Wan 3.0 now available on AI Gateway

Vercel’s AI Gateway now hosts Wan 3.0 from Alibaba, consolidating text-to-video, image-to-video, and frame-conditioning tasks into one model identified as alibaba/wan-v3.0-video. The update replaces separate Wan 2.7 endpoints for text-to-video and reference-to-video, which were limited to 15 seconds at 24fps. The new model supports clips up to 30 seconds at 30fps in resolutions of 480p, 720p, or 1080p, with audio included by default.

Users can generate videos asynchronously through AI Gateway, which processes requests in the background rather than keeping connections open. The process requires passing references under inputReferences, specifying URLs for audio and either URLs or base64-encoded data for images. Scene descriptions must be included in the prompt, and generation limits vary by reference type. First and last frame conditioning accepts one image each but cannot be combined with additional references.

The model playground on AI Gateway offers a no-code interface for testing Wan 3.0, while documentation provides guidance on asynchronous video generation and delivery verification. Vercel notes that audio references must be hosted URLs, and images can be provided as either URLs or base64-encoded strings. The updated model simplifies workflows by eliminating the need for multiple separate endpoints.

Vercel’s AI Gateway documentation outlines all supported methods for asynchronous video generation with Wan 3.0, including how to confirm successful delivery of outputs. The platform continues to expand its video model offerings, with Wan 3.0 joining a broader catalog accessible via the gateway. Users can explore additional video models through the gateway’s interface.