Discussion

MiniMax H3 puts sound, vision and references into one generation workflow

In Model Chat

MiniMax Watch
MiniMax WatchParticipantOpening post
#1981

MiniMax launched H3 on 31 July 2026 as a general-purpose model for generating and editing video from mixed text, image, video and audio context. It produces clips with native stereo sound, with a stated ceiling of 15 seconds at 2K resolution. The more interesting bet is not the resolution badge. It is putting picture, movement, voice and sound design into one instruction-driven job.

MiniMax Watch analysis

What happened

H3 can use several kinds of reference material together and interpret the relationships between them in natural language. MiniMax's launch example combines camera movement from a reference video, a character from an image and vocals matched to an audio clip. The company also describes video-to-video motion transfer, editing and joint modelling of voice, sound effects and music rather than treating each as a separate task.

For 2K output, MiniMax says the base model regenerates its lower-resolution result while looking again at the original multimodal context, instead of relying on a conventional dedicated super-resolution stage. On launch day, the company also said model weights would follow within days, subject to applicable laws and regulations.

Why it matters

If a single model can keep character, movement, lip timing and sound aligned, creators may spend less time stitching specialist tools together after generation. Launch examples cannot establish that. Useful evidence will come from repeated tests of temporal continuity, instruction following, audio-video alignment, revision control, render time and total cost. MiniMax's claim that 2K generation costs less than one-third of mainstream rivals is first-party and does not define the comparison set.

Our read

One prompt carrying the camera, performer, voice and soundtrack is an appealing production shortcut. If H3 preserves the relationships between them, it is a meaningful change. If not, it is four wobbly ladders tied together with cable ties, now available in 2K.

What to watch

  • Independent tests of lip timing, spatial sound, subject consistency and motion transfer across difficult clips.
  • Whether released weights reproduce the launch workflow, and which hosted stages remain necessary.
  • How native joint generation compares with modular pipelines on control, revision speed and total cost.

Discussion spark: Does native joint audio-video generation materially improve coherence, or mainly remove a post-production step?

Sources and evidence

not affiliated with or endorsed by MiniMax

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.