# Wan v2.5 Text-to-Video Preview

The foundation of Alibaba Cloud's Wan text-to-video lineup. Wan v2.5 Text-to-Video Preview established the generation pipeline that later informed the 2.6 series, and it remains available as a practical entry point for teams exploring AI video generation at lower resolution tiers.

- **Model ID:** `alibaba/wan-v2.5-t2v-preview`
- **Type:** video
- **Providers:** alibaba
- **Context window:** Not applicable
- **Maximum output tokens:** 0
- **Pricing:** See the provider details below
- **Canonical page:** https://vercel.com/ai-gateway/models/wan-v2.5-t2v-preview

## Supported parameters

- **Parameters:** `aspectRatio`, `resolution`, `duration`, `fps`, `seed`
- **Operations:** `text-to-video`
- **Input modalities:** `text`
- **Output modalities:** `video`
- **Aspect ratios:** `16:9`, `1:1`, `16:9 landscape`, `9:16`
- **Resolutions:** `1280x720`, `960x960`
- **Durations:** `5s`, `10s`
- **Frame rates:** `24 fps`
- **Maximum outputs per request:** Not reported

## Example

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

const result = await generateVideo({
  model: 'alibaba/wan-v2.5-t2v-preview',
  prompt: 'A serene mountain lake at sunrise.'
});
```

## About

Wan v2.5 Text-to-Video Preview marked Alibaba Cloud's initial public release of the Wan text-to-video architecture. Given only a free-form text prompt, it produces single-shot video clips up to 10 seconds long, with output available at 480p, 720p, or 1080p across three aspect ratios: landscape (16:9), portrait (9:16), and square (1:1).

What set this release apart from many early text-to-video models was its integrated audio generation. Rather than rendering silent video and requiring a separate dubbing pass, the 2.5 pipeline synthesizes ambient sound, effects, and even prompted character dialogue with lip-sync, all within a single generation call. For workflows that need audio-visual output, this removes an entire post-processing step.

The preview designation means the model is intended primarily for evaluation and prototyping. Teams can use it to develop prompt strategies, validate resolution and aspect ratio choices, and estimate costs at the 480p tier before scaling up to the production-grade Wan 2.6 models.

## What to consider

Preview-tier models may exhibit higher generation latency than their stable counterparts. If your workflow has strict turnaround requirements, benchmark generation times at your target resolution before integrating into production.

## When to use

### Best For

- Evaluating Alibaba Cloud's Wan text-to-video pipeline without committing to the higher-capability 2.6 series
- Generating short social clips (up to 10 seconds) from text descriptions, particularly in portrait 9:16 format
- Workflows that need audio baked into the generated video without a separate dubbing tool
- Running cost explorations at the 480p resolution tier before budgeting for higher-fidelity output

### Consider Alternatives When

- If your project calls for clips longer than 10 seconds or automatic scene transitions, wan-v2.6-t2v extends to 15 seconds with multi-shot storytelling
- If you already have a source image to animate, wan-v2.6-i2v and wan-v2.6-i2v-flash handle image-to-video conversion
- If consistent character identity across scenes is essential, wan-v2.6-r2v provides reference-based identity transfer

## Best for

- **Evaluating the wan pipeline:** Exploring Alibaba Cloud's text-to-video system without committing to the higher-capability 2.6 series
- **Short social clips:** Generating videos up to 10 seconds from text descriptions, particularly in portrait 9:16 format
- **Baked-in audio workflows:** Use cases that need audio in the generated video without a separate dubbing tool
- **480p cost exploration:** Running cost tests at the lower resolution tier before budgeting for higher-fidelity output

## Consider alternatives

- **Longer or multi-shot clips:** wan-v2.6-t2v extends to 15 seconds with multi-shot storytelling and automatic scene transitions
- **Image-to-video conversion:** wan-v2.6-i2v and wan-v2.6-i2v-flash handle animation from a source image
- **Consistent character identity:** wan-v2.6-r2v provides reference-based identity transfer across scenes

## Frequently asked questions

### Why would I choose the v2.5 preview over the newer v2.6 model?

The v2.5 preview supports 480p output (which v2.6 T2V doesn't), making it a lower-cost option for draft-quality renders and prompt experimentation. It also serves as a lighter-weight entry point for teams still evaluating the Wan pipeline.

### Can this model generate vertical video for mobile platforms?

Yes. The 9:16 aspect ratio produces portrait-oriented output suitable for platforms like TikTok, Instagram Reels, and YouTube Shorts.

### How does the built-in audio feature work?

Audio is generated in the same rendering pass as the video. The model produces ambient sound, effects, and, if the prompt describes speech, character dialogue with lip-sync, all without requiring a separate audio generation tool.

### What kind of text prompts work best with this model?

Descriptive scene prompts that specify setting, action, and mood tend to produce the most coherent output. Including details about lighting, camera angle, and desired audio cues gives the model more information to work with.

### Is there a way to control video duration precisely?

You can request specific durations within the model's range. The maximum is 10 seconds; for longer output, use the Wan v2.6 T2V model.

### Do I need special Vercel plan access to use this model?

Access requires a Pro or Enterprise plan or paid AI Gateway usage.

## Links

- [Model page](https://vercel.com/ai-gateway/models/wan-v2.5-t2v-preview)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://wan.video/)
- [Provider pricing](https://modelstudio.console.alibabacloud.com/ap-southeast-1/?tab=doc#/doc/?type=model&url=prices)