One prompt, every medium. MiniMax H3 is the open general-purpose video model that reads text, images, video clips, and audio as a single context: borrow a camera move from one video, cast a character from one photo, match a vocal from one audio take — and it composes the shot with native stereo sound at up to 2K. 4 to 15 seconds, 100 credits.
H3 treats every reference as a dial you set — the workflow:
Combine images for identity, video clips for motion and camera language, and audio for voice or rhythm — up to 12 files across types in one generation, mixed freely.
Write the relationship in words: 'use the camera move from Video 1, the character from Image 2, the vocal from Audio 3.' H3's context engine handles the cross-modal wiring itself.
Skip the references and go positional: give the first frame, the last frame, or both, and H3 renders the transition between them with sound attached.
Output lands at 768P or 2K — the 2K tier is H3's default sweet spot, with dialogue, ambience, and score generated natively in the same pass. 4 to 15 seconds per clip.
The capabilities that make it the指挥台 of multimodal direction on Oddbeam:
Text, images, video, and audio enter the same creative context instead of separate tool silos — the model reasons over all of them at once, which is what makes 'borrow that camera move, sing like this take' a single instruction.
H3 outputs 2K as its standard tier with native 32 kHz stereo audio generated alongside the picture — resolution and sound are the baseline, not an upsold add-on.
Stable native dialogue across Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish — multilingual voiceover ships inside the render.
V2V motion transfer carries a performance from a reference clip onto a new subject — the dance, the gesture, the fight move — without rotoscoping or hand-keying the animation.
H3's context engine parses fuzzy, conversational direction and fills sensible gaps itself — complex multimodal instructions survive contact with short 4-6 second clips better than keyword-stuffed prompts.
Teams whose references never fit in one box.
Dropping the demo vocal in as a reference and getting performance and lip-sync matched to it — that alone moved two post steps off my schedule.
Music Video Director
Post Production
I feed the character sheet, a camera-move clip, and a sound mood in one H3 run. The context does the assembly that used to take three different tools.
Game Cinematics Artist
Studio Team
2K default output is the quiet win — product films come out sharp enough for detail-page zoom without an upscale pass, at 100 credits a run.
Ecommerce Video Lead
Marketplace Brand
Dropping the demo vocal in as a reference and getting performance and lip-sync matched to it — that alone moved two post steps off my schedule.
Music Video Director
Post Production
I feed the character sheet, a camera-move clip, and a sound mood in one H3 run. The context does the assembly that used to take three different tools.
Game Cinematics Artist
Studio Team
2K default output is the quiet win — product films come out sharp enough for detail-page zoom without an upscale pass, at 100 credits a run.
Ecommerce Video Lead
Marketplace Brand
Cancel anytime • Unused credits roll over
Perfect for getting started
Great for power users and teams
First choice for professionals
Feed H3 your references and direct the whole ensemble — 4-15s at up to 2K stereo, from 100 credits.