MiniMax H3

MiniMax H3 (Hailuo 03), MiniMax’s video model: clips with native sound at 768p and 2K, 4 to 15 seconds, from a prompt and up to 9 reference images, 3 reference videos and 3 reference audio clips.

Prices

A clip costs its length times the rate: a 5 second 768p clip is $0.325. It’s charged only when the video is delivered.

What it takes

• A prompt of up to 7,000 characters.
• Up to 9 reference_images, 3 reference_videos and 3 reference_audio clips (MP3 or WAV), in any mix, at no extra charge. Reference videos run up to 15 seconds in all, and so do audio clips. It takes no first_frame or last_frame: give the image as a reference and say in the prompt how the video opens.
• Lengths from 4 to 15 seconds, in whole seconds.
• 16:9, 9:16, 1:1, 4:3, 3:4 and 21:9. Set the shape with aspect_ratio; it’s 16:9 if you leave it out, whatever the reference images.
• Sound is always on.
• 768p and 2K.

Example

{
  "model": "minimax-h3",
  "reference_images": ["https://example.com/presenter.jpg"],
  "reference_audio": ["https://example.com/voice.mp3"],
  "prompt": "The woman from the image holds up a serum bottle and says, in the voice from the audio: \"Two weeks with this serum and my skin is glowing.\" Handheld phone video.",
  "seconds": 8,
  "resolution": "768p",
  "aspect_ratio": "9:16"
}

Prompting

• Say what each reference is for: “the woman from the image”, “the motion of the video”.
• To make someone speak, write the words in the prompt, in quotes. A voice clip in reference_audio sets how they sound; without the words, the speech comes out as made-up phrases.
• To use a music clip as the soundtrack, say so: “use the reference audio itself as the whole soundtrack, instrumental only, no singing, no text on screen”. Without that, the model may write its own song, lyrics and subtitles included.
• A reference video lends its motion and camera work.

Good to know

• Clips usually take two to five minutes; references and 2K take longest.
• Reference images, videos and audio are included in the price.
• Reference videos (MP4 or MOV) and audio clips (MP3 or WAV) each run 2 to 15 seconds. There’s no negative prompt or seed, and sound is always on.
• Prompts or images that may break the content rules can be declined (content_policy); declined videos are free.
• Videos are kept for 30 days; download the ones you want to keep.

Full API details: Video generation. Try it in the dashboard’s Playground, which shows the code for each video.