Video Generation

AI Gateway supports video generation, letting you create videos from text prompts, images, or video input. You can control resolution, duration, aspect ratio, and audio through a unified API across multiple providers.

To see all supported video models, use the Video filter at the AI Gateway Models page.

Some video models are tagged by capability in their model name. You can also see capability tags on the AI Gateway Models page or via the /v1/models endpoint, which is useful for models that support multiple capabilities:

For example, klingai/kling-v2.6-t2v is a text-to-video model, and alibaba/wan-v2.6-i2v is an image-to-video model.

These parameters work across all video models, though support varies by provider.

Video models return results in result.videos. Each video object contains:

• uint8Array: Raw video data as Uint8Array
• base64: Base64-encoded video data

import { experimental_generateVideo as generateVideo } from 'ai';
import fs from 'node:fs';
 
const result = await generateVideo({
  model: 'google/veo-3.1-generate-001',
  prompt: 'A serene mountain landscape at sunset',
  duration: 8,
});
 
fs.writeFileSync('output.mp4', result.videos[0].uint8Array);

Video generation can take several minutes. In Node.js, the default fetch implementation (via Undici) enforces a 5-minute timeout. This can cause requests to fail before the video finishes generating.

To extend these timeouts, create a custom gateway instance with a longer Undici Agent timeout:

import { createGateway } from 'ai';
import { Agent } from 'undici';
 
export const gateway = createGateway({
  fetch: (url, init) =>
    fetch(url, {
      ...init,
      dispatcher: new Agent({
        headersTimeout: 15 * 60 * 1000, // 15 minutes
        bodyTimeout: 15 * 60 * 1000,
      }),
    } as RequestInit),
});

Then use the custom gateway instance:

import { experimental_generateVideo as generateVideo } from 'ai';
import { gateway } from './lib/gateway';
 
const { videos } = await generateVideo({
  model: gateway.video('google/veo-3.1-generate-001'),
  prompt: 'A timelapse of a flower blooming',
  duration: 8,
});

To use plain string model IDs with extended timeouts, set your custom gateway as the global default provider. In a Next.js app, add this to instrumentation.ts:

import { createGateway } from 'ai';
import { Agent } from 'undici';
 
export async function register() {
  globalThis.AI_SDK_DEFAULT_PROVIDER = createGateway({
    fetch: (url, init) =>
      fetch(url, {
        ...init,
        dispatcher: new Agent({
          headersTimeout: 15 * 60 * 1000,
          bodyTimeout: 15 * 60 * 1000,
        }),
      } as RequestInit),
  });
}