ByteDance's Seedance 2.5 Now Available on Vercel AI Gateway

ByteDance's Seedance 2.5 is now available on Vercel AI Gateway. It generates videos up to 30 seconds long in a single pass, maintaining camera movement and scene continuity without stitching multiple scenes together. Short videos can be extended by continuing characters, scenes, and camera movements.

Text, image, video, and audio inputs can be combined in a single request, allowing the appearance of one asset to be combined with the motion or camera work of another. In the final video, backgrounds, products, and characters can be replaced while maintaining frames, camera, and video pacing to create multiple versions, and prompts can be written in over 10 languages.

In the AI SDK, set model to bytedance/seedance-2.5 and call generateVideo to generate videos. Resolution, length, and audio generation options can also be specified in the request.

For image references, pass image URLs in inputReferences and specify them in the prompt as [Image 1], [Image 2], etc. Video references are numbered separately from images, so [Video 1] and [Image 1] can be used in the same prompt.

Reference assets serve to provide motion, camera work, pacing, etc., while generating new videos. In contrast, editing features directly modify completed videos, so mixing these two roles in a single prompt may reduce the stability of the results.