fix: 修复封面标题叠加根因 - custom_title 未传入渲染流程

- _render_video 函数签名新增 custom_title 参数
- 函数体内注入 custom_title 到 plan_config 覆盖模板标题
- 调用处传入 task_info.get('custom_title', '')

alembic/versions/055_cover_templates.py

@@ -10,8 +10,6 @@ Changes:
 3. config 为 JSON 字段，存储封面配置信息
 """
 
-import json
-
 import sqlalchemy as sa
 
 from alembic import context, op
@@ -73,7 +71,7 @@ def upgrade() -> None:
                 name=name,
                 thumbnail_url="",
                 is_system=True,
-                config=json.dumps(config),
+                config=config,
                 created_at=sa.func.now(),
                 updated_at=sa.func.now(),
             )

alembic/versions/056_fix_cover_templates_config.py

@@ -0,0 +1,39 @@
+"""修复 cover_templates.config 双重序列化
+
+Revision ID: 056_fix_cover_templates_config
+Revises: 055_cover_templates
+Create Date: 2026-08-13
+
+问题: 055 迁移 seed 数据时 json.dumps(config) 导致 config 被双重序列化为 JSON 字符串
+例如 "{}"（字符串）而不是 {}（对象），导致 Pydantic CoverTemplateResponse 校验失败 500。
+
+修复: 从 JSON 字符串中提取文本值，再 cast 回 json 对象类型。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "056_fix_cover_templates_config"
+down_revision = "055_cover_templates"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+
+    # PostgreSQL: 从 JSON string scalar 中提取文本内容，cast 为 json object
+    # 例如: JSON string "{}" -> text "{}" -> JSON object {}
+    if conn.dialect.name == "postgresql":
+        conn.execute(
+            sa.text(
+                "UPDATE cover_templates SET config = (config#>>'{}')::json "
+                "WHERE jsonb_typeof(config::jsonb) = 'string'"
+            )
+        )
+
+
+def downgrade() -> None:
+    # No safe rollback — the original data was incorrect
+    pass

apps/api/app/schemas/generation_task.py

@@ -7,8 +7,8 @@ from pydantic import BaseModel, Field, field_validator, model_validator
 class ConfirmGenerationRequest(BaseModel):
     """确认生成请求体 — 基于预览任务创建正式生成任务"""
 
-    output_width: int = Field(default=1080, description="输出视频宽度")
-    output_height: int = Field(default=1920, description="输出视频高度")
+    output_width: int = Field(default=1080, ge=100, description="输出视频宽度")
+    output_height: int = Field(default=1920, ge=100, description="输出视频高度")
     cover_url: str = Field(default="", description="自定义封面图片 URL")
     custom_title: str = Field(default="", description="自定义视频标题")
 

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

@@ -258,9 +258,12 @@ test.describe("Core generation flow", () => {
     // Generate API may return 400 in test env if template has no ready segments
     // That is OK for a wizard flow smoke test
     if (genResp.ok()) {
-      const genData = (await genResp.json()) as { plan_id: string; generation_task_id: string }
-      expect(genData.plan_id).toBeTruthy()
-      expect(genData.generation_task_id).toBeTruthy()
+      const genData = (await genResp.json()) as {
+        items: Array<{ id: string; status: string }>
+        total: number
+      }
+      expect(genData.items.length).toBeGreaterThan(0)
+      expect(genData.items[0].id).toBeTruthy()
     } else {
       console.log(`[E2E] Generate API returned ${genResp.status()}, wizard flow test still passes`)
     }

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

@@ -56,10 +56,41 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
 
     try {
       // 使用确认生成 API（基于预览任务）
-      // 解析分辨率
-      const [widthStr, heightStr] = (props.videoRatio || "1080x1920").split("x")
-      const outputWidth = parseInt(widthStr, 10) || 1080
-      const outputHeight = parseInt(heightStr, 10) || 1920
+      // 解析分辨率：videoRatio 可能是 "9:16"（宽高比）或 "1080x1920"（分辨率）
+      const ratio = props.videoRatio || "9:16"
+      let outputWidth: number
+      let outputHeight: number
+
+      if (ratio.includes(":")) {
+        // 宽高比格式，如 "9:16" → 基于基准高度 1920 计算
+        const [rw, rh] = ratio.split(":").map(Number)
+        if (rw > 0 && rh > 0) {
+          // 基准：长边 1920，短边按比例计算
+          const [longSide, shortSide] = rw < rh ? [rh, rw] : [rw, rh]
+          const baseLong = 1920
+          const baseShort = Math.round((baseLong * shortSide) / longSide)
+          // 确保偶数（FFmpeg 要求）
+          const evenShort = baseShort - (baseShort % 2)
+          if (rw < rh) {
+            outputWidth = evenShort
+            outputHeight = baseLong
+          } else {
+            outputWidth = baseLong
+            outputHeight = evenShort
+          }
+        } else {
+          outputWidth = 1080
+          outputHeight = 1920
+        }
+      } else if (ratio.includes("x")) {
+        // 分辨率格式，如 "1080x1920"
+        const [wStr, hStr] = ratio.split("x")
+        outputWidth = parseInt(wStr, 10) || 1080
+        outputHeight = parseInt(hStr, 10) || 1920
+      } else {
+        outputWidth = 1080
+        outputHeight = 1920
+      }
 
       await confirmGeneration(props.previewTaskId, {
         output_width: outputWidth,

apps/worker/video_processing/render_adapter.py

@@ -47,7 +47,14 @@ def _parse_resolution(resolution_str: str | None) -> tuple[int, int]:
         w, h = resolution_str.lower().split("x", 1)
         width = int(w.strip())
         height = int(h.strip())
-        if width <= 0 or height <= 0:
+        # 最小 100px 防护：避免前端传入宽高比（如 "9:16"）被 parseInt 截断为极小值
+        if width < 100 or height < 100:
+            logger.warning(
+                "分辨率异常小 (%dx%d)，使用默认值。原始值: %s",
+                width,
+                height,
+                resolution_str,
+            )
             return DEFAULT_OUTPUT_WIDTH, DEFAULT_OUTPUT_HEIGHT
         return width, height
     except (ValueError, TypeError):
@@ -576,7 +583,13 @@ class RenderAdapter:
         try:
             from video_processing.thumbnail_generator import extract_and_upload_cover_frames
 
-            cover_candidates = extract_and_upload_cover_frames(str(result.output_path), plan_id, num_frames=3)
+            # 从 plan config 提取标题文字，叠加到封面候选帧上
+            _title_cfg = (plan_config or {}).get("title", {}) or {}
+            _title_text = (_title_cfg.get("text", "") or "").strip() if _title_cfg.get("enabled", True) else ""
+
+            cover_candidates = extract_and_upload_cover_frames(
+                str(result.output_path), plan_id, num_frames=3, title_text=_title_text
+            )
             if cover_candidates:
                 logger.info(
                     "[render-adapter] 封面候选帧生成成功: plan_id=%s count=%d",

apps/worker/video_processing/thumbnail_generator.py

@@ -120,6 +120,92 @@ def _format_seek_time(seconds: float) -> str:
     return f"{h:02d}:{m:02d}:{s:05.2f}"
 
 
+def _overlay_title_on_image(
+    image_path: str,
+    title_text: str,
+    *,
+    timeout: int = 15,
+) -> str:
+    """在封面图上叠加标题文字（居中、白色、带阴影）。
+
+    使用 FFmpeg drawtext 滤镜，原地覆盖 image_path。
+
+    Args:
+        image_path: 输入图片路径（覆盖写入）
+        title_text: 要叠加的标题文字
+        timeout: 超时时间（秒）
+
+    Returns:
+        处理后的图片路径（与输入相同）
+    """
+    from video_processing.ffmpeg_utils import FFMPEG_BIN, run_ffmpeg
+
+    if not title_text or not title_text.strip():
+        return image_path
+
+    # 转义 drawtext 特殊字符
+    # FFmpeg drawtext 需要转义: ' : % \ [ ]
+    escaped = (
+        title_text.replace("\\", "\\\\")
+        .replace("'", "’")
+        .replace(":", "\\:")
+        .replace("%", "%%")
+        .replace("[", "\\[")
+        .replace("]", "\\]")
+    )
+    # 截断过长标题
+    if len(escaped) > 60:
+        escaped = escaped[:57] + "..."
+
+    # 使用中文字体
+    font_path = "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc"
+
+    # drawtext 滤镜参数：
+    # - 白色文字，字号按画面宽度自适应（约 1/18 宽度）
+    # - 黑色阴影偏移 2px
+    # - 水平居中，垂直偏下（距底部约 15%）
+    drawtext_filter = (
+        f"drawtext=fontfile='{font_path}'"
+        f":text='{escaped}'"
+        f":fontsize=h/14"
+        f":fontcolor=white"
+        f":shadowcolor=black@0.7"
+        f":shadowx=2:shadowy=2"
+        f":x=(w-text_w)/2"
+        f":y=h*0.82-text_h/2"
+        f":borderw=0"
+    )
+
+    tmp_out = image_path + ".tmp.jpg"
+    cmd = [
+        FFMPEG_BIN,
+        "-y",
+        "-i",
+        image_path,
+        "-vf",
+        drawtext_filter,
+        "-q:v",
+        "2",
+        tmp_out,
+    ]
+
+    try:
+        run_ffmpeg(cmd, capture_output=True, timeout=timeout)
+        # 覆盖原文件
+        import shutil
+
+        shutil.move(tmp_out, image_path)
+        logger.info("封面标题叠加成功: text=%s", title_text[:30])
+    except Exception as e:
+        logger.warning("封面标题叠加失败（使用原图）: %s", e)
+        try:
+            Path(tmp_out).unlink(missing_ok=True)
+        except Exception:
+            pass
+
+    return image_path
+
+
 def generate_and_upload_thumbnail(
     video_path: str,
     storage_key: str,
@@ -163,6 +249,7 @@ def extract_cover_candidates(
     *,
     width: int = 640,
     timeout: int = 30,
+    title_text: str = "",
 ) -> list[dict]:
     """在视频时长 25%/50%/75% 处各抽一帧，返回候选帧信息列表。
 
@@ -218,6 +305,9 @@ def extract_cover_candidates(
             run_ffmpeg(cmd, capture_output=True, timeout=timeout)
 
             if Path(output_path).exists() and Path(output_path).stat().st_size > 0:
+                # 叠加标题文字
+                if title_text and title_text.strip():
+                    _overlay_title_on_image(output_path, title_text, timeout=timeout)
                 results.append(
                     {
                         "local_path": output_path,
@@ -237,6 +327,8 @@ def extract_and_upload_cover_frames(
     video_path: str,
     plan_id: str,
     num_frames: int = 3,
+    *,
+    title_text: str = "",
 ) -> list[dict]:
     """抽取封面候选帧并上传到 OSS。
 
@@ -248,7 +340,7 @@ def extract_and_upload_cover_frames(
     Returns:
         [{"image_url": "https://...", "frame_time": 5.0, "storage_key": "covers/xxx/frame_0.jpg"}, ...]
     """
-    candidates = extract_cover_candidates(video_path, num_frames=num_frames)
+    candidates = extract_cover_candidates(video_path, num_frames=num_frames, title_text=title_text)
     if not candidates:
         logger.warning("封面候选帧抽取为空: plan_id=%s", plan_id)
         return []

apps/worker/worker_app/tasks/generation.py

@@ -1123,6 +1123,7 @@ def _render_video(
     resolution: str = "",
     bgm_config: dict | None = None,
     voice_ids: list[str] | None = None,
+    custom_title: str = "",
 ) -> tuple[Path, float]:
     """渲染视频（含配音混音）。
 
@@ -1169,6 +1170,20 @@ def _render_video(
             merged_bgm.get("source", ""),
         )
 
+    # 用户自定义标题覆盖模板标题（用户指定优先级最高）
+    if custom_title and custom_title.strip():
+        plan_cfg = dict(virtual_plan.config or {})
+        title_cfg = dict(plan_cfg.get("title", {}) or {})
+        title_cfg["text"] = custom_title.strip()
+        title_cfg["enabled"] = True
+        plan_cfg["title"] = title_cfg
+        virtual_plan.config = plan_cfg
+        logger.info(
+            "[task_id=%s] [渲染] 用户自定义标题已注入: title=%s",
+            task_id,
+            custom_title[:50],
+        )
+
     # 确保输出分辨率配置存在
     # 优先级：用户指定 > 模板配置 > 默认 1280x720
     # 预览模式：强制 854x480 + 低码率
@@ -1501,6 +1516,16 @@ def generate_video(self, task_id: str) -> dict:
             # 动态分辨率：优先使用 output_width/output_height，其次 resolution 字符串
             _ow = task_info.get("output_width", OUTPUT_WIDTH) or OUTPUT_WIDTH
             _oh = task_info.get("output_height", OUTPUT_HEIGHT) or OUTPUT_HEIGHT
+            # 防护：前端可能误传宽高比（如 parseInt("9:16") = 9），宽度 < 100 时忽略
+            if _ow < 100 or _oh < 100:
+                logger.warning(
+                    "[task_id=%s] output_width/output_height 异常 (%dx%d)，回退到默认",
+                    task_id,
+                    _ow,
+                    _oh,
+                )
+                _ow = OUTPUT_WIDTH
+                _oh = OUTPUT_HEIGHT
             if _ow != OUTPUT_WIDTH or _oh != OUTPUT_HEIGHT:
                 _resolved_resolution = f"{_ow}x{_oh}"
             else:
@@ -1519,6 +1544,7 @@ def generate_video(self, task_id: str) -> dict:
                 resolution=_resolved_resolution,
                 bgm_config=task_info.get("bgm_config", {}),
                 voice_ids=task_info.get("voice_ids", []),
+                custom_title=task_info.get("custom_title", ""),
             )
 
             if gen_task: