Seedance 2.5 Flash now available on AI Gateway
ByteDance's Seedance 2.5 Flash is now available on Vercel AI Gateway, generating 4 to 30 second clips at up to 720p and 24 frames per second with audio.
In short
AI SDK 7 developers on AI Gateway · 2.5 Flash, clips of 4 to 30 seconds · Now on Vercel AI Gateway · Call bytedance/seedance-2.5-flash
By Dillip Chowdary • Nov 16, 2026 • Source: Vercel Blog

At a glance
- Who it is for
- AI SDK 7 developers on AI Gateway
- The number
- 2.5 Flash, clips of 4 to 30 seconds
- Available
- Now on Vercel AI Gateway
- What to do
- Call bytedance/seedance-2.5-flash
- Model id
- bytedance/seedance-2.5-flash
- Frame rate
- 24 frames per second
In this post
ByteDance's Seedance 2.5 Flash is now available on AI Gateway, and Vercel Blog's report sets clip length at 4 to 30 seconds, resolution at up to 720p, and frame rate at 24 frames per second.
This guide is for developers using AI SDK 7 on AI Gateway who need the model id, the clip limits, and the sample calls. They can generate video with synchronized audio from a text prompt, pass image and video references, or try Seedance 2.5 Flash in the model playground.
01Faster lower-cost model variant
Seedance 2.5 Flash is a faster, lower-cost version of Seedance 2.5. The output is close to the standard model in human realism, reference-based editing, and prompt adherence. On AI Gateway the samples set model to bytedance/seedance-2.5-flash and run that call with AI SDK 7.
Speed and cost are the two differences named against Seedance 2.5. Flash is faster, and Flash is lower-cost. Human realism, reference-based editing, and prompt adherence stay close to the standard model, and the samples pass bytedance/seedance-2.5-flash to experimental_generateVideo, imported as generateVideo from the ai package.
| Measure | Seedance 2.5 | Seedance 2.5 Flash |
|---|---|---|
| Speed | Standard Seedance 2.5 | Faster |
| Cost | Standard Seedance 2.5 | Lower-cost |
| Human realism | Standard model | Close to the standard model |
| Reference-based editing | Standard model | Close to the standard model |
| Prompt adherence | Standard model | Close to the standard model |
02Clip length and frame formats
Seedance 2.5 Flash generates video with synchronized audio. A clip can start from a text prompt, or from image, video, and audio references. Length runs from 4 to 30 seconds at up to 720p and 24 frames per second, in landscape, square, or portrait.
Developers can edit an existing video or extend one, and they can provide images that set the first frame and the last frame. Those controls sit in the same Seedance 2.5 Flash description as the 4 to 30 second length. The stated ceiling remains up to 720p at 24 frames per second, in landscape, square, or portrait.
The text-prompt sample sets aspectRatio to 16:9 and duration to 5. A duration of 5 sits inside the stated range of 4 to 30 seconds. That sample still asks for video with synchronized audio, under the same up to 720p and 24 frames per second limits.
| Spec | Stated value |
|---|---|
| Clip length | 4 to 30 seconds |
| Resolution | Up to 720p |
| Frame rate | 24 frames per second |
| Formats | Landscape, square, or portrait |
| Sample aspectRatio | 16:9 |
| Sample duration | 5 |
| Output audio | Synchronized audio |
| Reference inputs | Image, video, and audio |
03Text prompt calls in AI SDK
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
AI SDK 7 is the version named for a video with audio from a text prompt. The sample imports experimental_generateVideo as generateVideo from the ai package, awaits generateVideo, and reads videos from the result. The prompt is "A paper boat drifts down a rain-soaked city street at dusk.", with model bytedance/seedance-2.5-flash, aspectRatio 16:9, and duration 5.
import { experimental_generateVideo as generateVideo } from 'ai';
const { videos } = await generateVideo({
model: 'bytedance/seedance-2.5-flash',
prompt: 'A paper boat drifts down a rain-soaked city street at dusk.',
aspectRatio: '16:9',
duration: 5,
});The object passed to generateVideo carries model, prompt, aspectRatio, and duration. The model string is bytedance/seedance-2.5-flash, the prompt is the paper-boat line, aspectRatio is 16:9, and duration is 5. That call is the text-prompt path for a clip with synchronized audio, and the awaited result is videos.
04Image and video reference inputs
Reference images and videos are passed in inputReferences, which the changelog describes as their URLs. Images and videos are numbered separately in the prompt as [Image 1], [Video 1], and so on. The sample prompt is "The lion from [Image 1] walks through a sunlit meadow, filmed like [Video 1]."
import { experimental_generateVideo as generateVideo } from 'ai';
const { videos } = await generateVideo({
model: 'bytedance/seedance-2.5-flash',
prompt: 'The lion from [Image 1] walks through a sunlit meadow, filmed like [Video 1].',
aspectRatio: '16:9',
duration: 5,
inputReferences: [
{
data: '...',
mediaType: 'image/jpeg',
},
{
data: '...',
mediaType: 'video/mp4',
},
],
});The inputReferences array has two objects: the first with data '...' and mediaType image/jpeg, and the second with data '...' and mediaType video/mp4. The model stays bytedance/seedance-2.5-flash, aspectRatio stays 16:9, and duration stays 5. The awaited call still returns videos from generateVideo, the alias of experimental_generateVideo imported from the ai package.
Audio references are named with image and video references as inputs that can drive a clip. This sample shows one image/jpeg and one video/mp4, cited in the prompt as [Image 1] and [Video 1]. Edit, extend, and images for the first and last frames are the other controls named beside that reference list.
05Model playground and quickstart
Developers can try Seedance 2.5 Flash in the model playground, or follow the video generation quickstart. The model id for that trial is bytedance/seedance-2.5-flash on AI Gateway. The quickstart pattern in the entry is experimental_generateVideo from the ai package, awaited as generateVideo, returning videos.
A text-prompt run sends model, prompt, aspectRatio, and duration, using the paper-boat prompt, 16:9, and duration 5 in the sample. A reference run adds inputReferences with data and mediaType, using image/jpeg and video/mp4, numbered in the prompt as [Image 1] and [Video 1]. Both follow the published clip rules of 4 to 30 seconds, up to 720p, and 24 frames per second, with synchronized audio, in landscape, square, or portrait.
Note: The helper in both samples is experimental_generateVideo from the ai package, aliased as generateVideo, and each reference data value is printed as '...'. Clips run 4 to 30 seconds at up to 720p and 24 frames per second, with synchronized audio, and developers can edit or extend a video or set the first and last frames with images.
Frequently asked questions
What is ByteDance Seedance 2.5 Flash?
Seedance 2.5 Flash is a faster, lower-cost version of Seedance 2.5, now available on Vercel AI Gateway as bytedance/seedance-2.5-flash. The output is close to the standard model in human realism, reference-based editing, and prompt adherence.
How long can Seedance 2.5 Flash clips be?
Clips can be 4 to 30 seconds long at up to 720p and 24 frames per second, in landscape, square, or portrait, with synchronized audio. The AI SDK samples set duration to 5 and aspectRatio to 16:9.
How do you call Seedance 2.5 Flash in the AI SDK?
With AI SDK 7, import experimental_generateVideo as generateVideo from the ai package and await generateVideo. Set model to bytedance/seedance-2.5-flash, and pass prompt, aspectRatio, and duration. The call returns videos.
How do image and video references work?
Pass them in inputReferences. The sample uses data '...' with mediaType image/jpeg and mediaType video/mp4. Number images and videos separately in the prompt as [Image 1], [Video 1], and so on.
Where can you try Seedance 2.5 Flash?
Vercel names the model playground and the video generation quickstart. The gateway model string is bytedance/seedance-2.5-flash. Developers can also edit or extend existing videos, or provide images for the first and last frames.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement