L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Données et recherche · SKILL.md

minimax-video

minimax-video

Generate MiniMax H3 videos from text and optional image, video, or audio references through AceDataCloud. Use for text-to-video, first/last-frame video, multimodal reference video, and MiniMax H3 task polling.

Independent GitHub snapshot · Automated classification · Original documentation governs

Aacedatacloud / skillsApache-2.0

Documentation de la Skill

GitHub

name: minimax-video description: Generate MiniMax H3 videos from text and optional image, video, or audio references through AceDataCloud. Use for text-to-video, first/last-frame video, multimodal reference video, and MiniMax H3 task polling. license: Apache-2.0 metadata: author: acedatacloud version: "1.0" compatibility: Requires ACEDATACLOUD_API_TOKEN in .env (see _shared/authentication.md). Optionally pair with mcp-minimax for tool-use.

MiniMax H3 Video Generation

Generate 4–15 second videos through POST https://api.acedata.cloud/minimax/videos. Use the V2 multimodal content array to supply the prompt and optional reference media.

Setup: See authentication. The HTTP API waits for the final result by default. Agents and MCP clients should send "async": true, save the returned task_id, then use async task polling with POST /minimax/tasks. MiniMax MCP generation tools expose async and default it to true.

Contract

ParameterValuesDefault
modelMiniMax-H3required
contentarray of prompt and optional media itemsrequired
resolution768P, 2Krequired
durationinteger 4–15required
ratioadaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16omitted
asynctrue, falsefalse at the HTTP API; true in MiniMax MCP tools
callback_urlpublic HTTP(S) webhook; also enables async modeomitted

Each content item has a type of text, image_url, video_url, or audio_url; set the matching field to the text or public URL. Media items use a role:

RoleUse
first_frameStarting image
last_frameEnding image
reference_imageReference image
reference_videoReference video
reference_audioReference audio

Include a text item with the generation prompt in every request.

Text to video

curl -X POST https://api.acedata.cloud/minimax/videos \
  -H "Authorization: Bearer $ACEDATACLOUD_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-H3",
    "content": [
      {
        "type": "text",
        "text": "A red fox running through a snowy forest at dawn, low tracking shot"
      }
    ],
    "resolution": "768P",
    "ratio": "16:9",
    "duration": 4,
    "async": true
  }'

First and last frame video

{
  "model": "MiniMax-H3",
  "content": [
    {
      "type": "text",
      "text": "A camera slowly pushes in as the character walks into the sunrise."
    },
    {
      "type": "image_url",
      "image_url": {"url": "https://cdn.acedata.cloud/first-frame.png"},
      "role": "first_frame"
    },
    {
      "type": "image_url",
      "image_url": {"url": "https://cdn.acedata.cloud/last-frame.png"},
      "role": "last_frame"
    }
  ],
  "resolution": "768P",
  "ratio": "9:16",
  "duration": 8,
  "async": true
}

Reference-guided video

{
  "model": "MiniMax-H3",
  "content": [
    {
      "type": "text",
      "text": "A dancer moves naturally to the rhythm."
    },
    {
      "type": "image_url",
      "image_url": {"url": "https://cdn.acedata.cloud/reference.png"},
      "role": "reference_image"
    },
    {
      "type": "audio_url",
      "audio_url": {"url": "https://cdn.acedata.cloud/reference.wav"},
      "role": "reference_audio"
    }
  ],
  "resolution": "768P",
  "ratio": "9:16",
  "duration": 8,
  "async": true
}

Poll a task

curl -X POST https://api.acedata.cloud/minimax/tasks \
  -H "Authorization: Bearer $ACEDATACLOUD_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"action":"retrieve","id":"TASK_ID"}'

Continue polling about every five seconds while task.status is queued or running. Only task.status="succeeded" is success; then read the final video URL from task.content.url. If the task is failed or cancelled, stop polling and report task.error.code / task.error.message when present. A returned task_id or HTTP 200 means acceptance, not a finished video; never report a URL as final before success.

Use retrieve_batch with ids to check several tasks; inspect each item in items independently. Use delete with id to remove a task record, not to cancel generation. Batch listing also accepts limit, offset, created_at_min, and created_at_max.

MCP: pip install mcp-minimax | Hosted: https://minimax.mcp.acedata.cloud/mcp | See MCP setup. Current generation supports MiniMax-H3 with 768P or 2K; keep querying the returned task ID until terminal success or failure.

Gotchas

  • Do not send action to /minimax/videos; the API infers the mode from media inputs.
  • duration must be an integer, not a decimal.
  • Put the prompt in a content item with type: "text"; do not send a top-level prompt.
  • Use first_frame, last_frame, and reference roles explicitly; do not rely on item order to determine an image's role.
  • Use public URLs that the generation service can download for every media item.
  • Returned videos are served from AceDataCloud CDN.
Skills enregistrées