auto-save 2026-05-14 04:04 (~6)

api/main.py

@@ -89,6 +89,7 @@ JobStatus = Literal[
 
 KEYFRAME_COUNT = int(os.getenv("KEYFRAME_COUNT", "5"))
 FrameExtractTarget = Literal["balanced", "subject", "transition", "expression", "motion"]
+FrameExtractMode = Literal["replace", "append"]
 FRAME_TARGET_LABELS: dict[FrameExtractTarget, str] = {
     "balanced": "综合关键帧",
     "subject": "清晰主体",
@@ -582,6 +583,7 @@ async def pipeline_analyze(
     job_id: str,
     frame_count: int = KEYFRAME_COUNT,
     target: FrameExtractTarget = "balanced",
+    mode: FrameExtractMode = "replace",
 ) -> None:
     """阶段 2：拆音轨 + 抽关键帧。ASR/翻译是独立文案轨，不阻塞视觉素材流。"""
     job = JOBS[job_id]
@@ -591,13 +593,16 @@ async def pipeline_analyze(
         if not mp4.exists():
             raise RuntimeError("source.mp4 不存在，先完成下载")
 
-        update(job, status="splitting", message="ffmpeg 拆分音轨…", progress=35)
         wav = d / "audio.wav"
-        run([
-            "ffmpeg", "-y", "-i", str(mp4),
-            "-vn", "-ac", "1", "-ar", "16000", "-c:a", "pcm_s16le",
-            str(wav),
-        ])
+        if wav.exists():
+            update(job, status="splitting", message="复用音轨 · 准备抽帧…", progress=35)
+        else:
+            update(job, status="splitting", message="ffmpeg 拆分音轨…", progress=35)
+            run([
+                "ffmpeg", "-y", "-i", str(mp4),
+                "-vn", "-ac", "1", "-ar", "16000", "-c:a", "pcm_s16le",
+                str(wav),
+            ])
 
         n = max(1, min(int(frame_count), 20))
         target_label = FRAME_TARGET_LABELS.get(target, FRAME_TARGET_LABELS["balanced"])
@@ -607,7 +612,9 @@ async def pipeline_analyze(
 
         update(job, message=f"低清扫描候选 · {target_label} · 约 {estimated_scan_count} 帧…", progress=45)
         frames_dir = d / "frames"
-        if frames_dir.exists():
+        replacing = mode == "replace"
+        existing_frames = list(job.frames) if not replacing else []
+        if replacing and frames_dir.exists():
             shutil.rmtree(frames_dir)
         frames_dir.mkdir(parents=True)
         scan_dir = d / "frame_scan"
@@ -637,15 +644,23 @@ async def pipeline_analyze(
             raise RuntimeError("候选帧评分失败")
 
         # 2) 目标化筛选：pHash 去重 + 清晰度 / 中心细节 / 转场变化 / 动作强度 + 时序分桶。
+        selection_count = n if replacing else min(len(candidates), max(n * 4, n + len(existing_frames) + 2))
         update(job, message=f"{target_label}筛选 {n} / {len(candidates)} 张…", progress=60)
-        chosen = _select_keyframes(candidates, n, target)
+        chosen = _select_keyframes(candidates, selection_count, target)
 
         # 3) 只对最终选中的时间点，从原视频抽高质量关键帧。
         renamed: list[KeyFrame] = []
         chosen_sorted = sorted(chosen, key=lambda it: float(it["timestamp"]))
-        for i, item in enumerate(chosen_sorted):
-            dst = frames_dir / f"{i:03d}.jpg"
+        existing_timestamps = [float(f.timestamp) for f in existing_frames]
+        next_idx = max((int(f.index) for f in existing_frames), default=-1) + 1
+        for item in chosen_sorted:
+            if len(renamed) >= n:
+                break
             t = float(item["timestamp"])
+            if not replacing and any(abs(t - old) < 0.35 for old in existing_timestamps):
+                continue
+            idx = next_idx + len(renamed)
+            dst = frames_dir / f"{idx:03d}.jpg"
             run([
                 "ffmpeg", "-y", "-ss", f"{t:.3f}", "-i", str(mp4),
                 "-frames:v", "1",
@@ -653,20 +668,24 @@ async def pipeline_analyze(
                 str(dst),
             ])
             renamed.append(KeyFrame(
-                index=i,
+                index=idx,
                 timestamp=round(t, 2),
-                url=f"/jobs/{job_id}/frames/{i}.jpg",
+                url=f"/jobs/{job_id}/frames/{idx}.jpg",
             ))
+            existing_timestamps.append(t)
 
         # 4) 清理扫描目录
         shutil.rmtree(scan_dir, ignore_errors=True)
 
+        merged_frames = sorted(existing_frames + renamed, key=lambda f: f.timestamp)
+        action_label = "追加" if not replacing else "抽取"
+
         update(
             job,
             status="frames_extracted",
-            frames=renamed,
+            frames=merged_frames,
             progress=70,
-            message=f"已按「{target_label}」抽取 {len(renamed)} 张关键帧 · 可继续清洗 / 提取元素 / 分镜编排",
+            message=f"已按「{target_label}」{action_label} {len(renamed)} 张关键帧 · 共 {len(merged_frames)} 张",
         )
 
     except Exception as e:
@@ -1040,13 +1059,14 @@ async def trigger_analyze(
     bg: BackgroundTasks,
     frames: int = KEYFRAME_COUNT,
     target: FrameExtractTarget = "balanced",
+    mode: FrameExtractMode = "replace",
 ) -> Job:
     job = JOBS.get(job_id)
     if not job:
         raise HTTPException(404, "job not found")
     if job.status not in {"downloaded", "frames_extracted", "transcribed", "failed"}:
         raise HTTPException(409, f"status must be downloaded/failed, got {job.status}")
-    bg.add_task(pipeline_analyze, job_id, frames, target)
+    bg.add_task(pipeline_analyze, job_id, frames, target, mode)
     return job