MiniMax H3

MiniMax H3 Features and Generation Modes

Explore MiniMax H3 features: text, image, frame-pair and subject-reference workflows, available 2K output, native audio and camera-level direction.

Try MiniMax H3View pricing

Quick answer

MiniMax H3 combines multiple creative inputs in one video workflow so creators can choose between open-ended generation and tighter visual control.

Updated August 1, 2026Independent service · Not affiliated with MiniMax

Capability sheet

MiniMax H3 specifications and workflows

Generation modes
Text · Image · First & Last Frame · Subject Reference
Primary output
Up to 2K AI video where available
Image inputs
Starting plate, frame pair or subject image
Camera control
Prompt-level pan, push-in, tracking, orbit and static framing
Aspect ratios
Landscape, square and vertical options in the generator
Delivery
Browser preview followed by download

Workflow table

Modes, inputs and practical use

CapabilityInputControlBest use
Text to VideoPromptOpen scene and camera directionConcepts and storyboards
Image to VideoImage + promptStarting composition and motionProduct and key-art animation
First & Last FrameTwo images + promptKnown endpoints and transitionReveals and transformations
Subject ReferenceSubject image + promptIdentity plus directed actionCharacter-led variants
Native audio workflowPrompt and supported referencesScene sound directionMore complete short-form concepts
Up to 2K outputSelected generation settingsResolution and format choiceDetailed final concepts
The same workflow can start from a written direction, an image, endpoint frames or a subject reference.

01

Generation and output

Four input paths let creators begin with the amount of direction they actually have, then generate a detailed short-form output in the browser.

  • Text-led ideation
  • Still-image animation
  • Endpoint-controlled transitions
  • Subject-led character scenes

02

Inputs and consistency

Images can define composition, endpoints or subject identity. Explicitly state what should remain stable while the prompt drives action and environment.

  • Preserve products and logos
  • Reuse one clean subject plate
  • Match endpoint composition
  • Keep references visually unambiguous

03

Motion and camera direction

Prompt-level camera language helps the model choose one readable visual idea instead of improvising several conflicting moves.

  • Static framing
  • Pan and push-in
  • Tracking movement
  • Controlled orbit

04

Subject reference

A subject plate provides a stronger identity anchor than text description alone. It is suited to creator ads, presenter scenes and character-led campaign variants.

  • Face and character continuity
  • Separate wardrobe from identity
  • Prefer one continuous action
  • Avoid conflicting references

05

First-and-last-frame transitions

Two visual endpoints direct where a scene begins and resolves. The prompt controls the bridge, pacing, camera and light continuity.

  • Product reveals
  • Before-and-after stories
  • Day-to-night changes
  • Style transformations

06

Production workflow

Creators choose a mode, write the brief, see available settings, generate, preview and download without a traditional timeline editor.

  • Cost visible before submit
  • Task status and history
  • Browser preview
  • Downloadable output

07

Known practical limits

Complex multi-scene instructions, dense readable text, fast interactions and exact brand reproduction require careful prompting and review.

  • Use one clear action per short clip
  • Inspect hands, faces, labels and logos
  • Do not assume every output is production-ready
  • Retain source assets and prompt records

Common questions

Features FAQ

What are the four main H3 generation modes?

Text to Video, Image to Video, First & Last Frame and Subject Reference.

Does MiniMax H3 support 2K?

The H3 workflow targets output up to 2K where the selected model settings make that option available.

Can H3 animate an existing image?

Yes. Image to Video uses the still as the starting composition while the prompt directs movement and atmosphere.

How does first-and-last-frame control work?

You provide the opening and closing images, then describe the motion and visual change connecting those endpoints.

How do I keep a person recognizable?

Use Subject Reference with one clear subject image and avoid instructions that contradict identity.

Can I combine every input type in one task?

Available combinations depend on the active H3 API mode. The generator displays only the inputs supported by the chosen workflow.

Related resources

How to use every mode Prompt examples

Create your first H3 video

Start with text or an image, direct one clear shot, then preview and refine the result in your browser.

Open the generator