diff --git a/.memory/worklog.json b/.memory/worklog.json
index cdb0efb..c128a88 100644
--- a/.memory/worklog.json
+++ b/.memory/worklog.json
@@ -2421,6 +2421,19 @@
       "type": "session-heartbeat",
       "message": "Codex 会话活跃 · 最近命令：codex · 3 项未提交变更 · 最近提交：auto-save 2026-05-13 21:18 (~2)",
       "files_changed": 3
+    },
+    {
+      "ts": "2026-05-13T21:24:32+08:00",
+      "type": "commit",
+      "message": "auto-save 2026-05-13 21:24 (~6)",
+      "hash": "2befdf4",
+      "files_changed": 6
+    },
+    {
+      "ts": "2026-05-13T13:29:31Z",
+      "type": "session-heartbeat",
+      "message": "Codex 会话活跃 · 最近命令：codex · 7 项未提交变更 · 最近提交：auto-save 2026-05-13 21:24 (~6)",
+      "files_changed": 7
     }
   ]
 }
diff --git a/api/main.py b/api/main.py
index 7a6ebd5..a763853 100644
--- a/api/main.py
+++ b/api/main.py
@@ -125,6 +125,8 @@ class StoryboardScene(BaseModel):
     v2: 4 图槽 + 时长（复制粘贴模式）— 主体 / 场景 / 产品 / 动作 各一张图
     v1 字段保留兼容（subject/product/scene/action/reference_ids）"""
     duration: float = 0
+    first_image: dict | None = None
+    last_image: dict | None = None
     # 4 图槽：dict 含 {kind, frame_idx, element_id?, cutout_id?, label}
     subject_image: dict | None = None
     scene_image: dict | None = None
@@ -1647,6 +1649,8 @@ class UpdateStoryboardReq(BaseModel):
 class GenerateStoryboardVideoReq(BaseModel):
     prompt: str
     duration: float = 4
+    first_image: dict | None = None
+    last_image: dict | None = None
     subject_image: dict | None = None
     scene_image: dict | None = None
     product_image: dict | None = None
@@ -1758,7 +1762,15 @@ def ark_reference_data_url(ref_img: Path) -> str:
     return f"data:{mime};base64,{base64.b64encode(ref_img.read_bytes()).decode('ascii')}"
 
 
-def submit_video_create(client, url: str, headers: dict, ref_img: Path, payload: dict, source_ref: VideoSourceRef | None = None):
+def submit_video_create(
+    client,
+    url: str,
+    headers: dict,
+    ref_img: Path,
+    payload: dict,
+    source_ref: VideoSourceRef | None = None,
+    last_img: Path | None = None,
+):
     if video_uses_ark():
         content = [{"type": "text", "text": payload["prompt"]}]
         if source_ref and source_ref.kind == "source_video" and source_ref.url:
@@ -1776,6 +1788,14 @@ def submit_video_create(client, url: str, headers: dict, ref_img: Path, payload:
                 "role": "first_frame",
             }
         )
+        if last_img and last_img.exists():
+            content.append(
+                {
+                    "type": "image_url",
+                    "image_url": {"url": ark_reference_data_url(last_img)},
+                    "role": "last_frame",
+                }
+            )
         data = {
             "model": payload["model"],
             "content": content,
@@ -1801,17 +1821,33 @@ def submit_video_create(client, url: str, headers: dict, ref_img: Path, payload:
         )
 
 
-def render_storyboard_video(job_id: str, local_id: str, provider_id: str, ref_path: Path, prompt: str, model: str, seconds: str, size: str, source_ref: VideoSourceRef | None = None) -> None:
+def render_storyboard_video(
+    job_id: str,
+    local_id: str,
+    provider_id: str,
+    ref_path: Path,
+    prompt: str,
+    model: str,
+    seconds: str,
+    size: str,
+    source_ref: VideoSourceRef | None = None,
+    last_ref_path: Path | None = None,
+) -> None:
     import httpx
 
     out_dir = job_dir(job_id) / "storyboard_videos" / local_id
     ref_img = out_dir / "reference.jpg"
+    last_img = out_dir / "last_reference.jpg"
     out_mp4 = out_dir / "video.mp4"
     base = video_api_base()
     headers = {"Authorization": f"Bearer {video_api_key()}"}
 
     try:
         prepare_video_reference(ref_path, ref_img)
+        prepared_last_img: Path | None = None
+        if last_ref_path and last_ref_path.exists():
+            prepare_video_reference(last_ref_path, last_img)
+            prepared_last_img = last_img
         update_generated_video(job_id, local_id, status="in_progress", progress=5)
         with httpx.Client(timeout=120) as client:
             payload = {"model": model, "prompt": prompt, "size": size}
@@ -1819,10 +1855,13 @@ def render_storyboard_video(job_id: str, local_id: str, provider_id: str, ref_pa
             create = None
             create_errors: list[str] = []
             for create_path in VIDEO_CREATE_PATHS:
-                resp = submit_video_create(client, f"{base}{video_path(create_path)}", headers, ref_img, payload, source_ref)
+                resp = submit_video_create(client, f"{base}{video_path(create_path)}", headers, ref_img, payload, source_ref, prepared_last_img)
                 if video_uses_ark() and source_ref and resp.status_code in {400, 422}:
                     create_errors.append(f"{video_path(create_path)} + reference_video -> HTTP {resp.status_code}: {resp.text[:160]}")
-                    resp = submit_video_create(client, f"{base}{video_path(create_path)}", headers, ref_img, payload, None)
+                    resp = submit_video_create(client, f"{base}{video_path(create_path)}", headers, ref_img, payload, None, prepared_last_img)
+                if video_uses_ark() and prepared_last_img and resp.status_code in {400, 422}:
+                    create_errors.append(f"{video_path(create_path)} + last_frame -> HTTP {resp.status_code}: {resp.text[:160]}")
+                    resp = submit_video_create(client, f"{base}{video_path(create_path)}", headers, ref_img, payload, None, None)
                 if resp.status_code < 400:
                     create = resp
                     break
@@ -1879,11 +1918,12 @@ def generate_storyboard_video(job_id: str, idx: int, req: GenerateStoryboardVide
     if not prompt:
         raise HTTPException(400, "prompt required")
 
-    ref = req.product_image or req.subject_image or req.scene_image or req.action_image
+    ref = req.first_image or req.subject_image or req.product_image or req.scene_image or req.action_image
     ref_path = storyboard_ref_path(job_id, ref) or (job_dir(job_id) / "frames" / f"{idx:03d}.jpg")
     if not ref_path.exists():
         raise HTTPException(404, "reference image missing")
     poster = storyboard_ref_url(job_id, ref) or f"/jobs/{job_id}/frames/{idx}.jpg"
+    last_ref_path = storyboard_ref_path(job_id, req.last_image)
 
     local_id = uuid.uuid4().hex[:12]
     model = resolve_video_model(req.model)
@@ -1905,7 +1945,7 @@ def generate_storyboard_video(job_id: str, idx: int, req: GenerateStoryboardVide
     source_ref = req.source_ref
     if source_ref and source_ref.kind == "source_video" and not source_ref.url:
         source_ref = None
-    bg.add_task(render_storyboard_video, job_id, local_id, "", ref_path, prompt, model, seconds, req.size, source_ref)
+    bg.add_task(render_storyboard_video, job_id, local_id, "", ref_path, prompt, model, seconds, req.size, source_ref, last_ref_path)
     return job
 
 
diff --git a/docs/source-analysis.html b/docs/source-analysis.html
index 8c27291..a7527ad 100644
--- a/docs/source-analysis.html
+++ b/docs/source-analysis.html
@@ -830,6 +830,18 @@ api/main.py
         <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>2026-05-13 · 分镜编排改为首尾帧生成</h3>
+              <span class="tag violet">StoryboardWorkbench</span>
+              <span class="tag blue">API</span>
+            </header>
+            <div class="body">
+              <p><strong>问题：</strong>赶交付时，顶部横向分镜缩略条占空间，4 图槽也不如“首帧到尾帧”直接；用户希望直接做首尾帧视频生成。</p>
+              <p><strong>改动：</strong>移除 <code>StoryboardBar</code> 的横向分镜缩略图区域，只保留标题栏和展开按钮；<code>StoryboardWorkbench</code> 改成首帧 / 尾帧两个槽，首帧默认当前分镜，尾帧默认下一张已选分镜，也可从剪贴板粘贴指定结束画面。后端 <code>/storyboard/video</code> 支持 <code>first_image</code>/<code>last_image</code>，Ark 请求同时传 <code>first_frame</code>/<code>last_frame</code>，如果接口不接受尾帧字段则自动回退到单首帧。</p>
+              <p><strong>影响：</strong><code>web/components/storyboard-bar.tsx</code>、<code>web/components/storyboard-workbench.tsx</code>、<code>web/app/page.tsx</code>、<code>web/lib/api.ts</code>、<code>api/main.py</code>。</p>
+            </div>
+          </article>
           <article class="change">
             <header>
               <h3>2026-05-13 · 生视频携带原视频链接做节奏参考</h3>
diff --git a/web/app/page.tsx b/web/app/page.tsx
index 93d0bfd..7484ac8 100644
--- a/web/app/page.tsx
+++ b/web/app/page.tsx
@@ -232,6 +232,20 @@ export default function Home() {
     if (!frame) return
 
     const labelOf = (ref?: ImageRef | null, fallback = "未提供") => ref?.label || fallback
+    const keyframeRef: ImageRef = {
+      kind: "keyframe",
+      frame_idx: frameIdx,
+      label: `分镜 ${frameIdx + 1} 首帧`,
+    }
+    const orderedSelected = job.frames
+      .filter((f) => selectedFrames.has(f.index))
+      .sort((a, b) => a.timestamp - b.timestamp)
+    const nextFrame = orderedSelected.find((f) => f.timestamp > frame.timestamp) ?? null
+    const defaultLastRef: ImageRef | null = nextFrame
+      ? { kind: "keyframe", frame_idx: nextFrame.index, label: `分镜 ${nextFrame.index + 1} 尾帧` }
+      : null
+    const firstRef = scene.first_image ?? keyframeRef
+    const lastRef = scene.last_image ?? defaultLastRef
     const duration = scene.duration && scene.duration > 0 ? scene.duration : 5
     const sourceScene = frame.description?.scene ? `参考画面识别：${frame.description.scene}` : ""
     const sourceStyle = frame.description?.style ? `参考风格：${frame.description.style}` : ""
@@ -245,21 +259,21 @@ export default function Home() {
     const sceneDirection = scene.scene?.trim()
       || "借鉴参考画面的构图、可信感和空间层次，但改造成适合 SKG 产品广告的现代家居、办公或零售场景。"
     const actionDirection = scene.action?.trim()
-      || "一镜到底缓慢推进，先建立画面，再出现自然手部互动，最后停在产品细节或使用状态特写。"
+      || "按首帧到尾帧做平滑过渡，动作连续自然，镜头运动稳定，最后准确停在尾帧意图。"
     const prompt = [
       `竖屏 9:16，${duration.toFixed(1)} 秒，SKG 产品短视频广告。`,
-      "直接根据当前分镜关键帧生成视频。必须使用输入的完整视频关键帧作为第一帧和视觉锚点：第一帧构图、主体位置、透视关系和光线方向保持稳定，然后从这一帧自然动起来。",
-      "生成一段单镜头连续视频，一镜到底，不要跳切，不要突然换场景，不要突然换主体，不要蒙太奇，不要多镜头拼接。",
+      "使用首帧和尾帧生成连续过渡视频：首帧必须严格作为视频开始画面，尾帧必须作为视频结束目标画面，中间只做自然运动补间。",
+      "生成一段单镜头连续视频，一镜到底，从首帧平滑过渡到尾帧；不要跳切，不要突然换场景，不要突然换主体，不要蒙太奇，不要多镜头拼接。",
       "如果提供了原视频链接，把它只作为节奏、镜头运动、动作顺序和画面调度参考；不要照搬原视频里的品牌、文字、水印、竞品产品或具体人物。",
-      "时间线：0%-25% 保持首帧构图并轻微启动；25%-70% 做一个清晰、缓慢、可信的产品展示动作；70%-100% 镜头自然停稳在 SKG 产品或使用效果特写。",
+      "时间线：0%-15% 锁住首帧构图并轻微启动；15%-85% 做平滑连续运动；85%-100% 缓慢贴近尾帧并稳定收住。",
       `主体改造：${subjectDirection}`,
       `产品替换：${productDirection}`,
       `场景改造：${sceneDirection}`,
       `连续动作和镜头：${actionDirection}`,
-      `参考主体图槽：${labelOf(scene.subject_image, "产品演示主体或手部姿态")}`,
-      `参考场景图槽：${labelOf(scene.scene_image, "现代健康生活场景")}`,
-      `SKG 产品图槽：${labelOf(scene.product_image, "SKG 产品视觉主角")}`,
-      `参考动作图槽：${labelOf(scene.action_image, "自然拿取、佩戴、展示或靠近产品的动作")}`,
+      `首帧：${labelOf(firstRef, "当前分镜关键帧")}`,
+      `尾帧：${labelOf(lastRef, "未指定，按首帧小幅自然运动收尾")}`,
+      `SKG 产品参考：${labelOf(scene.product_image, "SKG 产品视觉主角")}`,
+      `动作参考：${labelOf(scene.action_image, "自然拿取、佩戴、展示或靠近产品的动作")}`,
       sourceScene,
       sourceStyle,
       sourceObjects,
@@ -270,16 +284,13 @@ export default function Home() {
 
     try {
       toast.info(`已提交 ${model} 生视频 · 分镜 ${frameIdx + 1}`)
-      const keyframeRef: ImageRef = {
-        kind: "keyframe",
-        frame_idx: frameIdx,
-        label: `分镜 ${frameIdx + 1} 关键帧`,
-      }
       const sourceUrl = job.url?.trim()
       const updated = await generateStoryboardVideo(job.id, frameIdx, {
         prompt,
         duration,
-        subject_image: keyframeRef,
+        first_image: firstRef,
+        last_image: lastRef,
+        subject_image: firstRef,
         scene_image: null,
         product_image: null,
         action_image: null,
@@ -293,7 +304,7 @@ export default function Home() {
     } catch (e) {
       toast.error("提交视频失败：" + (e instanceof Error ? e.message : String(e)))
     }
-  }, [job, setJob])
+  }, [job, selectedFrames, setJob])
 
   // URL ?job=xxx,yyy 自动恢复多个 job
   useEffect(() => {
diff --git a/web/components/storyboard-bar.tsx b/web/components/storyboard-bar.tsx
index 661eeb4..d49539f 100644
--- a/web/components/storyboard-bar.tsx
+++ b/web/components/storyboard-bar.tsx
@@ -1,8 +1,6 @@
 "use client"
-import { useEffect, useRef, useState } from "react"
-import { createPortal } from "react-dom"
-import { LayoutGrid, ChevronDown, ChevronUp, Sparkle } from "lucide-react"
-import { type Job, effectiveFrameUrl, hasCutout } from "@/lib/api"
+import { LayoutGrid, ChevronDown, ChevronUp } from "lucide-react"
+import { type Job, hasCutout } from "@/lib/api"
 
 interface Props {
   job: Job | null
@@ -15,19 +13,12 @@ interface Props {
 }
 
 export function StoryboardBar({ job, selectedFrames, focusedFrame, onFocusFrame, workbenchOpen = false, onOpenWorkbench, onCloseWorkbench }: Props) {
-  const [collapsed, setCollapsed] = useState(false)
-  const [mounted, setMounted] = useState(false)
-  useEffect(() => setMounted(true), [])
-  const [hover, setHover] = useState<{ src: string; topLabel: string; subLabel: string; rect: DOMRect } | null>(null)
-  const btnRefs = useRef<Record<number, HTMLButtonElement | null>>({})
-
   if (!job) return null
 
   const frames = job.frames
     .filter((f) => selectedFrames.has(f.index))
     .sort((a, b) => a.timestamp - b.timestamp)
 
-  const aspect = job.height > 0 ? `${job.width}/${job.height}` : "9/16"
   const totalElements = frames.reduce(
     (sum, f) => sum + (f.elements?.filter((e) => hasCutout(e)).length ?? 0),
     0,
@@ -73,7 +64,6 @@ export function StoryboardBar({ job, selectedFrames, focusedFrame, onFocusFrame,
               if (frames.length === 0) return
               const nextFrame = focusedFrame ?? frames[0].index
               if (focusedFrame === null) onFocusFrame(nextFrame)
-              setCollapsed(false)
               onOpenWorkbench?.(nextFrame)
             }}
             disabled={frames.length === 0}
@@ -83,123 +73,8 @@ export function StoryboardBar({ job, selectedFrames, focusedFrame, onFocusFrame,
             {workbenchOpen ? <ChevronUp className="h-3 w-3" /> : <ChevronDown className="h-3 w-3" />}
             {workbenchOpen ? "收起编排" : "展开编排"}
           </button>
-          <button
-            onClick={() => {
-              const nextCollapsed = !collapsed
-              setCollapsed(nextCollapsed)
-              if (nextCollapsed) onCloseWorkbench?.()
-            }}
-            className="text-white/50 hover:text-white text-[11px] inline-flex items-center gap-1"
-            title={collapsed ? "展开" : "折叠"}
-          >
-            {collapsed ? <ChevronDown className="h-3 w-3" /> : <ChevronUp className="h-3 w-3" />}
-            {collapsed ? "展开" : "折叠"}
-          </button>
         </div>
       </div>
-
-      {/* thumbnails row */}
-      {!collapsed && (
-        frames.length === 0 ? (
-          <div className="px-4 pb-3 text-[11px] text-white/40">
-            还没选用分镜 · 在「关键帧」节点上点击缩略图右下勾选「选用此帧」，被选用的分镜按时间序出现在这里
-          </div>
-        ) : (
-          <div className="px-4 pb-3 flex gap-2 overflow-x-auto">
-            {frames.map((f, i) => {
-              const elementCount = f.elements?.filter((e) => hasCutout(e)).length ?? 0
-              const totalElCount = f.elements?.length ?? 0
-              const cleaned = f.cleaned_applied
-              const isFocused = focusedFrame === f.index
-              return (
-                <button
-                  key={f.index}
-                  ref={(el) => { btnRefs.current[f.index] = el }}
-                  onClick={() => {
-                    onFocusFrame(f.index)
-                    setCollapsed(false)
-                  }}
-                  onMouseEnter={() => {
-                    const el = btnRefs.current[f.index]
-                    if (el) setHover({
-                      src: effectiveFrameUrl(job.id, f),
-                      topLabel: `分镜 ${i + 1}`,
-                      subLabel: `${f.timestamp.toFixed(2)}s`,
-                      rect: el.getBoundingClientRect(),
-                    })
-                  }}
-                  onMouseLeave={() => setHover(null)}
-                  title={`分镜 ${i + 1} · ${f.timestamp.toFixed(2)}s${cleaned ? " · 已清洗" : ""} · ${elementCount}/${totalElCount} 元素 · 点击聚焦`}
-                  className={`relative shrink-0 rounded-md border transition shadow-lg hover:-translate-y-0.5 ${
-                    isFocused
-                      ? "border-violet-300 ring-2 ring-violet-300/70"
-                      : "border-white/15 hover:border-violet-300/60"
-                  }`}
-                  style={{ width: 88, aspectRatio: aspect }}
-                >
-                  <img
-                    src={effectiveFrameUrl(job.id, f)}
-                    alt={`frame ${f.index}`}
-                    className="absolute inset-0 w-full h-full object-cover rounded-md"
-                  />
-                  <div className="absolute top-1 left-1 text-[9.5px] font-bold text-white bg-violet-500/85 backdrop-blur px-1.5 py-0.5 rounded">
-                    #{i + 1}
-                  </div>
-                  {cleaned && (
-                    <div className="absolute top-1 right-1 text-[9px] text-white bg-cyan-500/85 backdrop-blur px-1 py-0.5 rounded font-bold" title="已清洗">
-                      ✨
-                    </div>
-                  )}
-                  <div className="absolute bottom-0 right-0 left-0 px-1.5 py-0.5 text-[9px] font-mono text-white bg-gradient-to-t from-black/85 to-transparent flex items-center justify-between rounded-b-md">
-                    <span>{f.timestamp.toFixed(1)}s</span>
-                    {totalElCount > 0 && (
-                      <span className="inline-flex items-center gap-0.5">
-                        <Sparkle className="h-2 w-2" />
-                        {elementCount}/{totalElCount}
-                      </span>
-                    )}
-                  </div>
-                </button>
-              )
-            })}
-          </div>
-        )
-      )}
-
-
-      {/* Hover 预览 · 浮在缩略图正下方（bar 在顶部 fixed，下方是 DAG 画布区） */}
-      {mounted && hover && (() => {
-        const vidAspect = job.height > 0 ? job.height / job.width : 16 / 9
-        const w = 280
-        const h = w * vidAspect
-        const gap = 10
-        const centerX = hover.rect.left + hover.rect.width / 2
-        const left = Math.max(12, Math.min(window.innerWidth - w - 12, centerX - w / 2))
-        const top = hover.rect.bottom + gap
-        return createPortal(
-          <div
-            className="fixed z-[120] pointer-events-none"
-            style={{
-              left, top,
-              animation: "drawer-in 0.18s cubic-bezier(0.32, 0.72, 0, 1)",
-            }}
-          >
-            <div className="rounded-lg overflow-hidden border-2 border-violet-300/50 bg-white shadow-2xl">
-              <img
-                src={hover.src}
-                alt="preview"
-                className="block"
-                style={{ width: w, height: h, objectFit: "cover" }}
-              />
-              <div className="px-2 py-1 bg-black/80 text-white text-[10.5px] flex items-center justify-between gap-2">
-                <span className="truncate">{hover.topLabel}</span>
-                <span className="text-white/60 font-mono shrink-0">{hover.subLabel}</span>
-              </div>
-            </div>
-          </div>,
-          document.body,
-        )
-      })()}
     </div>
   )
 }
diff --git a/web/components/storyboard-workbench.tsx b/web/components/storyboard-workbench.tsx
index 1e85efd..c672cc2 100644
--- a/web/components/storyboard-workbench.tsx
+++ b/web/components/storyboard-workbench.tsx
@@ -82,6 +82,13 @@ export function StoryboardWorkbench({ job, selectedFrames, open, onClose, onJobU
     .sort((a, b) => a.timestamp - b.timestamp)
   const focusFrame = focusedIdx !== null ? job.frames.find((f) => f.index === focusedIdx) ?? null : null
   const focusSeq = focusFrame ? frames.findIndex((f) => f.index === focusFrame.index) + 1 : 0
+  const defaultFirstRef: ImageRef | null = focusFrame
+    ? { kind: "keyframe", frame_idx: focusFrame.index, label: `分镜 ${focusSeq || focusFrame.index + 1} 首帧` }
+    : null
+  const nextFrame = focusFrame ? frames.find((f) => f.timestamp > focusFrame.timestamp) ?? null : null
+  const defaultLastRef: ImageRef | null = nextFrame
+    ? { kind: "keyframe", frame_idx: nextFrame.index, label: `分镜 ${frames.findIndex((f) => f.index === nextFrame.index) + 1} 尾帧` }
+    : null
 
   const queueSave = (next: StoryboardScene) => {
     setForm(next)
@@ -176,21 +183,20 @@ export function StoryboardWorkbench({ job, selectedFrames, open, onClose, onJobU
                 </div>
               </div>
 
-              {/* 4 图槽 grid：图片是参考，不是最终复刻素材 */}
-              <div className="grid grid-cols-4 gap-4">
+              {/* 首尾帧：图片直接参与视频生成 */}
+              <div className="grid grid-cols-2 gap-4">
                 {([
-                  { key: "subject_image" as const, label: "参考主体", placeholder: "人物 / 手部 / 模特姿态" },
-                  { key: "scene_image" as const, label: "参考场景", placeholder: "药店柜台 / 卧室 / 浴室" },
-                  { key: "product_image" as const, label: "SKG 产品", placeholder: "产品图 / 包装 / 使用状态" },
-                  { key: "action_image" as const, label: "参考动作", placeholder: "拿起 / 佩戴 / 展示 / 递给顾客" },
+                  { key: "first_image" as const, label: "首帧", placeholder: "默认当前分镜关键帧" },
+                  { key: "last_image" as const, label: "尾帧", placeholder: defaultLastRef ? "默认下一张已选分镜" : "粘贴一张结束画面" },
                 ]).map(({ key, label, placeholder }) => {
-                  const ref = form[key]
+                  const fallback = key === "first_image" ? defaultFirstRef : defaultLastRef
+                  const ref = form[key] ?? fallback
                   const url = ref ? resolveImageRefUrl(job.id, ref) : ""
                   return (
                     <div key={key} className="rounded-lg bg-white/[0.04] border border-white/10 p-2.5">
                       <div className="text-[12px] text-white font-semibold mb-2 flex items-center justify-between">
                         <span>{label}</span>
-                        {ref && (
+                        {form[key] && (
                           <button
                             onClick={() => queueSave({ ...form, [key]: null })}
                             className="text-[10px] text-white/40 hover:text-rose-300"
@@ -238,6 +244,9 @@ export function StoryboardWorkbench({ job, selectedFrames, open, onClose, onJobU
                   )
                 })}
               </div>
+              <div className="rounded-md border border-violet-300/20 bg-violet-500/10 px-3 py-2 text-[11px] leading-relaxed text-violet-100/75">
+                现在按首尾帧出片：首帧默认当前分镜，尾帧默认下一张已选分镜。需要指定结尾时，在任意关键帧或生成图点 📋，再粘贴到「尾帧」。
+              </div>
 
               {/* 改造 brief：明确“借鉴参考 → 变成 SKG 产品视频”，避免直接复刻 */}
               <section className="rounded-lg bg-white/[0.035] border border-white/10 p-3">
@@ -314,13 +323,13 @@ export function StoryboardWorkbench({ job, selectedFrames, open, onClose, onJobU
                     }
                   }}
                   className="w-full py-3 rounded-lg text-[13.5px] font-semibold inline-flex items-center justify-center gap-2 bg-gradient-to-r from-rose-500 to-violet-500 text-white border border-violet-300/40 shadow-lg shadow-violet-500/20 hover:from-rose-400 hover:to-violet-400 disabled:opacity-40 disabled:cursor-not-allowed"
-                  title={`用当前分镜关键帧作为首帧，调用 ${currentModelLabel} 生视频 API`}
+                  title={`用首帧和尾帧调用 ${currentModelLabel} 生视频 API`}
                 >
                   {generating ? <Loader2 className="h-4 w-4 animate-spin" /> : <Wand2 className="h-4 w-4" />}
-                  用当前关键帧生成视频
+                  用首尾帧生成视频
                 </button>
                 <div className="mt-2 text-[10.5px] text-white/35 leading-relaxed">
-                  直接用当前分镜关键帧作为首帧快速出片；4 图槽和改造目标只作为提示词参考，生成中的进度和完成后的 MP4 会显示在 Video Gen 节点。
+                  直接用首帧 + 尾帧快速生成连续过渡视频；改造目标和原视频链接只作为节奏 / 镜头参考，生成进度和 MP4 会显示在 Video Gen 节点。
                 </div>
               </section>
           </div>
diff --git a/web/lib/api.ts b/web/lib/api.ts
index d0a94f1..ff55e2e 100644
--- a/web/lib/api.ts
+++ b/web/lib/api.ts
@@ -57,6 +57,8 @@ export interface ImageRef {
 
 export interface StoryboardScene {
   duration: number
+  first_image?: ImageRef | null
+  last_image?: ImageRef | null
   subject_image?: ImageRef | null
   scene_image?: ImageRef | null
   product_image?: ImageRef | null
@@ -378,6 +380,8 @@ export async function generateStoryboardVideo(
   body: {
     prompt: string
     duration?: number
+    first_image?: ImageRef | null
+    last_image?: ImageRef | null
     subject_image?: ImageRef | null
     scene_image?: ImageRef | null
     product_image?: ImageRef | null