auto-save 2026-05-14 12:37 (~6)

.memory/worklog.json

@@ -1,19 +1,5 @@
 {
   "entries": [
-    {
-      "files_changed": 1,
-      "hash": "714db7d",
-      "message": "auto-save 2026-05-13 05:46 (~1)",
-      "ts": "2026-05-13T05:46:19+08:00",
-      "type": "commit"
-    },
-    {
-      "files_changed": 1,
-      "hash": "91d2d64",
-      "message": "auto-save 2026-05-13 05:52 (~1)",
-      "ts": "2026-05-13T05:52:14+08:00",
-      "type": "commit"
-    },
     {
       "files_changed": 1,
       "hash": "4a9264a",
@@ -3289,6 +3275,19 @@
       "type": "session-heartbeat",
       "message": "Codex 会话活跃 · 最近命令：codex · 2 项未提交变更 · 最近提交：auto-save 2026-05-14 12:26 (~4)",
       "files_changed": 2
+    },
+    {
+      "ts": "2026-05-14T12:32:00+08:00",
+      "type": "commit",
+      "message": "auto-save 2026-05-14 12:31 (~2)",
+      "hash": "01ab67e",
+      "files_changed": 2
+    },
+    {
+      "ts": "2026-05-14T04:36:11Z",
+      "type": "session-heartbeat",
+      "message": "Codex 会话活跃 · 最近命令：codex · 6 项未提交变更 · 最近提交：auto-save 2026-05-14 12:31 (~2)",
+      "files_changed": 6
     }
   ]
 }

docs/source-analysis.html

@@ -629,7 +629,7 @@ api/main.py
           </div>
           <div class="flow-row">
             <div><strong>你看到的区域</strong><span>关键帧素材审核面板</span></div>
-            <div><strong>主要源码</strong><span><code>FrameLightbox</code>；按“原图/清洗、主体资产、首尾帧、产品融合、审核”五个页签组织；左侧只放主图/框选画布，但主体资产页左侧改为全部已清洗/已选参考帧网格，首尾帧页左侧显示全部关键帧并可勾选人物/机位参考。主体识别页会显示透明骨架人目标和 Vision 验收分数。清洗页右侧支持一键清洗未处理帧、单张替换清洗版和一键替换全部待应用清洗版；批量替换顺序调用 <code>applyCleanedFrame</code>，不新增后端接口。产品融合页左侧是纵向 6 行镜头工作表：每行直接显示首帧、尾帧、同一产品 4 个角度图、描述词、秒数和单条生成按钮，便于一次看完 6 条视频。产品融合槽位的“粘贴”优先使用应用内 <code>clipboard</code>，也支持选中槽位后 Cmd+V 粘贴系统图片。右侧保留 GPT Image 2 / Seedance 固定模型、当前镜头状态、桌面四角度一键填充、20 条产品使用描述模板、批量排队和产品图库选用；产品图库选中后会填入当前镜头下一个产品角度槽。主体资产页只确认一个统一主体，后端按参考重绘六张纯背景、占满画面的标准站立透明骨架人资产图；首尾帧页通过地点、风格、参考要素和可编辑 prompt 做文字生图，生成结果写入 <code>scene_assets</code> 但以 <code>asset_role=first_frame/last_frame</code> 标记，并自动传入当前产品融合镜头。相关接口包括 <code>cleanupFrame</code>、<code>applyCleanedFrame</code>、<code>addElement</code>、<code>generateSubjectAssets</code>、<code>generateSceneAsset</code>、<code>listProductLibrary</code>、<code>copyProductLibraryAsset</code> 和 <code>generateProductFusionDescriptions</code>。</span></div>
+            <div><strong>主要源码</strong><span><code>FrameLightbox</code>；按“原图/清洗、主体资产、首尾帧、产品融合、审核”五个页签组织；左侧只放主图/框选画布，但主体资产页左侧改为全部已清洗/已选参考帧网格，首尾帧页左侧显示全部关键帧并可勾选人物/机位参考。主体识别页会显示透明骨架人目标和 Vision 验收分数。清洗页右侧支持一键清洗未处理帧、单张替换清洗版和一键替换全部待应用清洗版；批量替换顺序调用 <code>applyCleanedFrame</code>，不新增后端接口。产品融合页左侧是纵向 6 行镜头工作表：每行只显示首帧、尾帧、已预填动作描述、秒数、生成按钮和对应视频结果；四张桌面 SKG 产品图作为固定产品参考，生成时通过 <code>copyProductLibraryAsset</code> 自动写入镜头，不再暴露产品角度槽、产品融合辅助栏或产品图库选择器。产品融合槽位的“粘贴”优先使用应用内 <code>clipboard</code>，也支持选中槽位后 Cmd+V 粘贴系统图片。主体资产页只确认一个统一主体，后端按参考重绘六张纯背景、占满画面的标准站立透明骨架人资产图；首尾帧页通过地点、风格、参考要素和可编辑 prompt 做文字生图，生成结果写入 <code>scene_assets</code> 但以 <code>asset_role=first_frame/last_frame</code> 标记，并自动传入当前产品融合镜头。相关接口包括 <code>cleanupFrame</code>、<code>applyCleanedFrame</code>、<code>addElement</code>、<code>generateSubjectAssets</code>、<code>generateSceneAsset</code> 和 <code>copyProductLibraryAsset</code>。</span></div>
             <div><strong>适合怎么描述</strong><span>“这一组关键帧如何共同生成一个统一主体包；某张关键帧的水印、去主体场景图、产品融合镜头组和质量风险应该如何审核”。</span></div>
           </div>
           <div class="flow-row">
@@ -748,7 +748,7 @@ SubjectAsset {
           </div>
           <div class="card">
             <h3>ProductFusionShot</h3>
-            <p>产品融合镜头组的单行数据。每个关键帧最多 6 行，首帧、尾帧、四张同一产品不同角度图、动作描述和秒数一一对应；生成时直接把首尾帧和产品角度图作为 Seedance 垫图提交。</p>
+            <p>产品融合镜头组的单行数据。每个关键帧最多 6 行，用户只补首帧、尾帧和必要时微调动作描述、秒数；四张桌面 SKG 产品角度图固定隐藏填充，生成时直接把首尾帧和固定产品图作为 Seedance 垫图提交。</p>
             <pre>ProductFusionShot {
   id,
   first_image,
@@ -806,7 +806,7 @@ SubjectAsset {
             <tr><td>产品图库</td><td><code>GET /product-library/skg</code></td><td><code>listProductLibrary</code></td><td>读取内置 SKG 白底图库 manifest，返回产品标题、品类、尺寸、白底评分和预览图 URL。</td></tr>
             <tr><td>产品图入库到 job</td><td><code>POST /jobs/{id}/assets/product-library</code></td><td><code>copyProductLibraryAsset</code></td><td>把一个内置产品图库条目复制为当前 job 的普通 asset，返回 <code>ImageRef(kind="asset")</code>，用于画面工作台产品融合和分镜产品参考组。</td></tr>
             <tr><td>产品融合引导图</td><td><code>POST /jobs/{id}/product-fusion/guide</code></td><td><code>createProductFusionGuide</code></td><td>旧流程兼容接口：读取产品图和白底人物图，按 <code>product_region</code> 合成位置引导图。当前首尾帧流程不再主动调用它。</td></tr>
-            <tr><td>产品融合描述词</td><td><code>POST /jobs/{id}/product-fusion/descriptions</code></td><td><code>generateProductFusionDescriptions</code></td><td>生成 20 条产品融合动作描述库，前端每次按 6 条轮换套用到 6 行镜头；输入重点变为首帧、尾帧和四张产品角度图，有 LLM 配置时用 <code>REWRITE_MODEL</code> 生成 JSON，无配置或失败时回退到本地 20 条精准模板。</td></tr>
+            <tr><td>产品融合描述词</td><td><code>POST /jobs/{id}/product-fusion/descriptions</code></td><td><code>generateProductFusionDescriptions</code></td><td>兼容接口：可生成 20 条产品融合动作描述库。当前前端默认直接用本地 20 条精准模板预填 6 行镜头，不再显示单独的 AI 草拟入口。</td></tr>
             <tr><td>分镜保存</td><td><code>PUT /frames/{idx}/storyboard</code></td><td><code>updateStoryboard</code></td><td>保存 4 图槽、时长和改造说明。</td></tr>
             <tr><td>生图</td><td><code>POST /frames/{idx}/generate</code></td><td><code>generateImage</code></td><td>基于关键帧或已选生成图做 image-to-image，目前可用。</td></tr>
           </tbody>
@@ -925,7 +925,7 @@ SubjectAsset {
             </header>
             <div class="body">
               <p><strong>问题：</strong>产品融合视频的动作描述不能泛泛写“人物使用产品”，需要稳定表达透明骨架人在具体场景中佩戴 SKG 产品，并呈现舒适享受状态。</p>
-              <p><strong>改动：</strong>前端内置 20 条产品使用描述模板，覆盖卧室、客厅、办公、浴室、阳台、影棚、阅读角等场景；“AI 草拟 6 条”每次从 20 条中按 6 条轮换套用，便于多次生成不同镜头组。</p>
+              <p><strong>改动：</strong>前端内置 20 条产品使用描述模板，覆盖卧室、客厅、办公、浴室、阳台、影棚、阅读角等场景；当前产品融合页默认把前 6 条预填到 6 行镜头，用户只在需要时直接改每行描述。</p>
               <p><strong>后端：</strong><code>generateProductFusionDescriptions</code> 的兜底模板同步扩为 20 条，LLM 提示也改为生成 20 条 35-70 字描述，要求包含场景、佩戴/展示动作和舒适表情，同时排除医疗治疗承诺。</p>
               <p><strong>影响：</strong><code>api/main.py</code>、<code>web/components/lightbox.tsx</code>、<code>docs/source-analysis.html</code>。</p>
             </div>
@@ -1089,6 +1089,18 @@ SubjectAsset {
               <p><strong>影响：</strong><code>web/components/lightbox.tsx</code>、<code>docs/source-analysis.html</code>。</p>
             </div>
           </article>
+          <article class="change">
+            <header>
+              <h3>2026-05-14 · 产品融合收敛为首尾帧 + 固定四产品图</h3>
+              <span class="tag violet">FrameLightbox</span>
+              <span class="tag orange">产品融合</span>
+            </header>
+            <div class="body">
+              <p><strong>问题：</strong>产品融合页继续显示产品角度槽、辅助栏和产品图库会把操作变复杂；当前工作流只需要用户手动补人物首尾帧，产品图固定来自桌面 4 张 SKG 图。</p>
+              <p><strong>改动：</strong>“产品融合”页每行只保留首帧、尾帧、已预填描述词、秒数、生成按钮和行末视频结果。生成单条或批量视频前，前端自动把内置的 4 张桌面 SKG 产品图复制为当前 job asset 并写入 <code>product_images[4]</code>；视频 prompt 增加 <code>产品融合镜头ID</code> 标记，用来把生成结果显示回对应行。</p>
+              <p><strong>影响：</strong><code>web/components/lightbox.tsx</code>、<code>web/app/page.tsx</code>、<code>web/components/nodes/index.tsx</code>、<code>web/components/dashboard.tsx</code>、<code>docs/source-analysis.html</code>。</p>
+            </div>
+          </article>
           <article class="change">
             <header>
               <h3>2026-05-14 · 产品融合镜头组改为纵向 6 行工作表</h3>

web/app/page.tsx

@@ -522,13 +522,14 @@ export default function Home() {
     if (!frame) return
     const productRefs = (shot.product_images ?? []).filter(Boolean).slice(0, 4) as ImageRef[]
     if (!shot.first_image || !shot.last_image || productRefs.length < 4 || !shot.action_text?.trim()) {
-      toast.error("产品融合镜头缺少首帧、尾帧、四张产品角度图或描述词")
+      toast.error("产品融合镜头缺少首帧、尾帧、固定产品图或描述词")
       return
     }
     const duration = shot.duration && shot.duration > 0 ? shot.duration : 5
     const labelOf = (ref?: ImageRef | null, fallback = "未提供") => ref?.label || fallback
     try {
       const prompt = [
+        `产品融合镜头ID：${shot.id || `shot-${frameIdx + 1}`}`,
         `竖屏 9:16，${duration.toFixed(1)} 秒，Seedance 图生视频。`,
         "图片模型固定为 GPT Image 2：首帧和尾帧已经由文字生图生成，用来锁定透明骨架人角色、场景构图和动作起止状态。",
         "视频模型固定为 Seedance：使用首帧作为起始画面、尾帧作为结束画面，并用四张同一 SKG 产品不同角度白底图作为垫图/产品身份参考。",

web/components/dashboard.tsx

@@ -320,6 +320,7 @@ export const Dashboard = forwardRef<DashboardHandle, Props>(function Dashboard({
                     embedded
                     jobId={data.job.id}
                     frames={data.job.frames}
+                    generatedVideos={data.job.generated_videos ?? []}
                     activeIndex={data.expandedFrame}
                     selected={data.selectedFrames}
                     onClose={data.onCloseExpandedFrame}
@@ -333,6 +334,7 @@ export const Dashboard = forwardRef<DashboardHandle, Props>(function Dashboard({
                     clipboard={data.clipboard}
                     onCopyImage={data.onCopyImage}
                     onGenerateProductFusionVideo={data.onGenerateProductFusionVideo}
+                    onDeleteVideo={data.onDeleteVideo}
                   />
                 ) : (
                   renderSection(t.key)

web/components/lightbox.tsx

@@ -5,7 +5,7 @@ import { X, ChevronLeft, ChevronRight, Check, Sparkles, Wand2, Loader2, Eye, Ref
 import {
   frameUrl, cleanedFrameUrl, apiAssetUrl,
   describeFrame, cleanupFrame, applyCleanedFrame, discardCleanedFrame, addElement, updateElement, deleteElement,
-  generateSceneAsset, generateSubjectAssets, generateProductFusionDescriptions, resolveImageRefUrl, uploadStoryboardAsset, updateStoryboard, copyProductLibraryAsset,
+  generateSceneAsset, generateSubjectAssets, resolveImageRefUrl, uploadStoryboardAsset, updateStoryboard, copyProductLibraryAsset,
   type AssetBackground, type AssetSize, type KeyFrame, type Job, type ImageRef, type ProductFusionShot, type SceneAssetRole, type SceneStyle, type SubjectKind,
 } from "@/lib/api"
 import { TRANSPARENT_HUMAN_FRAME_STANDARD, TRANSPARENT_HUMAN_UI_SUMMARY } from "@/lib/workflow-target"
@@ -200,8 +200,6 @@ export function FrameLightbox({ jobId, frames, generatedVideos = [], activeIndex
   const [fusionUploadTarget, setFusionUploadTarget] = useState<FusionUploadTarget | null>(null)
   const [fusionGenerating, setFusionGenerating] = useState<number | "all" | null>(null)
   const [fusionSaving, setFusionSaving] = useState(false)
-  const [fusionFillingProducts, setFusionFillingProducts] = useState<"current" | "all" | null>(null)
-  const [fusionDraftPage, setFusionDraftPage] = useState(0)
   const [editingElement, setEditingElement] = useState<{
     frameIndex: number
     id: string
@@ -338,7 +336,6 @@ export function FrameLightbox({ jobId, frames, generatedVideos = [], activeIndex
     sceneExtraKeywords.trim() ? `额外关键词：${sceneExtraKeywords.trim()}。` : "",
     "要求：单一透明骨架人清晰可见，人物占画面主体，首尾帧可连续生成视频；无文字、水印、平台 UI、恐怖解剖感。",
   ].filter(Boolean).join("\n")
-  const currentFusionShot = fusionShots[activeFusionShot] ?? fusionShots[0]
   const fusionReadyCount = fusionShots.filter((shot) =>
     shot.first_image && shot.last_image && shot.action_text?.trim()
   ).length
@@ -397,31 +394,7 @@ export function FrameLightbox({ jobId, frames, generatedVideos = [], activeIndex
     requestAnimationFrame(() => fusionFileInputRef.current?.click())
   }
 
-  const draftFusionDescriptions = async () => {
-    const actions = PRODUCT_FUSION_DESCRIPTION_PRESETS
-    let descriptions = actions
-    try {
-      const result = await generateProductFusionDescriptions(jobId, fusionShots)
-      descriptions = result.descriptions.length >= PRODUCT_FUSION_DESCRIPTION_PRESETS.length ? result.descriptions : actions
-    } catch (e) {
-      toast.error("AI 描述生成失败，已使用本地草稿")
-    }
-    const start = (fusionDraftPage * FUSION_SHOT_COUNT) % descriptions.length
-    const selectedDescriptions = Array.from({ length: FUSION_SHOT_COUNT }, (_, i) => (
-      descriptions[(start + i) % descriptions.length] || actions[i]
-    ))
-    const next = fusionShots.map((shot, i) => ({
-      ...shot,
-      action_text: selectedDescriptions[i] || shot.action_text || actions[i],
-    }))
-    setFusionShots(next)
-    setFusionDraftPage((prev) => prev + 1)
-    void persistFusionShots(next)
-    toast.success(`已套用 6 条动作描述 · 模板 ${start + 1}-${Math.min(start + FUSION_SHOT_COUNT, descriptions.length)}`)
-  }
-
   const ensureFixedProductAngles = async (indexes: number[]) => {
-    setFusionFillingProducts("all")
     try {
       const reusableRefs = fusionShots.find((shot) => (shot.product_images ?? []).filter(Boolean).length >= PRODUCT_ANGLE_COUNT)
         ?.product_images?.slice(0, PRODUCT_ANGLE_COUNT)
@@ -439,8 +412,6 @@ export function FrameLightbox({ jobId, frames, generatedVideos = [], activeIndex
     } catch (e) {
       toast.error("桌面产品角度填充失败：" + (e instanceof Error ? e.message : String(e)))
       return null
-    } finally {
-      setFusionFillingProducts(null)
     }
   }
 
@@ -1628,94 +1599,6 @@ export function FrameLightbox({ jobId, frames, generatedVideos = [], activeIndex
               </div>
             </section>
           )}
-          {activeTab === "product" && (
-            <>
-              <section className="rounded-lg border border-amber-300/18 bg-amber-500/[0.08] p-2.5 text-[10.5px] leading-relaxed text-white/58">
-                <div className="mb-2 flex items-center justify-between gap-2">
-                  <div className="text-[12px] font-semibold text-white">产品融合辅助</div>
-                  <span className="text-[9px] text-white/40">{fusionSaving ? "保存中" : "自动保存"}</span>
-                </div>
-                <div className="mb-2 grid grid-cols-2 gap-1.5">
-                  <div className="rounded border border-white/10 bg-black/25 px-2 py-1">
-                    <div className="text-[9px] text-white/35">图片模型</div>
-                    <div className="text-white/80">GPT Image 2</div>
-                  </div>
-                  <div className="rounded border border-white/10 bg-black/25 px-2 py-1">
-                    <div className="text-[9px] text-white/35">视频模型</div>
-                    <div className="text-white/80">Seedance</div>
-                  </div>
-                </div>
-                <div className="mb-2 rounded-md border border-white/10 bg-black/25 px-2 py-1.5">
-                  <div className="mb-1 flex items-center justify-between gap-2">
-                    <span className="font-mono text-[10px] text-amber-100">当前镜头 {activeFusionShot + 1}</span>
-                    <span className="text-[9px] text-white/38">{currentFusionShot?.duration ?? 5}s</span>
-                  </div>
-                  <div className="grid grid-cols-2 gap-x-2 gap-y-1 text-[9.5px]">
-                    <span className={currentFusionFirstUrl ? "text-emerald-200/80" : "text-white/35"}>首帧</span>
-                    <span className={currentFusionLastUrl ? "text-emerald-200/80" : "text-white/35"}>尾帧</span>
-                    <span className={currentFusionProductCount >= PRODUCT_ANGLE_COUNT ? "text-emerald-200/80" : "text-white/35"}>产品角度 {currentFusionProductCount}/{PRODUCT_ANGLE_COUNT}</span>
-                    <span className={currentFusionShot?.action_text?.trim() ? "text-emerald-200/80" : "text-white/35"}>描述词</span>
-                  </div>
-                  <div className={`mt-1 truncate text-[9.5px] ${currentFusionShot?.action_text?.trim() ? "text-white/58" : "text-white/32"}`}>
-                    {currentFusionShot?.action_text?.trim() || "描述词未填写"}
-                  </div>
-                </div>
-                <div className="mb-2 grid grid-cols-2 gap-1.5">
-                  <button
-                    type="button"
-                    onClick={() => void fillDesktopProductAngles("current")}
-                    disabled={!!fusionFillingProducts}
-                    className="rounded-md border border-amber-300/20 bg-amber-500/15 px-2 py-1.5 text-[10.5px] font-medium text-amber-50 transition hover:bg-amber-500/25 disabled:cursor-wait disabled:opacity-45 inline-flex items-center justify-center gap-1"
-                  >
-                    {fusionFillingProducts === "current" ? <Loader2 className="h-3 w-3 animate-spin" /> : <Upload className="h-3 w-3" />}
-                    填当前四角度
-                  </button>
-                  <button
-                    type="button"
-                    onClick={() => void fillDesktopProductAngles("all")}
-                    disabled={!!fusionFillingProducts}
-                    className="rounded-md border border-amber-300/20 bg-amber-500/15 px-2 py-1.5 text-[10.5px] font-medium text-amber-50 transition hover:bg-amber-500/25 disabled:cursor-wait disabled:opacity-45 inline-flex items-center justify-center gap-1"
-                  >
-                    {fusionFillingProducts === "all" ? <Loader2 className="h-3 w-3 animate-spin" /> : <Upload className="h-3 w-3" />}
-                    填满 6 镜头
-                  </button>
-                </div>
-                <div className="grid grid-cols-2 gap-1.5">
-                  <button
-                    type="button"
-                    onClick={draftFusionDescriptions}
-                    className="rounded-md bg-white/10 px-2 py-1.5 text-[11px] font-medium text-white/75 transition hover:bg-white/18 hover:text-white inline-flex items-center justify-center gap-1"
-                  >
-                    <Sparkles className="h-3 w-3" />
-                    AI 草拟 6 条
-                  </button>
-                  <button
-                    type="button"
-                    onClick={() => void runAllFusionVideos()}
-                    disabled={!!fusionGenerating || !onGenerateProductFusionVideo}
-                    className="rounded-md bg-white/10 px-2 py-1.5 text-[11px] font-medium text-white/75 transition hover:bg-white/18 hover:text-white disabled:cursor-not-allowed disabled:opacity-40 inline-flex items-center justify-center gap-1"
-                  >
-                    {fusionGenerating === "all" ? <Loader2 className="h-3 w-3 animate-spin" /> : <Sparkles className="h-3 w-3" />}
-                    批量排队
-                  </button>
-                </div>
-              </section>
-              <ProductLibraryPicker
-                jobId={jobId}
-                compact
-                buttonLabel="选用"
-                title={`镜头 ${activeFusionShot + 1} 产品角度图`}
-                onPick={(ref) => {
-                  const nextEmpty = PRODUCT_ANGLE_LABELS.findIndex((_, idx) => !currentFusionProducts[idx])
-                  assignFusionImage({
-                    shotIndex: activeFusionShot,
-                    slot: "product_images",
-                    productIndex: nextEmpty >= 0 ? nextEmpty : 0,
-                  }, ref)
-                }}
-              />
-            </>
-          )}
           {activeTab === "review" && (
             <section className="rounded-lg border border-white/10 bg-white/[0.035] p-2.5 text-[10.5px] leading-relaxed text-white/58">
               <div className="mb-2 text-[12px] font-semibold text-white">素材审核</div>

web/components/nodes/index.tsx

@@ -1955,6 +1955,7 @@ export function KeyframePanelNode({ data }: any) {
           embedded
           jobId={d.job.id}
           frames={d.job.frames}
+          generatedVideos={d.job.generated_videos ?? []}
           activeIndex={d.expandedFrame}
           selected={d.selectedFrames}
           onClose={d.onCloseExpandedFrame}
@@ -1964,6 +1965,7 @@ export function KeyframePanelNode({ data }: any) {
           clipboard={d.clipboard}
           onCopyImage={d.onCopyImage}
           onGenerateProductFusionVideo={d.onGenerateProductFusionVideo}
+          onDeleteVideo={d.onDeleteVideo}
         />
       </div>
       <button