# Seedance 2.5

The generation after Seedance 2.0. Seedance 2.5 produces 30-second clips at native 4K, and takes up to 50 reference inputs, so composition, motion, and style can be directed far more precisely than by prompt alone.

- **Model ID:** `bytedance/seedance-2.5`
- **Type:** video
- **Providers:** bytedance
- **Context window:** Not applicable
- **Maximum output tokens:** 0
- **Pricing:** See the provider details below
- **Canonical page:** https://vercel.com/ai-gateway/models/seedance-2.5

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { experimental_generateVideo as generateVideo } from 'ai';

const result = await generateVideo({
  model: 'bytedance/seedance-2.5',
  prompt: 'A serene mountain lake at sunrise.'
});
```

## About

Seedance 2.5 is the generation after Seedance 2.0 in ByteDance's Seed family, skipping several version numbers to signal the size of the jump.

Two capabilities define it. Single-shot length reaches 30 seconds, with multi-turn extension for longer sequences. Resolution is native 4K rather than upscaled from a smaller render, which matters for professional pipelines where an upscale is visible, and 10-bit colour depth leaves headroom for grading in post.

The larger change is reference capacity. Seedance 2.5 accepts up to 50 multimodal inputs in a single request, spanning images, video, audio clips, 3D white models, and style references, against 12 in its predecessor. That is what makes it directable: style, motion, and composition can be shown rather than described. ByteDance also reports roughly 20 percent better prompt adherence, which reduces the number of generations needed to reach a usable result.

Audio and video are generated together by one model rather than dubbed afterwards, inheriting the joint architecture introduced with Seedance 2.0.

Generate video with `generateVideo` from the AI SDK.

## What to consider

Reference capacity is the reason to choose Seedance 2.5, and it is also the work. Fifty inputs give fine control over style, motion, and composition, but assembling and curating that reference set is a real production step. On a simple prompt-to-clip task the earlier Seedance models reach a usable result with less setup.

Native 4K and 30-second duration both raise cost per generation compared with shorter, smaller output. Prototype at lower resolution and duration, then generate the final at full settings. See the pricing panel on this page for current rates.

Regional availability has been staged. Confirm the model is reachable for your account and region before you design a pipeline around it.

## When to use

### Best For

- **Native 4K delivery** where an upscale from a smaller render would be visible
- **Reference-heavy direction** using up to 50 multimodal inputs in one request
- **Longer single shots** reaching 30 seconds before extension is needed
- **Colour grading workflows** that benefit from 10-bit depth in post
- **Joint audio and video** produced together rather than dubbed afterwards

### Consider Alternatives When

- **Simple prompt-to-clip work** is the task, where earlier Seedance models need less setup
- **Cost per generation** is the constraint, since 4K and 30 seconds both raise it
- **Short clips** are all you need, where a lite Seedance variant is cheaper
- **Still images** are the output, where Seedream is the right family

## Best for

- **Native 4K Delivery:** No visible upscale from a smaller render
- **Reference-Heavy Direction:** Up to 50 multimodal inputs in one request
- **Longer Single Shots:** 30 seconds before extension is needed
- **Colour Grading Workflows:** 10-bit depth leaving headroom in post
- **Joint Audio And Video:** Generated together rather than dubbed

## Consider alternatives

- **Simple Prompt-To-Clip Work:** Earlier Seedance models need less setup
- **Cost-Constrained Generation:** 4K and 30 seconds both raise cost per clip
- **Short Clip Needs:** A lite Seedance variant is cheaper
- **Still Image Output:** Seedream is the right family

## Frequently asked questions

### How long can a Seedance 2.5 clip be?

Up to 30 seconds in a single generation, with multi-turn extension for longer sequences.

### Is Seedance 2.5 really 4K?

Yes, generated natively at 4K rather than upscaled from a lower resolution render. It also supports 10-bit colour depth for grading in post.

### How many reference inputs does Seedance 2.5 accept?

Up to 50 in one request, spanning images, video, audio clips, 3D white models, and style references. Its predecessor accepted 12.

### How do I call Seedance 2.5 on AI Gateway?

Use `generateVideo` from the AI SDK with `bytedance/seedance-2.5`. The Chat Completions, Responses, and Messages APIs do not serve video generation.

### Does Seedance 2.5 generate audio?

Yes. Audio and video are produced together by one model, inheriting the joint generation architecture introduced with Seedance 2.0.

### How does Seedance 2.5 compare to Seedance 2.0?

Longer single shots, native 4K, and a much larger reference budget of 50 inputs against 12. ByteDance also reports roughly 20 percent better prompt adherence.

### Does Seedance 2.5 support Zero Data Retention?

Zero Data Retention is not currently available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

## Links

- [Model page](https://vercel.com/ai-gateway/models/seedance-2.5)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://ai.byteplus.com/ark/promotion?modelName=seedance-2-5)


If you encounter incorrect information, can’t find what you need, need a capability Vercel doesn’t currently offer, or have any other feedback to report, please do it by following the instructions at https://vercel.com/feedback.md.