> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dimilinks.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Create video generation (MiniMax-H3)

> Generate 4-15 second videos with minimax-h3, with image, video, and audio reference support.

`minimax-h3` is DimiLinks' multimodal video generation model in the `seedance/h3` group. Billing is per second, and the resolution tier determines the per-second price. Tasks are asynchronous: creation returns a `task_id`, then use [Retrieve video task](https://docs.dimilinks.com/en/api-reference/videos/tasks.md) to poll status and download the result.

> [!NOTE]
>
> Video billing is per second: cost = per-second price × duration. Check `seconds` and `resolution` against the limits below before submitting, and see [Video groups and billing rules](https://docs.dimilinks.com/en/api-reference/videos/groups.md) for how the two video groups differ.

## Endpoint

```http theme={null}
POST https://api-direct.dimilinks.com/v1/videos
Content-Type: application/json
Authorization: Bearer <DIMILINKS_API_KEY>
```

## Text-to-video example

```bash theme={null}
curl "https://api-direct.dimilinks.com/v1/videos" \
  -H "Authorization: Bearer $DIMILINKS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3",
    "prompt": "A futuristic city in the rain, a black sports car crossing neon streets, low-angle tracking shot, cinematic lighting",
    "seconds": 6,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "generate_audio": true
  }'
```

Success response:

```json theme={null}
{
  "id": "task_xxx",
  "task_id": "task_xxx",
  "object": "video",
  "model": "minimax-h3",
  "status": "queued",
  "progress": 0
}
```

## Parameters

| Parameter | Type | Required | Description |
| - | - | - | - |
| `model` | string | yes | Always `minimax-h3`. |
| `prompt` | string | yes | Video description; describe how each reference is used when references are provided. |
| `seconds` | integer | no | Duration 4-15 seconds, default 4. |
| `resolution` | string | no | `480p` / `720p` / `1080p`, default `480p`; determines the per-second price. |
| `aspect_ratio` | string | no | `16:9` / `9:16`, default `16:9`. |
| `generate_audio` | boolean | no | Generate or keep audio. |
| `references` | array | no | Reference media array, see below. |

## Pricing

| Resolution | Per-second price (¥) |
| - | -: |
| 480p | 0.16 |
| 720p | 0.24 |
| 1080p | 0.70 |

Cost = per-second price × `seconds`. Prices may change; use `GET /api/pricing` as the runtime source of truth.

## Reference media

Put media into the `references` array. Each `source` must be a public HTTPS URL that the server can fetch without login or cookies:

```json theme={null}
{
  "references": [
    { "type": "image", "role": "reference", "source": "https://cdn.example.com/character.png" },
    { "type": "video", "role": "reference", "source": "https://cdn.example.com/motion.mp4" },
    { "type": "audio", "role": "reference", "source": "https://cdn.example.com/dialogue.mp3" }
  ]
}
```

Limits:

* Up to 9 images; up to 3 videos and 3 audios, with videos + audios combined ≤ 3; total media ≤ 12.
* Every `source` must be unique.
* Use only `reference` roles or omit `role`; this model does not support `first_frame` / `last_frame`.
* Supported formats: images `image/jpeg`, `image/png`, `image/webp`; videos `video/mp4`, `video/quicktime`; audio `audio/mpeg`, `audio/wav`, `audio/x-wav`.

## Errors

* `400`: invalid parameters or media; fix the request instead of retrying unchanged.
* `402`: insufficient balance or generation quota.
* `429`: capacity saturated; back off and retry.
* See [Errors](https://docs.dimilinks.com/en/api-reference/errors.md) for common error codes.

## Next step

Save the `task_id` from the creation response, then poll with [Retrieve video task](https://docs.dimilinks.com/en/api-reference/videos/tasks.md) until `completed` before downloading.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.