> ## Documentation Index
> Fetch the complete documentation index at: https://docs.haitoken.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Wan 3.0 Video Generation
> Generate, edit, and extend videos with native synchronized audio using Wan 3.0 All-in-One models
Wan 3.0 uses All-in-One models. Based on media types in `input.media` and the intent in your prompt, one model can handle text-to-video, first-frame or first-last-frame image-to-video, multimodal reference generation, video editing, and video extension. Unlike [Wan 2.7 Video Generation](/docs/en/api-reference/video/wan), you do not select separate text, image, or editing models for these capabilities.
This endpoint uses the Wan video task protocol. The `X-DashScope-Async` header is not required because async task semantics are built into the gateway.
## Request structure
The request body has three parts: `model`, `input`, and `parameters`.
| Field | Required | Description |
| -------------- | ------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| `model` | Yes | Platform model identifier: `wan3.0-video` or `wan3.0-video-prime` |
| `input.prompt` | At least one of `input.prompt` and `input.media` | Text prompt, up to 20,000 characters. In multimodal reference mode, use labels such as "image 1", "video 1", and "audio 1" to refer to assets in their array order |
| `input.media` | At least one of `input.prompt` and `input.media` | Media asset array. Every item contains `type` and `url` |
| `parameters` | No | Output resolution, ratio, duration, audio, random seed, prompt expansion, and watermark settings |
`media[].url` accepts a publicly accessible HTTP/HTTPS URL, an OSS temporary URL, or a base64 data URI. For a base64 data URI, the gateway stores the asset before sending it upstream.
## Authentication
Include the `Authorization` header in the format `Bearer YOUR_API_KEY`.
## Supported models
| Model identifier | Description |
| -------------------- | ---------------------------------------------- |
| `wan3.0-video` | Wan 3.0 general-purpose All-in-One video model |
| `wan3.0-video-prime` | Wan 3.0 Prime All-in-One video model |
## Capabilities and media types
Wan 3.0 can generate video with native synchronized audio, up to 30 seconds. Provide assets through `input.media`:
| `type` | Use | Limits and notes |
| ----------------- | -------------------------------------- | ---------------------------------------------------------------------------------------------- |
| `first_frame` | First-frame image-to-video | One image maximum; can be paired with `last_frame` for first-last-frame mode |
| `last_frame` | Last-frame image-to-video | One image maximum; may only be used with `first_frame` |
| `reference_image` | Reference images | Up to 10 images; each image must be 20 MB or smaller |
| `reference_video` | Reference, editing, or extension video | Up to 5 videos; 15 seconds total, 100 MB per video, and at least 16 fps |
| `reference_audio` | Reference audio | Up to 5 audio files; 15 seconds total and 15 MB per file |
| `file` | Document reference | One file maximum. Supports `docx`, `doc`, `xlsx`, `xls`, `pptx`, `ppt`, `pdf`, `txt`, and `md` |
| `link` | Public web-page reference | One public, no-login web page maximum |
`first_frame` / `last_frame` are mutually exclusive with `reference_image` / `reference_video` / `reference_audio` / `file` / `link`. Choose either `file` or `link`; either can be combined with reference images, video, and audio. The three `reference_*` types can be freely combined.
First-frame and first-last-frame mode only accept `first_frame` and an optional `last_frame`; do not add audio or other media types. To drive imagery with audio, use multimodal reference mode, for example `reference_image` plus `reference_audio`.
## Key parameters
| Parameter | Default | Description |
| --------------- | ---------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `resolution` | `1080P` | Output resolution: `480P`, `720P`, or `1080P` |
| `ratio` | `adaptive` | Aspect ratio: `16:9`, `4:3`, `1:1`, `3:4`, `9:16`, or `adaptive`. `adaptive` recommends a ratio from the input media and intent |
| `duration` | `5` | Output length in seconds. Without video input, use an integer from `2` to `30`; with video input, input-video length plus output length must not exceed 30 seconds. Use `-1` for smart duration |
| `audio` | `true` | Whether to include an audio track. `true` produces native synchronized audio; `false` omits the audio track. Both settings have the same price |
| `seed` | `-1` | Random seed. Use `-1` or an integer from `0` to `2147483647`; omit it or set `-1` for an automatically generated seed. The same seed does not guarantee identical output |
| `prompt_extend` | `true` | Whether to expand the prompt. Enabling it often improves short prompts but adds latency. When using `file` or `link`, enable it or omit it; do not set it to `false` |
| `watermark` | `false` | Whether to add a watermark |
For video editing, use `reference_video` and state the target edit, such as a replacement, removal, or modification, in the prompt. For video extension, explicitly state how to continue or extend the video. For either mode, `ratio: "adaptive"` and `duration: -1` are recommended.
## Quick request
This text-to-video request returns a task ID in `output.task_id`, which you then use to query the result.
```python Python theme={null}
import requests
response = requests.post(
"https://api.haitoken.ai/v1/alibaba/video/generations",
headers={
"Authorization": "Bearer YOUR_API_KEY",
"Content-Type": "application/json",
},
json={
"model": "wan3.0-video",
"input": {
"prompt": "A golden retriever runs along the beach at sunset; the camera follows slowly; ocean waves are clearly audible."
},
"parameters": {
"resolution": "720P",
"ratio": "16:9",
"duration": 5,
"prompt_extend": True,
},
},
)
print(response.json())
```
```javascript Node.js theme={null}
const response = await fetch('https://api.haitoken.ai/v1/alibaba/video/generations', {
method: 'POST',
headers: {
'Authorization': 'Bearer YOUR_API_KEY',
'Content-Type': 'application/json'
},
body: JSON.stringify({
model: 'wan3.0-video',
input: {
prompt: 'A golden retriever runs along the beach at sunset; the camera follows slowly; ocean waves are clearly audible.'
},
parameters: {
resolution: '720P',
ratio: '16:9',
duration: 5,
prompt_extend: true
}
})
});
console.log(await response.json());
```
```curl cURL theme={null}
curl -X POST 'https://api.haitoken.ai/v1/alibaba/video/generations' \
-H 'Authorization: Bearer YOUR_API_KEY' \
-H 'Content-Type: application/json' \
-d '{
"model": "wan3.0-video",
"input": {
"prompt": "A golden retriever runs along the beach at sunset; the camera follows slowly; ocean waves are clearly audible."
},
"parameters": {
"resolution": "720P",
"ratio": "16:9",
"duration": 5,
"prompt_extend": true
}
}'
```
## Scenario examples
### First-last-frame image to video
In first-last-frame mode, provide only the first and last frames. Do not mix in `reference_*`, `file`, or `link`.
```json theme={null}
{
"model": "wan3.0-video",
"input": {
"prompt": "The camera moves steadily through an empty street at dawn and ends at the entrance to a cafe, with cinematic lighting.",
"media": [
{
"type": "first_frame",
"url": "https://example.com/first-frame.jpg"
},
{
"type": "last_frame",
"url": "https://example.com/last-frame.jpg"
}
]
},
"parameters": {
"resolution": "720P",
"ratio": "adaptive",
"duration": 5,
"prompt_extend": true
}
}
```
### File-reference generation
Provide at most one `file`, and do not combine it with `link`. This mode must enable or omit `prompt_extend`.
```json theme={null}
{
"model": "wan3.0-video",
"input": {
"prompt": "Create a 10-second premium smart-glasses commercial from the product points in the file, with black, silver gray, and ice blue as the main palette.",
"media": [
{
"type": "file",
"url": "https://example.com/product-brief.pdf"
}
]
},
"parameters": {
"resolution": "720P",
"ratio": "16:9",
"duration": 10,
"prompt_extend": true
}
}
```
### Web-page and reference-image generation
Use a publicly accessible, no-login page for `link`. It can be combined with reference media such as `reference_image`.
```json theme={null}
{
"model": "wan3.0-video",
"input": {
"prompt": "Use the product information on the web page and the appearance in image 1 to create a concise product-launch video.",
"media": [
{
"type": "link",
"url": "https://example.com/product-page"
},
{
"type": "reference_image",
"url": "https://example.com/product-reference.jpg"
}
]
},
"parameters": {
"resolution": "720P",
"ratio": "16:9",
"duration": 8,
"prompt_extend": true
}
}
```
### Multimodal reference generation
Reference images, videos, and audio can be combined. Images, videos, and audio are each numbered in their array order.
```json theme={null}
{
"model": "wan3.0-video",
"input": {
"prompt": "Have the person in image 1 walk toward the camera in the beach setting from video 1, using the voice and rhythm from audio 1. Keep the result natural and realistic.",
"media": [
{
"type": "reference_image",
"url": "https://example.com/person.jpg"
},
{
"type": "reference_video",
"url": "https://example.com/beach.mp4"
},
{
"type": "reference_audio",
"url": "https://example.com/voice.wav"
}
]
},
"parameters": {
"resolution": "720P",
"ratio": "adaptive",
"duration": 5,
"prompt_extend": true
}
}
```
### Video editing
Use `reference_video` and state the requested edit directly.
```json theme={null}
{
"model": "wan3.0-video",
"input": {
"prompt": "Change the sky in the video to a sunset tone while preserving the person's movement and original dialogue.",
"media": [
{
"type": "reference_video",
"url": "https://example.com/input-video.mp4"
}
]
},
"parameters": {
"resolution": "720P",
"ratio": "adaptive",
"duration": -1,
"prompt_extend": true
}
}
```
### Video extension
Use `reference_video` and explicitly describe the continuation direction or next scene.
```json theme={null}
{
"model": "wan3.0-video",
"input": {
"prompt": "Extend the video forward. Keep panning right to reveal a broader mountain landscape and add natural ambient sound.",
"media": [
{
"type": "reference_video",
"url": "https://example.com/input-video.mp4"
}
]
},
"parameters": {
"resolution": "720P",
"ratio": "adaptive",
"duration": -1,
"prompt_extend": true
}
}
```
A successful creation returns the task ID:
```json theme={null}
{
"request_id": "8f3d2c1a-9b7e-4f5a-8c2d-1e6f0a9b3c5d",
"output": {
"task_id": "cgt-20260730120000-a1b2c3",
"task_status": "PENDING"
}
}
```
## Next steps
* See [Query Wan video task](/docs/en/api-reference/video/wan-status) to retrieve generation results
* Compare [Wan 2.7 Video Generation](/docs/en/api-reference/video/wan) for version differences
* See [Video generation](/docs/en/api-reference/video/generation) for the unified protocol
* See the [model list](/docs/en/api-reference/models/list-models) for available video models
## OpenAPI
````yaml en/api-reference/video/wan3/openapi.json POST /v1/alibaba/video/generations
openapi: 3.0.1
info:
title: Wan 3.0 Video Generation
version: 1.0.0
servers:
- url: https://api.haitoken.ai
security: []
paths:
/v1/alibaba/video/generations:
post:
summary: Create Wan 3.0 video task
description: >-
Create an async video task using Wan 3.0 All-in-One models. The
X-DashScope-Async header is not required; async task semantics are built
into the gateway.
parameters:
- name: Authorization
in: header
description: API key in the format Bearer YOUR_API_KEY
required: true
schema:
type: string
example: Bearer YOUR_API_KEY
- name: Content-Type
in: header
description: Request body type
required: true
schema:
type: string
example: application/json
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/AlibabaCreateTaskRequest'
responses:
'200':
description: Task created successfully
content:
application/json:
schema:
$ref: '#/components/schemas/AlibabaCreateTaskResponse'
example:
request_id: 8f3d2c1a-9b7e-4f5a-8c2d-1e6f0a9b3c5d
output:
task_id: cgt-20260730120000-a1b2c3
task_status: PENDING
components:
schemas:
AlibabaCreateTaskRequest:
type: object
required:
- model
- input
properties:
model:
type: string
description: >-
Wan 3.0 All-in-One model identifier: wan3.0-video or
wan3.0-video-prime
input:
$ref: '#/components/schemas/Input'
description: Input content (prompt and media elements)
parameters:
$ref: '#/components/schemas/Parameters'
description: Generation parameters
AlibabaCreateTaskResponse:
type: object
properties:
request_id:
type: string
description: Gateway request ID
output:
type: object
properties:
task_id:
type: string
description: Task ID for querying task status
task_status:
type: string
enum:
- PENDING
description: Task status
Input:
type: object
anyOf:
- required:
- prompt
- required:
- media
properties:
prompt:
type: string
maxLength: 20000
description: >-
Text prompt describing the video to generate, edit, or extend;
provide prompt or media at minimum. In multimodal reference mode,
use labels such as image 1, video 1, and audio 1 to refer to assets
in array order
media:
type: array
description: >-
Media element list (first frame / last frame / reference image /
reference video / reference audio / file / link)
items:
$ref: '#/components/schemas/MediaItem'
Parameters:
type: object
properties:
resolution:
type: string
enum:
- 480P
- 720P
- 1080P
default: 1080P
description: Output resolution
ratio:
type: string
enum:
- '16:9'
- '4:3'
- '1:1'
- '3:4'
- '9:16'
- adaptive
default: adaptive
description: Output aspect ratio. Use adaptive when providing input media
duration:
type: integer
oneOf:
- enum:
- -1
- minimum: 2
maximum: 30
default: 5
description: >-
Output duration in seconds. Without video input, use 2 to 30; with
video input, total input-video plus output duration cannot exceed
30. Use -1 for smart duration
prompt_extend:
type: boolean
default: true
description: >-
Whether to automatically expand the prompt. It can improve short
prompts but adds latency. Do not set it to false when using file or
link
audio:
type: boolean
default: true
description: >-
Whether to output native synchronized audio. Enabled by default; set
to false to disable (toggling audio does not change the price)
watermark:
type: boolean
default: false
description: Whether to add a watermark
seed:
type: integer
format: int64
minimum: -1
maximum: 2147483647
default: -1
description: >-
Random seed. Use -1 or 0 to 2147483647; omit it or use -1 for an
automatically generated seed. The same seed does not guarantee
identical output
MediaItem:
type: object
required:
- type
- url
properties:
type:
type: string
description: >-
Media type. first_frame / last_frame / reference_image /
reference_video (reference, editing, or extension) / reference_audio
/ file (mutually exclusive with link) / link (mutually exclusive
with file)
url:
type: string
format: uri
description: >-
Media URL or base64 data URI (base64 is uploaded to OBS by the
gateway and replaced with a URL)
````