MiniMax-H3 API

Use the exact model ID MiniMax-H3. The model field must match this page exactly.

Sample outputMiniMax-H3POST /v1/videos5.2s

Parameters: duration=5resolution=2Kratio=16:9

Prompt: A fleet of hot-air balloons drifts over mist-filled mountain valleys at dawn. Aerial camera glides slowly between the balloons as golden light breaks over the ridges and fog rolls through the pines below. Calm, majestic, photoreal landscape, steady motion, no text.

Request that produced this sample

{
  "model": "MiniMax-H3",
  "prompt": "A fleet of hot-air balloons drifts over mist-filled mountain valleys at dawn. Aerial camera glides slowly between the balloons as golden light breaks over the ridges and fog rolls through the pines below. Calm, majestic, photoreal landscape, steady motion, no text.",
  "seconds": "5",
  "duration": 5,
  "resolution": "2K",
  "ratio": "16:9",
  "aspect_ratio": "16:9"
}

Authentication and endpoint

Use a Modelflare API key with https://modelflare.dev/v1. The gateway entry point is POST /v1/videos.

HTTP

Authorization: Bearer YOUR_MODELFLARE_API_KEY

Create a task

bash

curl https://modelflare.dev/v1/videos \
  -H "Authorization: Bearer YOUR_MODELFLARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-H3",
    "content": [{"type": "text", "text": "A white cat runs through a neon-lit street in the rain, cinematic tracking shot"}],
    "resolution": "2K", "duration": 5, "ratio": "16:9"
  }'

Poll and download

Save the task ID. Poll GET /v1/videos/{task_id} and download with GET /v1/videos/{task_id}/content. Result URLs may expire, so transfer the file promptly.

bash

curl https://modelflare.dev/v1/videos/TASK_ID \
  -H "Authorization: Bearer YOUR_MODELFLARE_API_KEY"

Request notes

MiniMax-H3 uses content[] with resolution, duration, and ratio controlling output.

Request parameters

MiniMax-H3 uses a multimodal content[] array. Poll the MiniMax query path after task creation; first_frame / last_frame and reference_* roles are distinct workflows.

Field	Required	Notes
model	Yes	MiniMax-H3
content[]	Yes	Exactly one text item; may include first_frame, last_frame, reference_image, reference_video, or reference_audio
resolution	Yes	768P / 2K
duration	Yes	4–15 seconds
ratio	Conditional	21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16 / adaptive
content[].role	Conditional	Use first_frame / last_frame for image-to-video and reference_* for reference workflows; do not mix the groups
content[].url	No	Public HTTP/HTTPS URL required; Base64 media is not supported
aigc_watermark	No	Whether to add a watermark
callback_url	No	Task status callback URL

Input and output limits

Output is 768P or 2K for 4–15 seconds. Text-only requests need a concrete ratio; image-to-video allows one first frame and one last frame, while reference workflows require at least one reference image or video and cannot use audio alone. Do not mix the two role groups.

Documented workflow examples

MiniMax-H3 media URLs must be publicly accessible and cannot use Base64; first/end-frame and reference workflows must not be mixed.

bash

curl https://modelflare.dev/v1/videos \
  -H "Authorization: Bearer YOUR_MODELFLARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-H3",
    "content": [
      {"type": "text", "text": "让图像主体沿视频动作自然移动"},
      {"type": "reference_image", "role": "reference_image", "url": "https://example.com/subject.png"},
      {"type": "reference_video", "role": "reference_video", "url": "https://example.com/motion.mp4"}
    ],
    "resolution": "2K", "duration": 5, "ratio": "16:9"
  }'

Errors and failed tasks

Invalid model parameters return a 4xx with a request ID. An asynchronous task can also enter failed after creation succeeds; use error_code and error_message from the poll response as the final task reason. Do not retry the same request with a different JSON shape.

Supported scope

This page covers Modelflare model generation and editing calls. Search or map enrichment, voice management, and element-library management are resource APIs outside this public model-call surface and are not required request fields.

Pricing and billing

This model uses per-second video billing. See the model pricing page for the current customer price, reference list price, and complete specification.

Note

This page documents Modelflare’s public interface; the fields and limits shown here are the current contract.

Variants

Model ID	Difference
MiniMax-H3	The model documented on this page.