# Mixed Video Models

https://video-doc.1route.dev/en/docs/mixed-models



See the [unified video API](/en/docs/video-api/) for shared fields, errors and delivery. The [complete catalog](/video-models.json) and [OpenAPI](/openapi.json) cover all models.

Use a video-service key in the 低价混合 group and verify access with `GET /v1/models`. These models are configured; paid generation has not been tested.

[Model catalog](/mixed-models.json) · [OpenAPI](/mixed-openapi.json)

| API model ID                 | Seconds | Resolution              | Images/videos/audio | Price          |
| ---------------------------- | ------- | ----------------------- | ------------------- | -------------- |
| `sd2.0(fast903固定15秒720P)`    | 15      | 720p                    | 9/0/3               | ¥1.836/request |
| `sd2.5(全参过人脸720P)`           | 4–30    | 720p                    | 30/10/10            | ¥0.636/second  |
| `minimax-h3(933/4-15秒/2K)`   | 4–15    | 2k                      | 9/3/3               | ¥1.86/request  |
| `minimax-h3(933/4-15秒/768p)` | 4–15    | 768p                    | 9/3/3               | ¥1.38/request  |
| `sd2.0(933过人脸4-15秒)`         | 4–15    | 720p                    | 9/3/3               | ¥1.14/request  |
| `sd2.0(满血933过人脸4-15秒)`       | 4–15    | 720p                    | 9/3/3               | ¥2.40/request  |
| `wan3.0(满血10-5-5)`           | 4–30    | Default; omit parameter | 10/5/5              | ¥2.64/request  |

## Request fields [#request-fields]

Create with `POST /v1/videos`. Required fields are `model`, nonempty `prompt`, and integer `seconds` within the listed range. Fast903 requires exactly 15 seconds. Use `aspect_ratio` for all models that support ratio control; legacy `ratio` remains compatible. Allowed values depend on the model catalog.

References use public URL string arrays: `images`, `videos`, and `audios`. Audio must be real MP3 with a URL path ending in `.mp3`. Fast903 does not accept video references. The Seedance face-reference models accept optional boolean `face`; do not send it to other models.

WAN3 Full uses its model-default resolution: omit `resolution`. Its request and billing contract is separate from the existing WAN3 direct models; do not add reference-video seconds to its price.

## Billing [#billing]

Prices are CNY balance units for the 低价混合 group. The six per-request models charge once per successful task, without multiplying by duration or reference count. WAN3 Full is billed per request.

`sd2.5(全参过人脸720P)` reserves requested seconds × ¥0.636: 4 seconds reserves ¥2.544 and 10 seconds reserves ¥6.36. Valid measured completion usage replaces that estimate. Missing or invalid usage retains the reservation for manual reconciliation; a request echo is not measured usage. Failed tasks follow the service refund process.

## Example [#example]

```json
{
  "model": "sd2.0(933过人脸4-15秒)",
  "prompt": "A kite moves gently over a green field",
  "seconds": 4,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "face": false
}
```

Save the submitted task ID and poll `GET /v1/videos/{id}` every 15 seconds. Download from this service’s `metadata.url` or `url` on completion. Signatures last up to about 5 minutes, with about 10 minutes of storage. Do not automatically repeat creation after a 5xx or timeout.