MiniMax Format (MiniMax-H3)

MiniMax-H3 (the MiniMax video model on sale on this platform; for the exact model name see the Model Plaza) reuses the generic task-format endpoints and supports all three modes: text-to-video / image-to-video / video-to-video (for the general flow see Create Video Generation Task). The query flow is the same as Get Video Generation Task Status.

Endpoint

Create and query reuse the generic task endpoints: POST /v1/video/generations and GET /v1/video/generations/{task_id}.

Request Body

For the generic fields (prompt, metadata merge rules, etc.) see Create Video Generation Task; the table below lists what MiniMax-H3 requires and accepts, verified by testing:

ParameterLocationTypeDescription
modelTopstringMiniMax-H3, copied from the Model Plaza
promptTopstringRequired. Scene description for text-to-video; target motion and style for image/video-to-video
durationTopintegerNo. Duration in seconds; upstream validates an integer from 4 to 15, defaults to 5
imagesTopstring[]Required for image-to-video. Reference image (first frame) URL, see "Asset URL constraints" below
metadata.resolutionmetadatastringRequired. Resolution tier: 768p / 2K. ⚠️ 720p / 1080p are not supported; omitting it fails validation against 720p
metadata.ratiometadatastringRequired for text-to-video (error: ratio is required for text-to-video and cannot be adaptive); 16:9 verified. Optional for image / video-to-video
metadata.video_urlsmetadatastring[]Required for video-to-video. Direct mp4 link(s) of the source video, see "Asset URL constraints" below
NOTE

All these validations return synchronously at submit time and incur no charge — fix the parameters per the error and retry.

Asset URL constraints (image / video-to-video)

Reference images and source videos must be publicly reachable direct links; prefer your own site or a mainland CDN / OSS. Two common failure modes: overseas image hosts blocked by upstream (media url unreachable, error code 2013) and dead URLs (media not found (HTTP 404)). Failed tasks are not charged; fail_reason carries the upstream error.

Request Example

cURL
Python
Node.js
# Scenario 1: text-to-video (resolution + ratio required)
curl -X POST https://api-model.shuntianchuang.cn/v1/video/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-your-token" \
  -d '{
    "model": "MiniMax-H3",
    "prompt": "一只橘猫在草地上追蝴蝶,电影感镜头,缓慢推近",
    "duration": 5,
    "metadata": { "resolution": "768p", "ratio": "16:9" }
  }'
# Scenario 2: image-to-video (images carries the first frame; ratio optional)
curl -X POST https://api-model.shuntianchuang.cn/v1/video/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-your-token" \
  -d '{
    "model": "MiniMax-H3",
    "prompt": "让画面中的主体缓缓运动起来,镜头轻微推进,电影感光影",
    "images": ["https://example.com/first-frame.jpg"],
    "metadata": { "resolution": "768p" }
  }'
# Scenario 3: video-to-video (metadata.video_urls carries the source video)
curl -X POST https://api-model.shuntianchuang.cn/v1/video/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-your-token" \
  -d '{
    "model": "MiniMax-H3",
    "prompt": "把这段视频转换为水彩画风格,保持原有动作",
    "metadata": {
      "resolution": "768p",
      "video_urls": ["https://example.com/source.mp4"]
    }
  }'

Response Body

On success the submit returns the unified task object; keep task_id for querying.

Response Example

{
  "id": "task_xxxxxxxx",
  "task_id": "task_xxxxxxxx",
  "object": "video.generation",
  "model": "MiniMax-H3",
  "status": "queued",
  "progress": 0,
  "created_at": 1790651214
}

Query & Download

Query via the unified task endpoint GET /v1/video/generations/{task_id}; for the response structure and polling examples see Get Video Generation Task Status. Channel specifics:

  • Result: on SUCCESS, data.result_url is this platform's proxied download link; data.data.output.video_url is the upstream OSS signed direct link which expires in about 24 hours — save your output promptly
  • Mirror fields: data.data echoes the upstream task — size ("768P"), seconds ("5"), numeric progress, completed_at
  • Failure: FAILURE with the upstream error in data.fail_reason; failed tasks are not charged

Response example (success terminal state, abridged):

{
  "code": "success",
  "message": "",
  "data": {
    "task_id": "task_xxxxxxxx",
    "status": "SUCCESS",
    "progress": "100%",
    "result_url": "https://api-model.shuntianchuang.cn/v1/videos/task_xxxxxxxx/content",
    "fail_reason": "",
    "data": {
      "id": "task_yyyyyyyy",
      "size": "768P",
      "error": null,
      "model": "MiniMax-H3",
      "object": "video.generation",
      "output": {
        "video_url": "https://…aliyuncs.com/…/output_aigc.mp4?Expires=…"
      },
      "status": "completed",
      "seconds": "5",
      "progress": 100,
      "created_at": 1790651213,
      "completed_at": 1790651314
    }
  }
}
WARNING

The 2K tier is taken from the upstream error message and was not verified with an actual generation; pricing for higher resolutions / durations is subject to the Model Plaza.