MiniMax · Intelligence, with everyone
One studio for every sense. Direct films with Hailuo, design voices with Speech 2.8, score them with Music 3.0, build worlds on the open M3 model, and hand the boring parts to Agents.
The studio
Five instruments, one stage. Each is a flagship model with a working surface built for people who make things — not a demo behind a waitlist.
Hailuo · Video
Text, image, or a first-and-last frame in — a shot out. The H3 omni-modal engine holds character, light, and camera intent across a sequence, not just a lucky clip.
Speech 2.8 · Voice
Design a voice from ten seconds of audio — then direct it. Breath, pace, whisper, laughter. Speech 2.8 performs in over forty languages without flattening the accent.
Music 3.0 · Sound
Complete songs — arrangement, vocals, mix — written around your scene. Trade drafts like notes, then pull the stems into your own session when it lands.
MiniMax M3 · Open models
A frontier coding and agentic model on a novel MSA attention architecture. One million tokens of context — your whole script, codebase, or lore bible fits in one thought.
MiniMax Agent & Code · Autonomy
Spin up a team of agents that research, storyboard, render, and revise while you sleep. MiniMax Code is the harness that knows your habits — skills, memories, schedules — so the crew already works the way you do.
For creators
Feed Hailuo your shot list and a moodboard still. It holds your lead's face across forty takes, keeps the rain out of the close-up unless you asked for it, and hands your editor clean 1080p plates. One short-film team cut previz from three weeks to a weekend.
Hum a hook into Music 3.0 and argue with the arrangement like it's in the room. Verse too polite? Say so. When it finally snarls, export the stems and finish it in Ableton — your name on every track.
Design your instrument once — ten seconds is enough — then let Speech 2.8 carry the audiobook, the game barks, the trailer voice. You keep the rights and the royalties; it keeps the 3 a.m. sessions.
M3 reads your entire codebase in one pass — a million tokens deep — and MiniMax Code runs the grind work: tests, ports, the settings menu nobody wants to build. You stay on the fun 20%.
Your conlang grammar, four centuries of history, the map, the cast — M3 holds the whole bible at once and answers in your world's voice. Hailuo and Speech make the trailer and the tavern songs.
For developers
One API key opens the whole studio. REST when you want simple, SDKs when you want typed, webhooks when renders finish. Open weights when you want the whole thing on your metal.
Renders a 6-second 1080p shot. Webhook fires when the take is ready.
Community
Every day the feed fills with things no studio would have greenlit. That is the point.
“I pitched my short to three festivals with a Hailuo teaser. Two said yes before they saw a single human on camera.”
“Speech 2.8 reads my stage directions like a patient engineer. The audiobook shipped in eleven days — I narrated all of it, most of it asleep.”
“We pointed M3 at 400k lines of legacy game code. It found the desync bug our team had been romanticizing about for a year.”
Fresh from the lab
Start
Free tier, no waitlist, your first render in under a minute. The studio is already set up — it was missing one thing.