New 2K multimodal video model

MiniMax H3 video generator

MiniMax H3 makes 4-15 second 2K clips with stereo sound. Start from a prompt, animate one image or a first/last-frame pair, or combine images, video and audio when identity, motion and sound all need separate references.

Create from text Open reference mode

Generated video and video input cost 36.5 credits per second. The first 5 input images are free; each additional image costs 11 credits. Audio input is free. The final total is rounded to the nearest whole credit.

Product and campaign shots that need 2K detail plus sound in one generation.

First/last-frame animation when both ends of a transition matter.

Character, motion or voice matching from several reference assets.

Model video

Input, sample and route in one launch clip

Use this as the quick model check before generating: accepted input, generated output and the clearest route in Video Studio.

Input Pick the route from the asset you already have.

Sample Watch generated clips before you spend credits.

Route Clear prompt, clear pricing, clear next step.

Generated image-to-video sample

UlazAI interface launch shot

Input: one UlazAI interface image, 4 seconds, native 2K output with stereo sound.

Prompt

Turn this AI video studio interface into a premium launch shot: layered glass panels move forward in depth, cyan and lime light travels across the controls, the camera makes a slow clean push-in, subtle interface motion, preserve all existing design, no new text, no new logo, native stereo interface ambience.

What H3 controls

H3 is most useful when a prompt alone cannot carry the job. Separate references can define the subject, camera motion and audio while the prompt explains how they work together.

Native 2K video and stereo sound

All three Studio routes return 2K video.
Dialogue, ambience, music and sound effects can be directed in the same prompt.
Use explicit timing notes when a visible action must land on a sound cue.

First and last frame control

One image sets the opening composition.
A second image locks the final frame and gives the transition a clear destination.
This works well for product reveals, interface changes and before/after shots.

Multimodal references

Reference mode accepts images for identity or style, video for motion and camera rhythm, and audio for voice or sound direction.
Name each asset's role in the prompt instead of asking every reference to influence everything.
Use the fewest references that clearly define identity, motion and sound.

Precise changes without rebuilding the shot

Reference mode can change a subject, background, lighting, wardrobe or dialogue while preserving the parts named in the prompt.
State both the requested change and what must remain untouched.
Short, scoped edits are more reliable than replacing several unrelated parts at once.

When H3 earns the higher render cost

Output	Text / image	With 4s video input
4 seconds	146 credits	292 credits
8 seconds	292 credits	438 credits
15 seconds	548 credits	694 credits

Good jobs for MiniMax H3

Use H3 when the output needs a connected audiovisual scene, not just movement added to a still frame.

Product films

Combine material close-ups, feature shots, readable interface details and synchronized sound inside one short sequence.

Character-led scenes

Use separate identity, motion and voice references to keep the same person recognizable while changing the performance.

UI and game concepts

Animate menus, controls and camera movement while telling the model which interface details must stay intact.

Localized campaign edits

Replace dialogue, voice or selected visual elements while preserving the original composition and timing.

Studio routes

Text to Video

minimax_h3_t2v

4 to 15 seconds

Prompt only. Choose 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16.

Open route

Image to Video

minimax_h3_i2v

4 to 15 seconds

Use one first frame, or add a second image to lock the last frame.

Open route

Reference to Video

minimax_h3_r2v

4 to 15 seconds

Combine up to 9 images, 3 videos and 3 optional audio references. Give every reference a clear role in the prompt.

Open route

MiniMax H3 or Seedance 2?

Pick H3 when native 2K, stereo sound and mixed image/video/audio references justify the higher price. Seedance 2 Mini is the cheaper route for draft volume; Seedance 2 remains an option when native 2K output is not required.

A prompt that keeps references under control

Choose text, image or reference mode from the assets you already have.
Describe the shot order, camera movement and audio cues in the order they should happen.
For every reference, state one job: identity, product design, motion, camera language, voice or sound.
List the face, materials, colors, typography or composition that must not change.
Start with 4 seconds. Increase the duration only when the shot needs more time to read.

FAQ

What does MiniMax H3 generate?

It generates 4-15 second 2K video with native stereo sound from text, image frames or mixed references.

Can I use a first and last frame?

Yes. Image-to-video accepts one first frame and an optional last frame.

Which references work in reference mode?

Up to 9 images, 3 video references and 3 MP3 or WAV audio files.

How much does it cost?

Generated video and video input cost 36.5 credits per second. The first 5 input images are free, each additional image costs 11 credits, and audio input is free.

Does H3 create sound?

Yes. The model can generate stereo dialogue, music, ambience and effects together with the video.

When should I use a cheaper model?

Use a cheaper model for draft volume or simple image animation where 2K, native sound and multimodal control do not change the result.

Create from text Browse Video Studio

Test MiniMax H3 with one reference mode at a time

Start with text-to-video, image-to-video or reference-to-video—not all three assumptions at once. Use a clear action and camera move, then compare subject stability and pacing on the same short scene.

MiniMax H3 is currently available in UlazAI. Check its input controls and credit estimate before generating; third-party claims about free or unrestricted use are not product terms.