Prismix
Vercel — AI · · 2 min read

Wan 3.0 now available on AI Gateway

#model-release #version-bump #partnership #developer-tool #music

Mirrored from Vercel — AI for archival readability. Support the source by reading on the original site.

Like Read original ↗

Wan 3.0 from Alibaba is now available on AI Gateway as alibaba/wan-v3.0-video.

One model covers text to video, image to video, first and last frame conditioning, and reference-based generation, and it takes image, video, and audio as references. Clips run up to 30 seconds at 30fps, in 480p, 720p, or 1080p, and the output carries audio by default.

That consolidates what took several models before. Wan 2.7 shipped as separate -t2v and -r2v IDs, each capped at 15 seconds and 24fps.

Copy link to headingGenerating a video

Try the new asynchronous generation on AI Gateway, which runs the job in the background instead of holding a request open. Three routes do it, and they differ in what waits.

Pass poll and the gateway tracks the generation while the SDK makes short status requests until it lands. This is the smallest change from a plain call.

import { experimental_generateVideo as generateVideo } from 'ai';
import fs from 'node:fs';


const { videos } = await generateVideo({
  model: 'alibaba/wan-v3.0-video',
  prompt: 'A paper lantern drifting over a harbor at night',
  duration: 10,
  resolution: '1080p',
  poll: { intervalMs: 5000, timeoutMs: 600000 },
fs.writeFileSync('output.mp4', videos[0].uint8Array);

experimental_startVideo returns as soon as the gateway accepts the job, so nothing waits and the job outlives the process that started it. That suits serverless functions, queues, and fanning out a batch.

const { operation } = await startVideo({
  model: 'alibaba/wan-v3.0-video',
  prompt: 'A paper lantern drifting over a harbor at night',
// Later, from this process or another one:
const status = await getVideoStatus('alibaba/wan-v3.0-video', { operation });

Or pass a webhook and the gateway posts one event to your endpoint when the generation lands, so nothing polls at all.

const { videos } = await generateVideo({
  model: 'alibaba/wan-v3.0-video',
  prompt: 'A paper lantern drifting over a harbor at night',
  webhook: async () => ({ url: callbackUrl, received: waitForDelivery(token) }),

The docs cover all three, including how to verify a delivery.

Copy link to headingReferences

Pass references under inputReferences, each with its URL and media type, and describe the scene in the prompt. The limits differ by kind. Audio references have to be hosted URLs, where images also accept base64. First and last frame conditioning takes one image each and cannot be combined with references.

You can also try Wan 3.0 with no code in the model playground. To see more, browse all video models.

Discussion (0)

Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.

Sign in →

No comments yet. Sign in and be the first to say something.

More from Vercel — AI