From a8daf65dd45c2cf584a04ccdb9a68f2938657290 Mon Sep 17 00:00:00 2001
From: SecurityQQ <alexmsecurity@gmail.com>
Date: Wed, 19 Aug 2026 06:17:08 -0700
Subject: [PATCH] fix: resolve string video model IDs via
 AI_SDK_DEFAULT_PROVIDER
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

`<Video model="seedance_2_5" />` passed a string to generateVideo(),
which called model.doGenerate() directly — strings have no doGenerate,
causing "model.doGenerate is not a function" on every clip.

generateImage() from the Vercel AI SDK already resolves string IDs via
globalThis.AI_SDK_DEFAULT_PROVIDER (set by the render service before
render()). This adds the same resolution to generateVideo(), plus a
toVideoModelV3() shell for preview mode (matching toImageModelV3()).

VideoProps.model, TalkingHeadProps.model/lipsyncModel, and
DefaultModels.video now accept VideoModel (string | VideoModelV3).

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 src/ai-sdk/generate-video.ts  | 40 +++++++++++++++++++++++++++++++++--
 src/ai-sdk/index.ts           |  1 +
 src/react/renderers/render.ts | 30 ++++++++++++++++++++++++--
 src/react/types.ts            |  9 ++++----
 4 files changed, 72 insertions(+), 8 deletions(-)

diff --git a/src/ai-sdk/generate-video.ts b/src/ai-sdk/generate-video.ts
index 174dc36d..1223ab4a 100644
--- a/src/ai-sdk/generate-video.ts
+++ b/src/ai-sdk/generate-video.ts
@@ -15,8 +15,16 @@ export type GenerateVideoPrompt =
       video?: DataContent | Array<DataContent>;
     };
 
+/**
+ * A video model can be passed as a string ID (resolved via the default
+ * provider set on `globalThis.AI_SDK_DEFAULT_PROVIDER`) or as a full
+ * `VideoModelV3` object. This mirrors how the Vercel AI SDK's
+ * `generateImage` accepts string model IDs.
+ */
+export type VideoModel = VideoModelV3 | string;
+
 export interface GenerateVideoOptions {
-  model: VideoModelV3;
+  model: VideoModel;
   prompt: GenerateVideoPrompt;
   n?: number;
   resolution?: `${number}x${number}`;
@@ -122,11 +130,37 @@ function normalizePrompt(prompt: GenerateVideoPrompt): {
   };
 }
 
+/**
+ * Resolve a `VideoModel` (which may be a string ID) into a `VideoModelV3`
+ * object. String IDs are resolved through `globalThis.AI_SDK_DEFAULT_PROVIDER`
+ * (set by the render service before calling `render()`), mirroring how the
+ * Vercel AI SDK's `generateImage` resolves string image model IDs.
+ */
+function resolveVideoModel(model: VideoModel): VideoModelV3 {
+  if (typeof model !== "string") {
+    return model;
+  }
+
+  const provider = globalThis.AI_SDK_DEFAULT_PROVIDER as
+    | { videoModel?: (id: string) => VideoModelV3 }
+    | undefined;
+
+  if (!provider?.videoModel) {
+    throw new Error(
+      `Cannot resolve video model "${model}" — no default provider with videoModel() is set. ` +
+        'Either pass a VideoModelV3 object (e.g. varg.videoModel("seedance_2_5")) ' +
+        "or ensure globalThis.AI_SDK_DEFAULT_PROVIDER is configured.",
+    );
+  }
+
+  return provider.videoModel(model);
+}
+
 export async function generateVideo(
   options: GenerateVideoOptions,
 ): Promise<GenerateVideoResult> {
   const {
-    model,
+    model: modelArg,
     prompt: promptArg,
     n = 1,
     resolution,
@@ -141,6 +175,8 @@ export async function generateVideo(
 
   const { prompt, files } = normalizePrompt(promptArg);
 
+  const model = resolveVideoModel(modelArg);
+
   const result = await model.doGenerate({
     prompt: prompt ?? "",
     n,
diff --git a/src/ai-sdk/index.ts b/src/ai-sdk/index.ts
index 28fdde92..5ade1bde 100644
--- a/src/ai-sdk/index.ts
+++ b/src/ai-sdk/index.ts
@@ -26,6 +26,7 @@ export {
   type GenerateVideoPrompt,
   type GenerateVideoResult,
   generateVideo,
+  type VideoModel,
 } from "./generate-video";
 export {
   generatePlaceholder,
diff --git a/src/react/renderers/render.ts b/src/react/renderers/render.ts
index ee0acb3c..09374532 100644
--- a/src/react/renderers/render.ts
+++ b/src/react/renderers/render.ts
@@ -22,6 +22,7 @@ import type {
   Layer,
   VideoLayer,
 } from "../../ai-sdk/providers/editly/types";
+import type { VideoModelV3 } from "../../ai-sdk/video-model";
 import { ResolvedElement } from "../resolved-element";
 import type {
   ClipProps,
@@ -90,6 +91,29 @@ function toImageModelV3(
   };
 }
 
+function toVideoModelV3(
+  model: Parameters<typeof generateVideo>[0]["model"],
+): VideoModelV3 {
+  if (typeof model === "object" && model.specificationVersion === "v3") {
+    return model;
+  }
+  // for string IDs, create a shell that satisfies the type.
+  // in preview mode the middleware intercepts before doGenerate is called;
+  // in strict mode generateVideo() resolves the string via the default provider.
+  const modelId = typeof model === "string" ? model : model.modelId;
+  return {
+    specificationVersion: "v3",
+    provider: "placeholder",
+    modelId,
+    maxVideosPerCall: 1,
+    doGenerate: async () => {
+      throw new Error(
+        `toVideoModelV3 shell: doGenerate should not be called in preview mode (model: ${modelId})`,
+      );
+    },
+  };
+}
+
 export async function renderRoot(
   element: VargElement<"render">,
   options: RenderOptions,
@@ -179,7 +203,7 @@ export async function renderRoot(
       return cachedGenerateVideo({
         ...opts,
         model: wrapVideoModel({
-          model: opts.model,
+          model: toVideoModelV3(opts.model),
           middleware: placeholderFallbackMiddleware({
             mode: "preview",
             onFallback: () => {},
@@ -190,7 +214,9 @@ export async function renderRoot(
     }
 
     const result = await cachedGenerateVideo(opts);
-    emitPricing("video", opts.model.modelId, result);
+    const videoModelId =
+      typeof opts.model === "string" ? opts.model : opts.model.modelId;
+    emitPricing("video", videoModelId, result);
     return result;
   };
 
diff --git a/src/react/types.ts b/src/react/types.ts
index 874a5fd5..ec998221 100644
--- a/src/react/types.ts
+++ b/src/react/types.ts
@@ -6,6 +6,7 @@ import type {
 } from "@ai-sdk/provider";
 import type { CacheStorage } from "../ai-sdk/cache";
 import type { File } from "../ai-sdk/file";
+import type { VideoModel } from "../ai-sdk/generate-video";
 import type { MusicModelV3 } from "../ai-sdk/music-model";
 import type { FFmpegBackend } from "../ai-sdk/providers/editly/backends";
 import type {
@@ -164,7 +165,7 @@ export type VideoProps = BaseProps &
   TrimProps & {
     prompt?: VideoPrompt;
     src?: string;
-    model?: VideoModelV3;
+    model?: VideoModel;
     resize?: ResizeMode;
     cropPosition?: CropPosition;
     aspectRatio?: `${number}:${number}`;
@@ -206,9 +207,9 @@ export interface TalkingHeadProps extends BaseProps {
   /** Pre-resolved or lazy speech element to use as the audio track. */
   audio?: VargElement<"speech">;
   /** Lipsync video model (e.g. fal.videoModel("sync-v2-pro")). */
-  model?: VideoModelV3;
+  model?: VideoModel;
   /** Separate lipsync model override (defaults to `model`). */
-  lipsyncModel?: VideoModelV3;
+  lipsyncModel?: VideoModel;
   /** Video resolution for lipsync generation (default: "720p") */
   resolution?: "480p" | "720p" | "1080p";
   position?:
@@ -355,7 +356,7 @@ export type RenderMode = "strict" | "preview";
 
 export interface DefaultModels {
   image?: ImageModelV3;
-  video?: VideoModelV3;
+  video?: VideoModel;
   speech?: SpeechModelV3;
   music?: MusicModelV3;
   transcription?: TranscriptionModelV3;