Minimax H3

MiniMax H3 is a general-purpose multimodal AI video generator. Instead of treating text-to-video, image-to-video, reference-based creation, and editing as four separate tools, H3 handles all of them inside one model that reads text, images, video clips, and audio together as a single creative context — and returns a finished scene with sound already in it.

What it does

Every generation ships with native two-channel stereo audio: ambience, sound effects, music, and dialogue synced to the on-screen action. Most AI video tools hand back a silent clip and leave the sound design to you. H3 treats audio as part of the shot, not a post-production step.

You can feed it up to 12 files in a single request — as many as 9 reference images, 3 video clips, and 3 audio tracks, alongside a text prompt of up to 7,000 characters. H3 reads the faces, the choreography, the camera language, and the voices across all of them at once, then merges them into one coherent take. Point at a character photo to lock identity across shots, reference a motion clip to borrow its choreography, drop in a voice sample to clone its tone. All in the same prompt, not three separate passes stitched together afterward.

Editing works the same way. Describe the change in plain language and H3 applies it while leaving the rest of the frame pixel-stable. Swap a character, replace a background, relight a scene from day to night, change an outfit, remove an object, or rewrite a line of dialogue in a cloned voice. Because unedited regions stay stable, you can iterate shot by shot the way a director gives notes — rather than regenerating from scratch and hoping the next roll lands closer.