import sys
+    from types import SimpleNamespace
+
+    fake_mod = SimpleNamespace(GpuLipsyncService=gpu_service_cls)
+    monkeypatch.setitem(sys.modules, "app.services.gpu_lipsync_service", fake_mod)
+    monkeypatch.setattr(task_mod, "_get_db_session", lambda: fake_db)
+    monkeypatch.setattr(task_mod, "logger", MagicMock())
+
+    task_mod.lipsync_gpu_process_async.run("job-1", "u1", "gpu-task-1")
+
+    assert job.status == "cancelled"
+    assert not str(job.mediakit_task_id).startswith("mk-")
+    fake_db.commit.assert_called()
+    fake_gpu_svc.wait_for_result.assert_called_once()
+    gpu_service_cls.assert_called_once_with(fake_db)

tests/unit/test_lipsync_gpu_integration.py

@@ -323,3 +323,126 @@ class TestGpuServiceHelpers:
         svc = GpuLipsyncService(db=fake_db)
         fake_db.query.return_value.filter.return_value.first.return_value = None
         assert svc.has_available_worker() is False
+
+
+# ── cancel_job 取消链路 (#2009) ─────────────────────────────────────
+
+
+def _build_sqlite_session():
+    from sqlalchemy import create_engine
+    from sqlalchemy.orm import sessionmaker
+
+    from packages.adapters.sqlalchemy_impl import models as _  # noqa: F401
+    from packages.adapters.sqlalchemy_impl.models import Base
+
+    engine = create_engine("sqlite:///:memory:", future=True)
+    Base.metadata.create_all(engine)
+    Session = sessionmaker(bind=engine, future=True)
+    return Session()
+
+
+def _make_real_job(db, *, status="processing", mediakit_task_id="gpu:gpu-task-1"):
+    import uuid
+
+    from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
+
+    job = LipsyncJobModel(
+        id=str(uuid.uuid4()),
+        user_id="u1",
+        project_id="p1",
+        video_url="videos/v.mp4",
+        audio_url="audios/a.wav",
+        enable_video_loop=True,
+        mediakit_task_id=mediakit_task_id,
+        status=status,
+    )
+    db.add(job)
+    db.commit()
+    return job
+
+
+def test_cancel_processing_gpu_job_marks_gpu_task_cancelled():
+    """processing 的 GPU job 取消时，关联 GpuLipsyncTask 必须同步置 cancelled."""
+    import uuid
+
+    from packages.adapters.sqlalchemy_impl.models import GpuLipsyncTaskModel
+
+    db = _build_sqlite_session()
+    gpu_task_id = str(uuid.uuid4())
+    gpu_task = GpuLipsyncTaskModel(
+        id=gpu_task_id,
+        video_url="v",
+        audio_url="a",
+        status="processing",
+        worker_id="w-1",
+        attempt=1,
+    )
+    db.add(gpu_task)
+    db.commit()
+
+    job = _make_real_job(db, mediakit_task_id=f"gpu:{gpu_task_id}")
+    from app.services.lipsync_service import LipsyncService
+
+    svc = LipsyncService(db=db, client=MagicMock())
+    result = svc.cancel_job(job.id, "u1")
+
+    assert result.status == "cancelled"
+    db.refresh(gpu_task)
+    assert gpu_task.status == "cancelled"
+    assert gpu_task.error_msg == "用户取消"
+    assert gpu_task.finished_at is not None
+
+
+def test_cancel_processing_gpu_job_skips_non_processing_gpu_task():
+    """GPU task 已不在 processing（如已 done）时，取消 job 不应改它，也不报错."""
+    import uuid
+
+    from packages.adapters.sqlalchemy_impl.models import GpuLipsyncTaskModel
+
+    db = _build_sqlite_session()
+    gpu_task_id = str(uuid.uuid4())
+    gpu_task = GpuLipsyncTaskModel(
+        id=gpu_task_id, video_url="v", audio_url="a", status="done", worker_id="w-1", attempt=1
+    )
+    db.add(gpu_task)
+    db.commit()
+
+    job = _make_real_job(db, mediakit_task_id=f"gpu:{gpu_task_id}")
+    from app.services.lipsync_service import LipsyncService
+
+    svc = LipsyncService(db=db, client=MagicMock())
+    result = svc.cancel_job(job.id, "u1")
+
+    assert result.status == "cancelled"
+    db.refresh(gpu_task)
+    assert gpu_task.status == "done"  # 没被动
+
+
+def test_cancel_processing_non_gpu_job_does_not_touch_gpu_table():
+    """mediakit_task_id 不是 gpu: 前缀（普通 MediaKit 任务）时，不查 GPU task."""
+    db = _build_sqlite_session()
+    job = _make_real_job(db, mediakit_task_id="mk-task-99")
+    from app.services.lipsync_service import LipsyncService
+
+    svc = LipsyncService(db=db, client=MagicMock())
+    result = svc.cancel_job(job.id, "u1")
+    assert result.status == "cancelled"
+
+
+def test_cancel_completed_job_unchanged():
+    """completed 状态不可取消，cancel_job 原样返回."""
+    db = _build_sqlite_session()
+    job = _make_real_job(db, status="completed", mediakit_task_id="gpu:x")
+    from app.services.lipsync_service import LipsyncService
+
+    svc = LipsyncService(db=db, client=MagicMock())
+    result = svc.cancel_job(job.id, "u1")
+    assert result.status == "completed"
+
+
+def test_cancel_job_not_found_returns_none():
+    db = _build_sqlite_session()
+    from app.services.lipsync_service import LipsyncService
+
+    svc = LipsyncService(db=db, client=MagicMock())
+    assert svc.cancel_job("nonexistent", "u1") is None

tests/unit/test_lipsync_routes.py

@@ -298,6 +298,7 @@ class TestLipsyncServiceUnit:
 
         mock_db = MagicMock()
         svc = LipsyncService(mock_db, client=mock_mediakit)
+        svc.settings.use_gpu_lipsync = False
 
         with pytest.raises(MediaKitError, match="API 调用失败"):
             svc.create_job(
@@ -473,6 +474,7 @@ class TestLipsyncServiceUnit:
             cosyvoice_service=mock_cosyvoice,
             voice_clone_repo=MagicMock(),
         )
+        svc.settings.use_gpu_lipsync = False
         job = svc.create_job(
             user_id="user-1",
             video_url="https://example.com/video.mp4",
@@ -616,12 +618,14 @@ class TestSignMediaUrl403Fix:
     def _svc(self, mock_mediakit, mock_cosyvoice):
         from app.services.lipsync_service import LipsyncService
 
-        return LipsyncService(
+        svc = LipsyncService(
             MagicMock(),
             client=mock_mediakit,
             cosyvoice_service=mock_cosyvoice,
             voice_clone_repo=MagicMock(),
         )
+        svc.settings.use_gpu_lipsync = False
+        return svc
 
     def test_own_oss_unsigned_url_gets_resigned(self, mock_mediakit, mock_cosyvoice):
         """裸 public_url（不带签名，私有桶匿名 403）必须被重签."""

tests/unit/test_subtitle_style.py

@@ -399,3 +399,226 @@ class TestSubtitleSegment:
     def test_is_valid_zero_duration(self):
         seg = SubtitleSegment(start=1, end=1, text="hello")
         assert seg.is_valid is False
+
+
+# ── line_overrides 逐行样式覆盖测试 ─────────────────────────────────────────
+
+
+class TestBuildLineOverrideTag:
+    def test_empty_overrides_returns_empty(self):
+        style = SubtitleStyle()
+        assert style.build_line_override_tag(0, 1) == ""
+
+    def test_no_matching_line_returns_empty(self):
+        style = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 1, "color": "#FF0000"}]})
+        assert style.build_line_override_tag(0, 2) == ""
+
+    def test_out_of_range_index_returns_empty(self):
+        style = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 5, "color": "#FF0000"}]})
+        assert style.build_line_override_tag(0, 2) == ""
+
+    def test_negative_index_resolves_from_end(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [{"line_index": -1, "color": "#FF0000", "font_size": 48}],
+            }
+        )
+        tag = style.build_line_override_tag(-1, 3)  # last line, 3 lines total
+        assert tag
+        assert "\\c&H000000FF" in tag  # red
+        assert "\\fs48" in tag
+        assert tag.startswith("{") and tag.endswith("}")
+
+    def test_color_override(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [{"line_index": 0, "color": "#00FF00"}],
+            }
+        )
+        tag = style.build_line_override_tag(0, 1)
+        assert "\\c&H0000FF00" in tag  # green = 00FF00 → bgr 00FF00
+
+    def test_font_size_override(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [{"line_index": 0, "font_size": 60}],
+            }
+        )
+        tag = style.build_line_override_tag(0, 1)
+        assert "\\fs60" in tag
+
+    def test_font_override(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [{"line_index": 0, "font": "优设标题黑"}],
+            }
+        )
+        tag = style.build_line_override_tag(0, 1)
+        assert "\\fn优设标题黑" in tag
+
+    def test_bold_on_off(self):
+        style_bold = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0, "bold": True}]})
+        assert "\\b1" in style_bold.build_line_override_tag(0, 1)
+        style_nobold = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0, "bold": False}]})
+        assert "\\b0" in style_nobold.build_line_override_tag(0, 1)
+
+    def test_italic_on_off(self):
+        style_italic = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0, "italic": True}]})
+        assert "\\i1" in style_italic.build_line_override_tag(0, 1)
+        style_noitalic = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0, "italic": False}]})
+        assert "\\i0" in style_noitalic.build_line_override_tag(0, 1)
+
+    def test_stroke_override(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [{"line_index": 0, "stroke_color": "#0000FF", "stroke_width": 3.0}],
+            }
+        )
+        tag = style.build_line_override_tag(0, 1)
+        assert "\\3c&H00FF0000" in tag  # blue bgr
+        assert "\\bord3" in tag
+
+    def test_shadow_override(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [
+                    {"line_index": 0, "shadow_color": "#000000", "shadow_offset_x": 2, "shadow_offset_y": 3}
+                ],
+            }
+        )
+        tag = style.build_line_override_tag(0, 1)
+        assert "\\4c&H00000000" in tag  # black
+        assert "\\xshad2" in tag
+        assert "\\yshad3" in tag
+
+    def test_full_combo(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [
+                    {
+                        "line_index": 0,
+                        "color": "#FF0000",
+                        "font_size": 72,
+                        "font": "抖音美好体",
+                        "bold": True,
+                        "italic": False,
+                        "stroke_color": "#FFFFFF",
+                        "stroke_width": 2,
+                        "shadow_color": "#000000",
+                        "shadow_offset_x": 0,
+                        "shadow_offset_y": 4,
+                    }
+                ],
+            }
+        )
+        tag = style.build_line_override_tag(0, 1)
+        assert "\\c&H000000FF" in tag
+        assert "\\fs72" in tag
+        assert "\\fn抖音美好体" in tag
+        assert "\\b1" in tag
+        assert "\\i0" in tag
+        assert "\\3c&H00FFFFFF" in tag
+        assert "\\bord2" in tag
+        assert "\\4c&H00000000" in tag
+        assert "\\xshad0" in tag
+        assert "\\yshad4" in tag
+
+    def test_empty_override_dict_returns_empty(self):
+        style = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0}]})
+        assert style.build_line_override_tag(0, 1) == ""
+
+    def test_non_dict_items_filtered(self):
+        # 从 dict 解析时已过滤，这里手动构造测试
+        style = SubtitleStyle(line_overrides=["not a dict", {"line_index": 0, "color": "#FF0000"}])
+        tag = style.build_line_override_tag(0, 1)
+        assert "\\c&H000000FF" in tag
+
+    def test_total_lines_zero_returns_empty(self):
+        style = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0, "color": "#FF0000"}]})
+        assert style.build_line_override_tag(0, 0) == ""
+
+    def test_line_index_none_skipped(self):
+        style = SubtitleStyle.from_dict({"line_overrides": [{"color": "#FF0000"}]})  # 无 line_index
+        assert style.build_line_override_tag(0, 1) == ""
+
+
+class TestApplyLineOverrides:
+    def test_empty_text_returns_empty(self):
+        style = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0, "color": "#FF0000"}]})
+        assert style.apply_line_overrides("") == ""
+
+    def test_no_overrides_returns_original(self):
+        style = SubtitleStyle()
+        assert style.apply_line_overrides("hello\\Nworld") == "hello\\Nworld"
+
+    def test_single_line(self):
+        style = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0, "color": "#FF0000", "font_size": 60}]})
+        result = style.apply_line_overrides("单行标题")
+        assert result.startswith("{")
+        assert "单行标题" in result
+        assert "\\c&H000000FF" in result
+
+    def test_multi_line_only_first_line_tagged(self):
+        style = SubtitleStyle.from_dict({"line_overrides": [{"line_index": 0, "color": "#FF0000", "font_size": 72}]})
+        result = style.apply_line_overrides("第一行\\N第二行\\N第三行")
+        lines = result.split("\\N")
+        assert len(lines) == 3
+        assert lines[0].startswith("{\\c&H000000FF")
+        assert "第一行" in lines[0]
+        assert lines[1] == "第二行"  # 无标签
+        assert lines[2] == "第三行"
+
+    def test_multi_line_middle_and_last(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [
+                    {"line_index": 0, "font_size": 72, "bold": True},
+                    {"line_index": -1, "color": "#00FF00", "font_size": 36},
+                ],
+            }
+        )
+        result = style.apply_line_overrides("主标题\\N副标题\\N脚注")
+        lines = result.split("\\N")
+        assert lines[0].startswith("{\\fs72\\b1}")
+        assert lines[1] == "副标题"
+        assert "\\c&H0000FF00" in lines[2]
+        assert "\\fs36" in lines[2]
+
+    def test_line_without_override_kept_verbatim(self):
+        style = SubtitleStyle.from_dict(
+            {
+                "line_overrides": [{"line_index": 1, "bold": True}],
+            }
+        )
+        result = style.apply_line_overrides("第一行\\N第二行")
+        lines = result.split("\\N")
+        assert lines[0] == "第一行"
+        assert lines[1].startswith("{\\b1}")
+        assert "第二行" in lines[1]
+
+
+class TestFromDictLineOverrides:
+    def test_from_dict_parses_line_overrides(self):
+        cfg = {
+            "line_overrides": [
+                {"line_index": 0, "color": "#FF0000", "bold": True},
+                {"line_index": 1, "font_size": 36},
+            ],
+        }
+        style = SubtitleStyle.from_dict(cfg)
+        assert len(style.line_overrides) == 2
+        assert style.line_overrides[0]["line_index"] == 0
+        assert style.line_overrides[1]["font_size"] == 36
+
+    def test_from_dict_filters_non_dict_items(self):
+        cfg = {"line_overrides": [{"line_index": 0, "color": "#FF0000"}, "bad", None, 123]}
+        style = SubtitleStyle.from_dict(cfg)
+        assert len(style.line_overrides) == 1
+
+    def test_from_dict_no_field_defaults_empty(self):
+        style = SubtitleStyle.from_dict({"font": "微软雅黑"})
+        assert style.line_overrides == []
+
+    def test_default_has_empty_list(self):
+        style = SubtitleStyle()
+        assert style.line_overrides == []

tests/unit/test_tts_preview.py

@@ -101,6 +101,9 @@ class TestTTSPreviewEndpoint:
             text="你好世界",
             voice_id="longxiaochun",
             speed=1.0,
+            style="",
+            volume=50,
+            pitch=1.0,
             emotion="",
             language="zh-CN",
         )
@@ -148,6 +151,9 @@ class TestTTSPreviewEndpoint:
             text="测试",
             voice_id="v1",
             speed=1.5,
+            style="",
+            volume=50,
+            pitch=1.0,
             emotion="",
             language="zh-CN",
         )
@@ -334,6 +340,9 @@ class TestTTSPreviewEndpoint:
             text="克隆音色测试",
             voice_id="cosyvoice_actual_voice_123",
             speed=1.0,
+            style="",
+            volume=50,
+            pitch=1.0,
             emotion="",
             language="zh-CN",
         )
@@ -412,6 +421,9 @@ class TestTTSPreviewEndpoint:
             text="预设音色测试",
             voice_id="longxiaoxia_v3",
             speed=1.0,
+            style="",
+            volume=50,
+            pitch=1.0,
             emotion="",
             language="zh-CN",
         )