Gemini-VideoGeneration

Sleeping

App Files Files Community

LehongWu commited on 29 days ago

Commit

38b7ac0

verified ·

1 Parent(s): 7d2ad47

Upload folder using huggingface_hub

Browse files

Files changed (10) hide show

docs/SPEC_WEB_UI.md +2 -2
docs/WEB_DEV_GUIDE.md +1 -1
web/backend/routes/generate.py +5 -0
web/backend/services/generation_options.py +12 -0
web/backend/static/assets/index-J-30f0lA.js +0 -0
web/backend/static/index.html +1 -1
web/config/generation_options.json +4 -4
web/frontend/src/context/GenerationOptionsContext.tsx +8 -1
web/frontend/src/pages/VideoFrames.tsx +49 -23
web/frontend/src/pages/VideoGen.tsx +13 -7

docs/SPEC_WEB_UI.md CHANGED Viewed

@@ -10,9 +10,9 @@ Image and video generation from prompts and optional reference images.
 **A. 图片** — 0–3 张参考图 + 提示词 → 一张图。模型含 Nano Banana（`gemini-2.5-flash-image`，无长思考）、Nano Banana 2（`gemini-3.1-flash-image-preview`，可选长思考）、Nano Banana Pro（`gemini-3-pro-image-preview`，长思考）；默认选项为 Nano Banana 2 快速。宽高比、分辨率见 `generation_options.json`。
-**B. 视频** — 0–3 张参考图 + 提示词 → 短视频。Veo 模型可配置；**Veo 3.1 Lite** 在配置中设 **`supports_reference_images: false`**。有参考图时时长 **8s**；仅文案时 4/6/8s（以 API 为准）。
-**C. 首尾过渡** — 起始帧必填 + 提示词；结尾帧可选或「与起始相同」。时长 **固定 8s**。
 ## 辅助工具

 **A. 图片** — 0–3 张参考图 + 提示词 → 一张图。模型含 Nano Banana（`gemini-2.5-flash-image`，无长思考）、Nano Banana 2（`gemini-3.1-flash-image-preview`，可选长思考）、Nano Banana Pro（`gemini-3-pro-image-preview`，长思考）；默认选项为 Nano Banana 2 快速。宽高比、分辨率见 `generation_options.json`。
+**B. 视频** — 0–3 张参考图 + 提示词 → 短视频。Veo 模型可配置；**Veo 3 / Veo 3 Fast** 与 **Veo 3.1 Lite** 为 **`supports_reference_images: false`**（前两者仅纯提示词）。默认模型 **Veo 3.1 Fast**。有参考图时时长 **8s**；仅文案时 4/6/8s（以 API 为准）。
+**C. 首尾过渡** — 起始帧必填 + 提示词；时长 **固定 8s**。**Veo 3 / Veo 3 Fast**（`supports_end_frame: false`）仅起始帧；**Veo 3.1** 可选独立结尾帧或「与起始相同」。默认模型 **Veo 3.1 Fast**。
 ## 辅助工具

docs/WEB_DEV_GUIDE.md CHANGED Viewed

@@ -40,7 +40,7 @@ export SESSION_SECRET="$(openssl rand -hex 32)"
 Edit **`web/config/generation_options.json`** (or path from `GENERATION_OPTIONS_PATH`). UI loads it via **`GET /api/config/generation-options`** after login. JSON-only changes need no frontend rebuild.
 - **`image`**: `models`, `aspect_ratios`, `resolutions`.
-- **`video`** / **`video_frames`**: `models`, `aspect_ratios`, `resolutions`, `durations_seconds`. On **`video`**, **`supports_reference_images: false`** hides reference images (e.g. Veo 3.1 Lite). With reference images, duration is **8s**. **首尾帧** route is always **8s**.
 ## Build frontend

 Edit **`web/config/generation_options.json`** (or path from `GENERATION_OPTIONS_PATH`). UI loads it via **`GET /api/config/generation-options`** after login. JSON-only changes need no frontend rebuild.
 - **`image`**: `models`, `aspect_ratios`, `resolutions`.
+- **`video`** / **`video_frames`**: `models`, `aspect_ratios`, `resolutions`, `durations_seconds`. On **`video`**, **`supports_reference_images: false`** disables reference images for that model (e.g. Veo 3 / Veo 3 Fast — prompt-only; Veo 3.1 Lite — also no reference images). With reference images, duration is **8s**. **首尾帧** route is always **8s**. On **`video_frames`**, **`supports_end_frame: false`** (e.g. Veo 3 / Veo 3 Fast) means only the start frame is used; end-frame UI and last-frame conditioning are omitted.
 ## Build frontend

web/backend/routes/generate.py CHANGED Viewed

@@ -16,6 +16,7 @@ from web.backend.deps import SessionUser
 from web.backend.services import gemini_image, gemini_video
 from web.backend.services.generation_options import (
     video_model_supports_4k,
     video_model_supports_reference_images,
 )
@@ -217,6 +218,10 @@ async def generate_video_frames(
     same_as_start = _parse_form_bool(end_same_as_start)
     if same_as_start:
         end_raw = None
     # First/last frame conditioning: API requires 8s duration (same family of rules as reference images).
     dur = 8

 from web.backend.services import gemini_image, gemini_video
 from web.backend.services.generation_options import (
     video_model_supports_4k,
+    video_model_supports_end_frame,
     video_model_supports_reference_images,
 )
     same_as_start = _parse_form_bool(end_same_as_start)
     if same_as_start:
         end_raw = None
+    if not video_model_supports_end_frame(m):
+        # Veo 3 / Veo 3 Fast: start frame only (no last-frame conditioning).
+        end_raw = None
+        same_as_start = False
     # First/last frame conditioning: API requires 8s duration (same family of rules as reference images).
     dur = 8

web/backend/services/generation_options.py CHANGED Viewed

@@ -53,3 +53,15 @@ def video_model_supports_4k(model: str) -> bool:
             if isinstance(m, dict) and m.get("value") == model:
                 return bool(m.get("supports_4k", True))
     return True

             if isinstance(m, dict) and m.get("value") == model:
                 return bool(m.get("supports_4k", True))
     return True
+def video_model_supports_end_frame(model: str) -> bool:
+    """Read `video_frames.models[].supports_end_frame` (default True if key absent)."""
+    try:
+        data = load_generation_options()
+    except (OSError, ValueError, json.JSONDecodeError):
+        return True
+    for m in data.get("video_frames", {}).get("models", []):
+        if isinstance(m, dict) and m.get("value") == model:
+            return bool(m.get("supports_end_frame", True))
+    return True

web/backend/static/assets/index-J-30f0lA.js ADDED Viewed

The diff for this file is too large to render. See raw diff

web/backend/static/index.html CHANGED Viewed

@@ -53,7 +53,7 @@
         }
       })();
     </script>
-    <script type="module" crossorigin src="/assets/index-l1IkTo6p.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-JnUJDL9j.css">
   </head>
   <body>

         }
       })();
     </script>
+    <script type="module" crossorigin src="/assets/index-J-30f0lA.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-JnUJDL9j.css">
   </head>
   <body>

web/config/generation_options.json CHANGED Viewed

@@ -21,8 +21,8 @@
   },
   "video": {
     "models": [
-      { "value": "veo-3.0-fast-generate-001", "label": "Veo 3 Fast（快速）", "supports_reference_images": true },
-      { "value": "veo-3.0-generate-001", "label": "Veo 3（标准）", "supports_reference_images": true },
       { "value": "veo-3.1-lite-generate-preview", "label": "Veo 3.1 Lite（轻量）", "supports_reference_images": false, "supports_4k": false },
       { "value": "veo-3.1-fast-generate-preview", "label": "Veo 3.1 Fast（快速）", "supports_reference_images": true },
       { "value": "veo-3.1-generate-preview", "label": "Veo 3.1（标准）", "supports_reference_images": true }
@@ -33,8 +33,8 @@
   },
   "video_frames": {
     "models": [
-      { "value": "veo-3.0-fast-generate-001", "label": "Veo 3 Fast（快速）" },
-      { "value": "veo-3.0-generate-001", "label": "Veo 3（标准）" },
       { "value": "veo-3.1-lite-generate-preview", "label": "Veo 3.1 Lite（轻量）", "supports_4k": false },
       { "value": "veo-3.1-fast-generate-preview", "label": "Veo 3.1 Fast（快速）" },
       { "value": "veo-3.1-generate-preview", "label": "Veo 3.1（标准）" }

   },
   "video": {
     "models": [
+      { "value": "veo-3.0-fast-generate-001", "label": "Veo 3 Fast（快速）", "supports_reference_images": false },
+      { "value": "veo-3.0-generate-001", "label": "Veo 3（标准）", "supports_reference_images": false },
       { "value": "veo-3.1-lite-generate-preview", "label": "Veo 3.1 Lite（轻量）", "supports_reference_images": false, "supports_4k": false },
       { "value": "veo-3.1-fast-generate-preview", "label": "Veo 3.1 Fast（快速）", "supports_reference_images": true },
       { "value": "veo-3.1-generate-preview", "label": "Veo 3.1（标准）", "supports_reference_images": true }
   },
   "video_frames": {
     "models": [
+      { "value": "veo-3.0-fast-generate-001", "label": "Veo 3 Fast（快速）", "supports_end_frame": false },
+      { "value": "veo-3.0-generate-001", "label": "Veo 3（标准）", "supports_end_frame": false },
       { "value": "veo-3.1-lite-generate-preview", "label": "Veo 3.1 Lite（轻量）", "supports_4k": false },
       { "value": "veo-3.1-fast-generate-preview", "label": "Veo 3.1 Fast（快速）" },
       { "value": "veo-3.1-generate-preview", "label": "Veo 3.1（标准）" }

web/frontend/src/context/GenerationOptionsContext.tsx CHANGED Viewed

@@ -11,12 +11,14 @@ import { apiFetch } from "../api";
 export type LabeledModel = { value: string; label: string };
 export type ThinkingLevelOption = { value: string; label: string };
-/** Video model entry; `supports_reference_images` / `supports_4k` default to true if omitted. */
 export type VideoModelOption = {
   value: string;
   label: string;
   supports_reference_images?: boolean;
   supports_4k?: boolean;
 };
 export type GenerationOptions = {
@@ -135,6 +137,11 @@ export function modelSupports4k(models: VideoModelOption[], value: string): bool
   return m?.supports_4k !== false;
 }
 export function useGenerationOptions(): GenerationOptions {
   const ctx = useContext(GenerationOptionsContext);
   if (!ctx) {

 export type LabeledModel = { value: string; label: string };
 export type ThinkingLevelOption = { value: string; label: string };
+/** Video model entry; `supports_reference_images` / `supports_4k` / `supports_end_frame` default to true if omitted. */
 export type VideoModelOption = {
   value: string;
   label: string;
   supports_reference_images?: boolean;
   supports_4k?: boolean;
+  /** Start/end mode: if false, only start frame (e.g. Veo 3 family). */
+  supports_end_frame?: boolean;
 };
 export type GenerationOptions = {
   return m?.supports_4k !== false;
 }
+export function modelSupportsEndFrame(models: VideoModelOption[], value: string): boolean {
+  const m = models.find((x) => x.value === value);
+  return m?.supports_end_frame !== false;
+}
 export function useGenerationOptions(): GenerationOptions {
   const ctx = useContext(GenerationOptionsContext);
   if (!ctx) {

web/frontend/src/pages/VideoFrames.tsx CHANGED Viewed

@@ -4,21 +4,24 @@ import { DownloadMediaButton } from "../components/DownloadMediaButton";
 import { ElapsedTimer } from "../components/ElapsedTimer";
 import { ImageSlot } from "../components/ImageSlot";
 import {
-  defaultFastModelValue,
   modelSupports4k,
   useGenerationOptions,
 } from "../context/GenerationOptionsContext";
 export function VideoFrames() {
   const opts = useGenerationOptions();
   const vf = opts.video_frames;
   const [prompt, setPrompt] = useState("");
-  const [model, setModel] = useState(() =>
-    defaultFastModelValue(vf.models, "veo-3.0-fast-generate-001"),
-  );
   const [aspect, setAspect] = useState(vf.aspect_ratios[0] ?? "16:9");
   const [resolution, setResolution] = useState(
-    vf.resolutions[1] ?? vf.resolutions[0] ?? "1080p",
   );
   const [start, setStart] = useState<File | null>(null);
   const [end, setEnd] = useState<File | null>(null);
@@ -29,6 +32,7 @@ export function VideoFrames() {
   const formRef = useRef<HTMLFormElement>(null);
   const supports4k = modelSupports4k(vf.models, model);
   useEffect(() => {
     if (endSameAsStart) {
@@ -36,6 +40,13 @@ export function VideoFrames() {
     }
   }, [endSameAsStart]);
   useEffect(() => {
     if (!supports4k && resolution === "4k") {
       setResolution("1080p");
@@ -61,9 +72,13 @@ export function VideoFrames() {
     fd.append("aspect_ratio", aspect);
     fd.append("duration_seconds", "8");
     fd.append("start_frame", start);
-    fd.append("end_same_as_start", endSameAsStart ? "true" : "false");
-    if (!endSameAsStart && end) {
-      fd.append("end_frame", end);
     }
     setBusy(true);
@@ -91,7 +106,10 @@ export function VideoFrames() {
       <h1 className="font-display text-2xl font-semibold text-ink mb-2">视频生成（首尾过渡）</h1>
       <div className="mb-6 space-y-1.5">
         <p className="text-mist text-sm leading-relaxed">
-          起始帧（必选）；填写提示词（必填）。结尾帧（可选），或与起始帧相同。时长 8 秒。
         </p>
         <p className="text-xs text-mist leading-relaxed">
           Veo 3 与 Veo 3 Fast 为稳定版；Veo 3.1 系列为实验性，接口或效果可能变更。
@@ -175,20 +193,28 @@ export function VideoFrames() {
           <p className="text-sm font-semibold text-ink mb-2">帧</p>
           <div className="space-y-3">
             <ImageSlot label="起始帧" file={start} onChange={setStart} />
-            <label className="flex items-center gap-2 cursor-pointer text-sm text-ink select-none">
-              <input
-                type="checkbox"
-                className="rounded border-slate-300 text-ink focus:ring-clay/50"
-                checked={endSameAsStart}
-                onChange={(e) => setEndSameAsStart(e.target.checked)}
-              />
-              结尾帧与起始帧相同
-            </label>
-            {!endSameAsStart && (
-              <ImageSlot label="结尾帧" file={end} onChange={setEnd} />
-            )}
-            {endSameAsStart && (
-              <p className="text-xs text-mist pl-0.5">结尾帧与起始帧为同一张图。</p>
             )}
           </div>
         </div>

 import { ElapsedTimer } from "../components/ElapsedTimer";
 import { ImageSlot } from "../components/ImageSlot";
 import {
   modelSupports4k,
+  modelSupportsEndFrame,
   useGenerationOptions,
 } from "../context/GenerationOptionsContext";
+const DEFAULT_VIDEO_FRAMES_MODEL = "veo-3.1-fast-generate-preview";
 export function VideoFrames() {
   const opts = useGenerationOptions();
   const vf = opts.video_frames;
   const [prompt, setPrompt] = useState("");
+  const [model, setModel] = useState(() => {
+    const preferred = vf.models.find((m) => m.value === DEFAULT_VIDEO_FRAMES_MODEL);
+    return preferred?.value ?? vf.models[0]?.value ?? DEFAULT_VIDEO_FRAMES_MODEL;
+  });
   const [aspect, setAspect] = useState(vf.aspect_ratios[0] ?? "16:9");
   const [resolution, setResolution] = useState(
+    vf.resolutions[0] ?? "720p",
   );
   const [start, setStart] = useState<File | null>(null);
   const [end, setEnd] = useState<File | null>(null);
   const formRef = useRef<HTMLFormElement>(null);
   const supports4k = modelSupports4k(vf.models, model);
+  const supportsEndFrame = modelSupportsEndFrame(vf.models, model);
   useEffect(() => {
     if (endSameAsStart) {
     }
   }, [endSameAsStart]);
+  useEffect(() => {
+    if (!supportsEndFrame) {
+      setEndSameAsStart(false);
+      setEnd(null);
+    }
+  }, [supportsEndFrame]);
   useEffect(() => {
     if (!supports4k && resolution === "4k") {
       setResolution("1080p");
     fd.append("aspect_ratio", aspect);
     fd.append("duration_seconds", "8");
     fd.append("start_frame", start);
+    if (supportsEndFrame) {
+      fd.append("end_same_as_start", endSameAsStart ? "true" : "false");
+      if (!endSameAsStart && end) {
+        fd.append("end_frame", end);
+      }
+    } else {
+      fd.append("end_same_as_start", "false");
     }
     setBusy(true);
       <h1 className="font-display text-2xl font-semibold text-ink mb-2">视频生成（首尾过渡）</h1>
       <div className="mb-6 space-y-1.5">
         <p className="text-mist text-sm leading-relaxed">
+          起始帧（必选）；填写提示词（必填）。时长 8 秒。
+          <span className="block mt-1">
+            Veo 3 与 Veo 3 Fast 仅支持起始帧，不提供结尾帧选项。Veo 3.1 系列可选用独立结尾帧，或与起始帧相同。
+          </span>
         </p>
         <p className="text-xs text-mist leading-relaxed">
           Veo 3 与 Veo 3 Fast 为稳定版；Veo 3.1 系列为实验性，接口或效果可能变更。
           <p className="text-sm font-semibold text-ink mb-2">帧</p>
           <div className="space-y-3">
             <ImageSlot label="起始帧" file={start} onChange={setStart} />
+            {!supportsEndFrame ? (
+              <p className="text-xs text-mist pl-0.5">
+                当前模型（Veo 3 / Veo 3 Fast）仅使用起始帧，不设置结尾帧。
+              </p>
+            ) : (
+              <>
+                <label className="flex items-center gap-2 cursor-pointer text-sm text-ink select-none">
+                  <input
+                    type="checkbox"
+                    className="rounded border-slate-300 text-ink focus:ring-clay/50"
+                    checked={endSameAsStart}
+                    onChange={(e) => setEndSameAsStart(e.target.checked)}
+                  />
+                  结尾帧与起始帧相同
+                </label>
+                {!endSameAsStart && (
+                  <ImageSlot label="结尾帧" file={end} onChange={setEnd} />
+                )}
+                {endSameAsStart && (
+                  <p className="text-xs text-mist pl-0.5">结尾帧与起始帧为同一张图。</p>
+                )}
+              </>
             )}
           </div>
         </div>

web/frontend/src/pages/VideoGen.tsx CHANGED Viewed

@@ -10,18 +10,21 @@ import {
   useGenerationOptions,
 } from "../context/GenerationOptionsContext";
 export function VideoGen() {
   const opts = useGenerationOptions();
   const durs = opts.video.durations_seconds;
   const defaultDur = durs.includes(4) ? 4 : durs[0] ?? 4;
   const [prompt, setPrompt] = useState("");
-  const [model, setModel] = useState(() =>
-    defaultFastModelValue(opts.video.models, "veo-3.0-fast-generate-001"),
-  );
   const [aspect, setAspect] = useState(opts.video.aspect_ratios[0] ?? "16:9");
   const [resolution, setResolution] = useState(
-    opts.video.resolutions[1] ?? opts.video.resolutions[0] ?? "1080p",
   );
   const [duration, setDuration] = useState(defaultDur);
   const [f0, setF0] = useState<File | null>(null);
@@ -59,7 +62,9 @@ export function VideoGen() {
     );
     const fallback = defaultFastModelValue(
       support,
-      support[0]?.value ?? "veo-3.0-fast-generate-001",
     );
     if (fallback) setModel(fallback);
   }, [hasRefs, model, opts.video.models]);
@@ -112,7 +117,7 @@ export function VideoGen() {
       <div className="mb-6 space-y-1.5">
         <p className="text-mist text-sm leading-relaxed">
           填写提示词；参考图最多三张。有参考图、1080p 或 4K 时固定 8 秒；720p 仅文字时可选 4/6
-          秒。Veo 3.1 Lite 不支持参考图与 4K。
         </p>
         <p className="text-xs text-mist leading-relaxed">
           Veo 3 与 Veo 3 Fast 为稳定版；Veo 3.1 系列为实验性，接口或效果可能变更。
@@ -219,7 +224,8 @@ export function VideoGen() {
             参考图（最多三张）
             {hasRefs && (
               <span className="block text-xs font-normal text-mist mt-1">
-                有参考图时勿选 Veo 3.1 Lite；可选 Veo 3、Veo 3 Fast、Veo 3.1 Fast 或 Veo 3.1（标准）。
               </span>
             )}
           </p>

   useGenerationOptions,
 } from "../context/GenerationOptionsContext";
+const DEFAULT_VIDEO_MODEL = "veo-3.1-fast-generate-preview";
 export function VideoGen() {
   const opts = useGenerationOptions();
   const durs = opts.video.durations_seconds;
   const defaultDur = durs.includes(4) ? 4 : durs[0] ?? 4;
   const [prompt, setPrompt] = useState("");
+  const [model, setModel] = useState(() => {
+    const preferred = opts.video.models.find((m) => m.value === DEFAULT_VIDEO_MODEL);
+    return preferred?.value ?? defaultFastModelValue(opts.video.models, DEFAULT_VIDEO_MODEL);
+  });
   const [aspect, setAspect] = useState(opts.video.aspect_ratios[0] ?? "16:9");
   const [resolution, setResolution] = useState(
+    opts.video.resolutions[0] ?? "720p",
   );
   const [duration, setDuration] = useState(defaultDur);
   const [f0, setF0] = useState<File | null>(null);
     );
     const fallback = defaultFastModelValue(
       support,
+      support.find((m) => m.value === DEFAULT_VIDEO_MODEL)?.value ??
+        support[0]?.value ??
+        DEFAULT_VIDEO_MODEL,
     );
     if (fallback) setModel(fallback);
   }, [hasRefs, model, opts.video.models]);
       <div className="mb-6 space-y-1.5">
         <p className="text-mist text-sm leading-relaxed">
           填写提示词；参考图最多三张。有参考图、1080p 或 4K 时固定 8 秒；720p 仅文字时可选 4/6
+          秒。Veo 3 与 Veo 3 Fast 仅支持纯提示词（无参考图）；Veo 3.1 Lite 不支持参考图与 4K。
         </p>
         <p className="text-xs text-mist leading-relaxed">
           Veo 3 与 Veo 3 Fast 为稳定版；Veo 3.1 系列为实验性，接口或效果可能变更。
             参考图（最多三张）
             {hasRefs && (
               <span className="block text-xs font-normal text-mist mt-1">
+                有参考图时勿选 Veo 3 / Veo 3 Fast（仅纯提示词）与 Veo 3.1 Lite；可选 Veo 3.1 Fast 或
+                Veo 3.1（标准）。
               </span>
             )}
           </p>