Unified multimodal context
One request can carry text alongside up to nine images, three videos of up to 15 seconds and three audio clips of up to 15 seconds, so the model reads the whole brief rather than a prompt string.
MiniMax (Open Weights)
MiniMax H3
Open-weights unified multimodal video model. Up to 15 seconds at 2K with native stereo audio, driven by a unified context that accepts text, images, video and audio in the same request. The weights are open, so it can also be run on your own hardware.
The base MiniMax H3 endpoints are not in the Studio catalog yet — these buttons open our text-to-video and image-to-video workflows, and the speed-optimised MiniMax H3 Max is available in the Studio today at /models/minimax-h3-max.
- Max duration
- 15s
- Max resolution
- 2K (1440px short edge)
- Audio
- Native stereo
- Aspect ratios
- 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, adaptive
- Weights
- Open
- Commercial use
- Allowed
Features
What MiniMax H3 does differently.
Precise localized editing
Swap a product, rewrite signage, relight a scene or add and remove an element while the rest of the frame stays where it was.
Typography, UI and graphics
Legible type and animated interface work hold up at 2K, and prompts run to 7000 characters when a shot needs that much direction.
Native stereo audio and voice transfer
Audio is generated in stereo in the same pass as the picture, and a supplied voice can be transferred or cloned onto the speaking character.
Open weights from day one
The weights are published alongside the hosted release, so the same model can be run on your own infrastructure.
Examples
Prompts that show its range.
Vintage binocular brand film, Images 1 to 4 as sequential keyframes, Wes Anderson 35mm look, red typographic accents
Try this promptWuxia character film, Image 2 locked reference (hanfu and silver crown), 4K xianxia production value
Try this promptLive-action to voxel: preserve buildings, transform only trees and cars to Minecraft blocks
Try this promptNeon laundromat encounter, live-action plus hand-drawn animation, phone-camera feel
Try this promptCapybara recreation: replace three suited men with photoreal capybaras, exact movement path
Try this promptGreen-screen fairytale composite: remove green screen, match background to characters' actions
Try this promptThese clips and prompts are published by MiniMax as reference for what H3 can do. Run one in the Studio to see your own render.
Workflows
Every way to run MiniMax H3.
- Text to videoPrompt to video, stereo audio in the same passOpen workflow
- Image to videoAnimate a still, or drive a shot from sequential keyframesOpen workflow
- Reference to videoCharacter, motion and audio references, up to nine imagesOpen workflow
MiniMax has not published a public rate for H3, so there is no per-second figure to quote. Credits come from one shared balance across image, video and speech, and the exact estimate is shown in the Studio before you generate.
Built for
- Brand films with precise product placement
- Character-driven narrative work such as wuxia and fantasy
- Live-action converted to a stylised look — voxel, anime or clay
- UI and UX animation and motion graphics
- Green-screen compositing
- Motion reference recreation
FAQ
MiniMax H3, answered.
Limits, audio, licensing and price — the things worth knowing before you spend a credit.
Direct a shot with more than a prompt.
Open the Studio to start in text-to-video, or run the speed-optimised MiniMax H3 Max at /models/minimax-h3-max today.