auto-save 2026-05-17 22:14 (~3)

.memory/worklog.json

@@ -1,18 +1,5 @@
 {
   "entries": [
-    {
-      "files_changed": 1,
-      "hash": "6346329",
-      "message": "auto-save 2026-05-15 13:34 (~1)",
-      "ts": "2026-05-15T13:34:39+08:00",
-      "type": "commit"
-    },
-    {
-      "files_changed": 1,
-      "message": "Codex 会话活跃 · 最近命令：codex · 1 项未提交变更 · 最近提交：auto-save 2026-05-15 13:34 (~1)",
-      "ts": "2026-05-15T05:34:45Z",
-      "type": "session-heartbeat"
-    },
     {
       "files_changed": 1,
       "hash": "765745d",
@@ -3260,6 +3247,19 @@
       "type": "session-heartbeat",
       "message": "Codex 会话活跃 · 最近命令：codex · 分支 main · 1 项未提交变更 · 最近提交：auto-save 2026-05-17 21:58 (~2)",
       "files_changed": 1
+    },
+    {
+      "ts": "2026-05-17T22:03:37+08:00",
+      "type": "commit",
+      "message": "auto-save 2026-05-17 22:03 (~3)",
+      "hash": "87015e9",
+      "files_changed": 3
+    },
+    {
+      "ts": "2026-05-17T14:08:30Z",
+      "type": "session-heartbeat",
+      "message": "Codex 会话活跃 · 最近命令：codex · 分支 main · 1 项未提交变更 · 最近提交：auto-save 2026-05-17 22:03 (~3)",
+      "files_changed": 1
     }
   ]
 }

docs/source-analysis.html

@@ -649,8 +649,8 @@ api/main.py
           </div>
           <div class="flow-row">
             <div><strong>你看到的区域</strong><span>音频解析结果表</span></div>
-            <div><strong>主要源码</strong><span><code>AudioIntakePanel</code> / <code>AudioIntakeStatus</code> / <code>SourceReferenceBuildPanel</code> in <code>web/components/ad-recreation-board.tsx</code>；复用 <code>triggerTranscribe</code>、<code>AudioScript</code>、<code>analyzeJob</code> 和 <code>generateSubjectAssets</code>。</span></div>
+            <div><strong>主要源码</strong><span><code>AudioIntakePanel</code> / <code>AudioIntakeStatus</code> / <code>SourceReferenceBuildPanel</code> in <code>web/components/ad-recreation-board.tsx</code>；复用 <code>triggerTranscribe</code>、<code>AudioScript</code>、<code>analyzeJob</code>、<code>addManualFrame</code>、<code>deleteFrame</code> 和 <code>generateSubjectAssets</code>。</span></div>
-            <div><strong>适合怎么描述</strong><span>“原视频播放、12 张关键帧选择、相似主角 6 白底视图、连续响度波形、逐句时间轴滚动、高亮和跳转联动还需要怎么调整”。</span></div>
+            <div><strong>适合怎么描述</strong><span>“原视频播放尺寸、当前播放点手动抽帧、关键帧删除、相似主角 6 白底视图、连续响度波形、逐句时间轴滚动、高亮和跳转联动还需要怎么调整”。</span></div>
           </div>
           <div class="flow-row">
             <div><strong>你看到的区域</strong><span>信息流复刻分镜工作台</span></div>
@@ -879,7 +879,8 @@ ProductRefStateItem {
             <tr><td>分镜脚本改写</td><td><code>POST /jobs/{id}/script/rewrite</code></td><td><code>rewriteStoryboardScript</code></td><td>根据原参考文案、当前新口播、分镜角色、时间段和作者想法改写中文口播。<code>mode=segment</code> 只改一段；<code>mode=all</code> 一次改完整片，要求整片前后连贯。接口只返回 <code>items[index,text]</code>，前端暂存在当前页面状态里，生成本条视频时写入 <code>StoryboardScene.action</code>。</td></tr>
             <tr><td>原始音频文件</td><td><code>GET /jobs/{id}/audio.wav</code></td><td><code>sourceAudioUrl</code></td><td>返回拆轨得到的 wav；当前主界面不再渲染底部吸附音频条，右侧复刻工作表会读取该文件生成参考图式横向响度波形，并和原视频、逐句时间轴联动。</td></tr>
             <tr><td>改写配音文件</td><td><code>GET /jobs/{id}/audio-script.mp3</code></td><td><code>apiAssetUrl(job.audio_script.voice_url)</code></td><td>后续新配音阶段保留的 MiniMax T2A 产物。当前第一步不默认生成该文件。</td></tr>
-            <tr><td>手动加帧</td><td><code>POST /jobs/{id}/frames?t=</code></td><td><code>addManualFrame</code></td><td>按视频时间戳抽一帧，index 递增但 frames 按 timestamp 排序。</td></tr>
+            <tr><td>手动加帧</td><td><code>POST /jobs/{id}/frames?t=</code></td><td><code>addManualFrame</code></td><td>按视频时间戳抽一帧，index 递增但 frames 按 timestamp 排序。当前主界面会把原版视频播放秒数传给 <code>SourceReferenceBuildPanel</code> 的“当前点抽帧”。</td></tr>
+            <tr><td>删除关键帧</td><td><code>DELETE /jobs/{id}/frames/{idx}</code></td><td><code>deleteFrame</code></td><td>删除单张关键帧并清掉对应选择态；当前主界面每张缩略图右下角提供删除入口，方便手动抽错后直接修正。</td></tr>
             <tr><td>Vision 识别</td><td><code>POST /frames/{idx}/describe</code></td><td><code>describeFrame</code></td><td>写入 frame.description，后续可从 objects 加候选元素。</td></tr>
             <tr><td>清洗水印</td><td><code>POST /frames/{idx}/cleanup</code></td><td><code>cleanupFrame</code></td><td>支持全图和区域清洗，生成 cleaned 待应用版本；前端批量清洗会顺序调用该接口，不自动覆盖原图。单帧清洗状态按 frame.index 隔离，清洗某一张不会禁用其他关键帧的清洗按钮。</td></tr>
             <tr><td>应用清洗版</td><td><code>POST /frames/{idx}/cleanup/apply</code></td><td><code>applyCleanedFrame</code></td><td>把 cleaned 待应用版本覆盖到原关键帧，并保留首次原图备份；前端“一键替换待应用”会顺序调用该接口应用所有已有清洗版。</td></tr>
@@ -1002,6 +1003,18 @@ ProductRefStateItem {
         <h2>变更记录</h2>
         <p>这个记录不是 git log 的替代品。它记录“产品理解发生了什么变化、影响了哪些源码、你以后描述需求时该怎么说”。后续每次改功能都要补一条。</p>
         <div class="changelog">
+          <article class="change">
+            <header>
+              <h3>2026-05-17 · 原版视频放大并支持当前点抽帧删除</h3>
+              <span class="tag rose">UI</span>
+              <span class="tag cyan">Workflow</span>
+            </header>
+            <div class="body">
+              <p><strong>问题：</strong>原版视频区域偏小，只能看和播放；如果需要补某个具体画面，用户必须重新跑 12 帧抽取，并且抽错后缺少直接删除入口。</p>
+              <p><strong>改动：</strong><code>AudioIntakePanel</code> 放大原视频列和视频高度，并把逐句时间轴压成较窄侧栏；<code>SourceReferenceBuildPanel</code> 接收当前播放秒数，新增“当前点抽帧”按钮，复用 <code>addManualFrame</code>；关键帧缩略图右下角新增删除按钮，复用 <code>deleteFrame</code>，并展示全部已抽帧而不只截取前 12 张。</p>
+              <p><strong>影响：</strong><code>web/components/ad-recreation-board.tsx</code>、<code>docs/source-analysis.html</code>。后续描述这里时，应把原版视频旁的关键帧区理解为“人物参考帧池”，支持自动抽人物、按播放点补帧和单帧删除。</p>
+            </div>
+          </article>
           <article class="change">
             <header>
               <h3>2026-05-17 · 主动抽帧改成人物主体并压缩时间轴侧栏</h3>

web/components/ad-recreation-board.tsx

@@ -851,6 +851,8 @@ export function AdRecreationBoard({
                 selectedFrames={data.selectedFrames}
                 onToggleFrame={data.onToggleFrame}
                 onJobUpdate={data.onJobUpdate}
+                onAddFrame={data.onAddManualFrameForJob}
+                onDeleteFrame={data.onDeleteFrameForJob}
               />
               <AudioStoryboardPlanPanel
                 job={job}
@@ -990,11 +992,15 @@ function AudioIntakePanel({
   selectedFrames,
   onToggleFrame,
   onJobUpdate,
+  onAddFrame,
+  onDeleteFrame,
 }: {
   job: Job | null
   selectedFrames: Set<number>
   onToggleFrame: (idx: number) => void
   onJobUpdate: (job: Job) => void
+  onAddFrame?: (jobId: string, t: number) => Promise<void> | void
+  onDeleteFrame?: (jobId: string, idx: number) => Promise<void> | void
 }) {
   const [currentTime, setCurrentTime] = useState(0)
   const [mediaDuration, setMediaDuration] = useState(0)
@@ -1126,7 +1132,7 @@ function AudioIntakePanel({
           />
         </div>
 
-        <div className="grid gap-2 xl:grid-cols-[300px_minmax(340px,1fr)_430px] 2xl:grid-cols-[340px_minmax(520px,1fr)_480px]">
+        <div className="grid gap-2 xl:grid-cols-[340px_minmax(340px,1fr)_320px] 2xl:grid-cols-[410px_minmax(500px,1fr)_320px]">
           <div className="min-w-0">
             <div className="mb-2 flex items-center justify-between gap-3">
               <SectionTitle icon={<Play className="h-4 w-4" />} title="原版视频" />
@@ -1138,7 +1144,7 @@ function AudioIntakePanel({
                   ref={videoRef}
                   controls
                   playsInline
-                  className="h-[238px] w-full bg-black object-contain"
+                  className="h-[270px] w-full bg-black object-contain 2xl:h-[300px]"
                   src={videoSrcUrl}
                   onTimeUpdate={(event) => setCurrentTime(event.currentTarget.currentTime)}
                   onSeeked={(event) => setCurrentTime(event.currentTarget.currentTime)}
@@ -1152,7 +1158,7 @@ function AudioIntakePanel({
                   }}
                 />
               ) : (
-                <div className="flex h-[238px] items-center justify-center text-[12px] text-white/38">等待原视频</div>
+                <div className="flex h-[270px] items-center justify-center text-[12px] text-white/38 2xl:h-[300px]">等待原视频</div>
               )}
             </div>
           </div>
@@ -1162,6 +1168,10 @@ function AudioIntakePanel({
             selectedFrames={selectedFrames}
             onToggleFrame={onToggleFrame}
             onJobUpdate={onJobUpdate}
+            currentTime={currentTime}
+            duration={timelineDuration}
+            onAddFrame={onAddFrame}
+            onDeleteFrame={onDeleteFrame}
           />
 
           <div className="min-w-0">
@@ -1176,7 +1186,7 @@ function AudioIntakePanel({
                   <div>原文</div>
                   <div>中文</div>
                 </div>
-                <div className="max-h-[238px] overflow-y-auto">
+                <div className="max-h-[270px] overflow-y-auto 2xl:max-h-[300px]">
                   {job.transcript.map((segment) => {
                     const active = activeSegment?.index === segment.index
                     return (
@@ -1213,14 +1223,24 @@ function SourceReferenceBuildPanel({
   selectedFrames,
   onToggleFrame,
   onJobUpdate,
+  currentTime,
+  duration,
+  onAddFrame,
+  onDeleteFrame,
 }: {
   job: Job
   selectedFrames: Set<number>
   onToggleFrame: (idx: number) => void
   onJobUpdate: (job: Job) => void
+  currentTime: number
+  duration: number
+  onAddFrame?: (jobId: string, t: number) => Promise<void> | void
+  onDeleteFrame?: (jobId: string, idx: number) => Promise<void> | void
 }) {
   const [extracting, setExtracting] = useState(false)
+  const [manualBusy, setManualBusy] = useState(false)
   const [subjectBusy, setSubjectBusy] = useState(false)
+  const [deletingFrame, setDeletingFrame] = useState<number | null>(null)
   const frames = useMemo(() => [...job.frames].sort((a, b) => a.timestamp - b.timestamp), [job.frames])
   const selectedReferenceFrames = useMemo(
     () => frames.filter((frame) => selectedFrames.has(frame.index)),
@@ -1239,6 +1259,7 @@ function SourceReferenceBuildPanel({
     return null
   }, [frames, selectedReferenceFrames])
   const actorAssets = actorSource?.element.subject_assets ?? []
+  const manualFrameTime = clampNumber(currentTime, 0, Math.max(duration, 1))
 
   const extractKeyframes = async () => {
     setExtracting(true)
@@ -1301,6 +1322,26 @@ function SourceReferenceBuildPanel({
     }
   }
 
+  const addFrameAtCurrentTime = async () => {
+    if (!onAddFrame) return
+    setManualBusy(true)
+    try {
+      await onAddFrame(job.id, manualFrameTime)
+    } finally {
+      setManualBusy(false)
+    }
+  }
+
+  const deleteReferenceFrame = async (idx: number) => {
+    if (!onDeleteFrame) return
+    setDeletingFrame(idx)
+    try {
+      await onDeleteFrame(job.id, idx)
+    } finally {
+      setDeletingFrame(null)
+    }
+  }
+
   return (
     <div className="min-w-0">
       <div className="mb-2 flex items-center justify-between gap-3">
@@ -1309,8 +1350,8 @@ function SourceReferenceBuildPanel({
           {frames.length ? `${frames.length} 张` : "待抽帧"} · 已选 {selectedReferenceFrames.length}
         </span>
       </div>
-      <div className="h-[238px] overflow-y-auto rounded-md border border-white/10 bg-black/32 p-2">
+      <div className="h-[270px] overflow-y-auto rounded-md border border-white/10 bg-black/32 p-2 2xl:h-[300px]">
-        <div className="grid grid-cols-2 gap-1.5">
+        <div className="grid grid-cols-3 gap-1.5">
           <button
             type="button"
             onClick={() => void extractKeyframes()}
@@ -1320,6 +1361,16 @@ function SourceReferenceBuildPanel({
             {extracting || job.status === "splitting" ? <Loader2 className="h-3.5 w-3.5 animate-spin" /> : <Scissors className="h-3.5 w-3.5" />}
             抽人物 12 帧
           </button>
+          <button
+            type="button"
+            onClick={() => void addFrameAtCurrentTime()}
+            disabled={!job.video_url || !onAddFrame || manualBusy || job.status === "splitting"}
+            title={`按当前播放位置手动抽帧：${manualFrameTime.toFixed(1)}s`}
+            className="inline-flex h-7 items-center justify-center gap-1 rounded-md border border-white/10 bg-white/[0.055] px-2 text-[10.5px] font-semibold text-white/66 transition hover:border-emerald-300/35 hover:text-emerald-100 disabled:cursor-not-allowed disabled:opacity-35"
+          >
+            {manualBusy ? <Loader2 className="h-3.5 w-3.5 animate-spin" /> : <Plus className="h-3.5 w-3.5" />}
+            当前点抽帧
+          </button>
           <button
             type="button"
             onClick={() => void generateSimilarActor()}
@@ -1332,29 +1383,48 @@ function SourceReferenceBuildPanel({
         </div>
 
         <div className="mt-2 grid grid-cols-4 gap-1.5">
-          {frames.slice(0, 12).map((frame, index) => {
+          {frames.map((frame, index) => {
             const selected = selectedFrames.has(frame.index)
             return (
-              <button
+              <div
                 key={frame.index}
-                type="button"
-                onClick={() => onToggleFrame(frame.index)}
                 className={`group relative h-12 overflow-hidden rounded border bg-black transition ${
                   selected ? "border-emerald-300/70" : "border-white/10 hover:border-cyan-300/40"
                 }`}
                 title={`关键帧 ${index + 1} · ${frame.timestamp.toFixed(1)}s`}
               >
-                <img src={effectiveFrameUrl(job.id, frame)} alt="" className="h-full w-full object-cover" />
+                <button
+                  type="button"
+                  onClick={() => onToggleFrame(frame.index)}
+                  className="absolute inset-0 cursor-pointer overflow-hidden focus:outline-none focus:ring-1 focus:ring-cyan-200/70"
+                >
+                  <img src={effectiveFrameUrl(job.id, frame)} alt="" className="h-full w-full object-cover" />
+                </button>
                 <span className="absolute left-1 top-1 rounded bg-black/72 px-1 font-mono text-[9px] text-white/70">{String(index + 1).padStart(2, "0")}</span>
                 <span className="absolute right-1 top-1 rounded-full bg-black/72 p-0.5">
                   {selected ? <Check className="h-3 w-3 text-emerald-200" /> : <Circle className="h-3 w-3 text-white/50" />}
                 </span>
-              </button>
+                {onDeleteFrame && (
+                  <button
+                    type="button"
+                    onClick={(event) => {
+                      event.stopPropagation()
+                      void deleteReferenceFrame(frame.index)
+                    }}
+                    disabled={deletingFrame === frame.index}
+                    className="absolute bottom-1 right-1 inline-flex h-6 w-6 items-center justify-center rounded-full border border-rose-200/25 bg-black/78 text-rose-100 opacity-80 transition hover:border-rose-200/55 hover:bg-rose-500/25 hover:opacity-100 focus:opacity-100 focus:outline-none focus:ring-1 focus:ring-rose-100/70 disabled:cursor-not-allowed disabled:opacity-60"
+                    aria-label={`删除关键帧 ${index + 1}`}
+                    title="删除这张关键帧"
+                  >
+                    {deletingFrame === frame.index ? <Loader2 className="h-3.5 w-3.5 animate-spin" /> : <Trash2 className="h-3.5 w-3.5" />}
+                  </button>
+                )}
+              </div>
             )
           })}
           {!frames.length && (
             <div className="col-span-4 flex h-[106px] items-center justify-center rounded border border-dashed border-white/12 text-[11px] text-white/34">
-              点击“抽人物 12 帧”后，这里会展示原视频里主体人物清晰的关键画面。
+              点击“抽人物 12 帧”，或播放到指定时间后用“当前点抽帧”补充人物参考。
             </div>
           )}
         </div>