Skip to content
Luma AI Help Center home
InboxAsk a human

MiniMax H3: FAQ

Updated by Martin

1. What is MiniMax H3?

MiniMax H3 generates video with a native audio track. It runs in two modes: a Keyframe mode for text-to-video and image-to-video, and a Reference mode (MiniMax calls it Omni Reference) for multimodal generation from images, videos, and audio. It follows long, structured, shot-by-shot prompts well.

2. What resolution and duration does it support?

Resolutions:

  • 768P (lower cost)

  • 2K (1440p, maximum detail)

Durations (clip lengths):

  • Any whole number of seconds from 4s to 15s, at 24fps (longer clips cost more)

3. How does MiniMax H3 handle audio?

Every output includes a native stereo audio track, and no parameter disables it. Describe the sound you want (ambience, dialogue, score, timed music cues) directly in the prompt. Separately, reference audio in Reference mode lets you condition on audio you supply, including voice transfer or cloning.

4. What aspect ratios are available?

  • 16:9 (landscape, default)

  • 9:16 (portrait / vertical)

  • 1:1 (square)

  • 4:3, 3:4, 21:9

  • adaptive (reference mode only; text-to-video must name a concrete ratio, and keyframe mode follows the supplied image's ratio)

5. Can you combine inputs in one prompt?

Yes. The two modes are mutually exclusive:

Keyframe mode:

  • A Start Frame with an optional End Frame for interpolation.

Reference mode (Omni Reference, multimodal):

  • Up to 9 reference images

  • Up to 3 reference videos (each 2 to 15s; combined 15s or less)

  • Up to 3 reference audio clips (each 2 to 15s; cannot be used alone, must pair with a reference image or video)

  • 12 reference files maximum across all three slots combined

Address inputs in the prompt as prose ordinals within their modality ("Image 1", "Image 2", "Video 1", "Audio 1"), matching slot order. It does not parse @-mentions or bracket tokens. A reference video also enables editing workflows: replace or add a subject, swap wardrobe or products, relight, replace a background, or match another clip's motion and pacing.

6. Are keyframes available?

Yes, in Keyframe mode:

  • Start Frame: the model animates outward from this image, and the image's aspect ratio overrides the ratio picker.

  • End Frame: optional, requires a Start Frame, and drives interpolation between the two.

At a glance

Aspect

Options

Type

Video (native audio)

Modes

Keyframe, Reference (Omni Reference)

Resolutions

768P, 2K

Duration

4s to 15s, 24fps

Aspect ratios

16:9, 9:16, 1:1, 4:3, 3:4, 21:9, adaptive

Reference images

Up to 9

Reference videos

Up to 3 (combined 15s or less)

Reference audio

Up to 3 (must pair with image/video)

Reference files total

12 max across all slots

Audio

Native stereo, always on

Prompt length

Up to 7000 characters (English and Chinese)