Back to featured models

MiniMax H3

One model for text, images, video, and audio—producing 2K video with synchronized stereo sound for up to 15 seconds.

An open-weights, omni-modal video model.

Try MiniMax H3 in OneInfer Studio

Describe a scene or start from an image, and get a video. No code or API key; sign in with your OneInfer account.

Open in Studio(opens in a new tab)
Resolution
2K
Duration
5–15 seconds
Frame rate
24 FPS
Aspect ratios
7
Audio
Native stereo
Prompts
Up to 7,000 characters

One model, every reference

Move from multimodal references to targeted edits, stable graphics, and audio composed to picture.

One model, every reference

Combine up to nine images, three video clips, and three audio tracks. MiniMax H3 can carry identity, camera motion, composition, sound design, and editing rhythm into one coherent output.

Reference to Video: 9 images, 3 video clips, 3 audio tracks

Change one thing, keep everything else

Replace objects, rewrite signs, shift lighting, change dialogue, or restyle a scene while preserving the composition and details you want to keep.

Precise editing: localized edits, relighting, object control

Text and interfaces that hold together

Create readable captions, credits, brand marks, interface motion, game HUDs, and kinetic typography with prompts up to 7,000 characters.

Typography & graphics: legible type, UI motion, 7,000 characters

Sound composed to picture

Generate stereo score, dialogue, foley, and ambience timed to the video. Reference audio can guide voice and performance across a scene.

Native audio: stereo audio, dialogue, voice reference

Copy a structured prompt and adapt it for your own generation.

Aurora over the ice fjord

First & last frame

Open on a still fjord at dusk. Hold a slow static frame as the last light fades, then let green and violet aurora fill the sky while gentle snow drifts through frame.

Desert courier

Reference to video

Use Image 1 as the fixed character reference. Preserve the silhouette and costume exactly while the camera tracks alongside at a low angle across a dune ridge at golden hour.

Studio backdrop swap

Video editing

Replace the plain studio backdrop in Video 1 with the rain-soaked alley from Video 2. Match the cool lighting and reflections so the composite reads as one shot.

Generate, reference, and edit

text-to-video

MiniMax H3 Text to Video

Generate video directly from a prompt at up to 2K resolution, with durations from 5 to 15 seconds and multiple aspect ratios.

View model details
image-to-video

MiniMax H3 Image to Video

Animate an opening image, or provide first and last frames to control the visual transition and composition.

View model details
reference-to-video

MiniMax H3 Reference to Video

Guide subjects, visual style, motion, and sound using a combined set of image, video, and audio references.

View model details

A few lines to first frame

const response = await fetch(
  "https://api.oneinfer.ai/v1/ula/generate-video",
  {
    method: "POST",
    headers: {
      "Authorization": "Bearer $ONEINFER_API_KEY",
      "Content-Type": "application/json"
    },
    body: JSON.stringify({
      model: "MiniMaxAI/MiniMax-H3",
      prompt: "A gateway of light opens above a quiet coastline",
      aspect_ratio: "16:9",
      resolution: "2K",
      duration: 10
    })
  }
);

Use your OneInfer API key from the console.

Read the docs

Common questions

MiniMax H3 is an open-weight, general-purpose omni-modal system. It understands text, images, video, and audio together and generates video from those combined references.