MiniMaxVIDEOGuest engineNative audio

MiniMax H3 on VisionX.

Up to 2K, refined from a native 768p render, in one take of up to fifteen seconds with sound.

1model on rosterfrom 7 VXper shotup to 9reference imagesCast-readyidentity locked
01

What MiniMax H3 is on the roster

THE MODEL

MiniMax H3 renders natively at 768p and delivers up to 2K by refining that render in a second pass, so a keeper leaves as a 2K master without a separate finishing job. Its 720p tier is the native 768p render itself, for drafts.

It is also unusually flexible about what it will take direction from. A shot can open on a frame you supply and close on another, or it can be built from references — up to nine images for subject and wardrobe, three clips for camera rhythm, and voice tracks for delivery — with a generated audio track on the way out.

2K from a native 768p render

The 2K tier refines the native 768p render to 2K in a second pass. Block the shot at 720p, then run the keeper at 2K.

Fifteen seconds in one take

Any length from four to fifteen seconds, as a single continuous shot, so a beat that needs room does not have to be stitched from two renders.

Open and close on your frames

A first frame, a last frame, or both. The output inherits their geometry, which makes it the straightforward way to bridge two stills you already have.

References for subject, motion and voice

Nine reference images, three reference clips and three voice tracks, addressed positionally in the prompt — the character from Image 1 moving to the camera rhythm of Video 1.

Widescreen included

H3 renders true 21:9 as a first-class ratio rather than cropping a 16:9 frame down to it.

02

What MiniMax H3 actually does

CAPABILITIES

A 2K tier, one take up to fifteen seconds, and the references that brief it.

2K from a native 768p render

H3 renders natively at 768p, which is the 720p tier for drafts. The 2K tier refines that render to 2K in a second pass, so a keeper leaves as a 2K master without a separate finishing job.

2K

Any length from four to fifteen seconds

One continuous take, set to the second, so a beat gets the room it needs instead of being stitched from two renders.

Single take

Open and close on your frames

A first frame, a last frame, or both. The output inherits their geometry, which makes H3 the straightforward way to bridge two stills you already have.

First and last frame

Nine stills, three clips, three tracks

Up to nine reference images, three reference clips and three audio tracks, twelve references at most in one shot, each addressed by number in the prompt.

Multimodal reference

Sound on every take

H3 generates an audio track with the picture on every render, and on VisionX there is no silent mode, so plan to replace the track in the edit when the cut is scored.

Native audio

True 21:9

Widescreen 21:9 is a first-class ratio alongside 16:9, 4:3, 1:1, 3:4 and 9:16, rather than a 16:9 frame cropped down to it.

Aspect
03

Every variant, exactly as the studio runs it

SPEC SHEET

These rows are read live from the engine registry and the pricing engine. What you see here is what the composer quotes.

ModelModesResolutionsRatiosDurationsVX
MiniMax-H3MiniMax H3T2V · I2V · V2V720p · 2K · 4K21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:164 to 15s7 VX · 5s T2V at 720p

Self-serve rates run as low as $0.085 per VX at scale. The composer quotes the exact VX for your shot, references included, before you run it.

04

How to write for MiniMax H3

PROMPTING

H3 reads references by type and number, so the prompt’s job is to say what each one is for. Everything else is ordinary shot description.

PRODUCT FILM · 2K · 10S · 21:9
Image 1 is the watch. Image 2 is the wrist and sleeve.
Video 1 carries the camera move: a slow orbit, then a push.

The watch from Image 1 on the wrist from Image 2, turning into
low raking light. The move follows Video 1 and settles on the dial.

Address references by number

“Image 1”, “Video 1”, “Audio 1”, in the order you attached them. Naming what each one contributes beats “in the style of”.

Audio needs a picture beside it

A voice or audio reference cannot brief a shot on its own. Attach at least one image or clip with it.

Save 2K for the keepers

Block the shot on the 720p tier until the staging is right, then re-run the approved prompt at 2K.

Keep the brief under 7,000 characters

That is H3’s own prompt ceiling, and the composer counts against it as you write.

05

What teams route to MiniMax H3

USE CASES

The board lines that go to H3 first.

Deliverables that ship at 2K

Product close-ups, packaging and title cards that leave the render as a 2K master.

Bridging two approved stills

A first and a last frame from signed-off key art, with the move generated between them.

A shot with a voice brief

The Cast member from an image, the delivery from a voice track, the camera rhythm from a clip.

Cinematic widescreen

Title cards and brand films framed at 21:9 from the first render.

FAQ

MiniMax H3, answered

Is MiniMax H3’s 2K native?
No. H3 renders natively at 768p, which is the 720p tier. The 2K tier is that render refined to 2K in a second pass, not generated at 2K from the start.
How long can an H3 shot be?
Any whole number of seconds from four to fifteen, as one continuous take.
What references does H3 take?
Up to nine reference images, three reference clips totalling fifteen seconds, and three audio tracks totalling fifteen seconds, with twelve references at most in one shot. A first frame, a last frame or both can anchor the shot instead. An audio reference always needs an image or clip alongside it.
Can I switch H3’s audio off?
No. H3 renders an audio track on every take, so the composer shows audio as always on for it. Replace the track in the edit when the cut is scored.

Try MiniMax H3 on your Cast.

20 VX free on signup, no card required.