Reference Assets

Beyond images, MiniMax-H3 also accepts audio and video as reference assets, used to drive pacing or transfer style.

Audio Reference

Drive the visuals with audio — well suited to music, narration, or rhythm-based videos.

curl -X POST "http://103.36.63.156:3000/v1/videos" \
  -H "Authorization: Bearer sk-xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax_h3",
    "prompt": "Generate an abstract particle-flow animation synced to the audio rhythm",
    "audio_url": "https://example.com/bgm.mp3",
    "duration": 10
  }'

Video Reference

Use an existing video as reference and transfer its motion or style.

Video Reference Only

curl -X POST "http://103.36.63.156:3000/v1/videos" \
  -H "Authorization: Bearer sk-xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax_h3",
    "prompt": "Following the camera movement of the reference video, generate a city street scene",
    "reference_video": "https://example.com/ref.mp4",
    "duration": 6
  }'

Video + Audio Track Reference

Reuse the reference video's audio track at the same time:

{
  "model": "minimax_h3",
  "prompt": "Keep the rhythm of the reference video, replacing the subject with a robot",
  "reference_video": "https://example.com/ref.mp4",
  "reference_video_audio": true,
  "duration": 6
}

Parameter Reference

Parameter	Type	Description
`audio_url`	string	Audio reference URL
`reference_video`	string	Video reference URL
`reference_video_audio`	boolean	Whether to reuse the reference video's audio track

Reference Asset Limits

Type	Limit	Supported Formats
Audio	1	mp3 / wav / m4a
Video	1	mp4 / mov

Reference assets and the prompt should be semantically consistent. The closer the camera movement, pacing, and reference assets described in the prompt, the more stable the result.