MiniMax H3
MiniMax H3 is an open general multimodal video model built for creators who want to generate, reference, edit, and reshape complete audiovisual content in one workflow.
Unlike a simple text-to-video or image-to-video model, MiniMax H3 treats text, images, videos, and audio as a unified creative context. It can understand characters, actions, camera movement, composition, emotion, sound, dialogue, voice, atmosphere, and visual style, then use that understanding to generate or modify a complete video scene.
Key Features
l Native Multimodal Fusion
MiniMax H3 can combine text, images, videos, and audio as creative references. It supports up to 9 images, 3 video clips, and 3 audio clips, with up to 12 mixed reference assets in total. This allows creators to guide the model with richer context instead of relying on one prompt alone.
l Fine-Grained Reference Control
H3 can reference different parts of your creative input separately. Use one image for character identity, another for visual style, a video for motion, another clip for camera movement, and an audio file for mood or voice. This makes it more flexible than basic reference-to-video generation.
l Generate and Edit in One Workflow
MiniMax H3 is designed to support both creation and modification. You can generate new videos, edit existing clips, replace characters, add or objects, change backgrounds, adjust light and effects, or rewrite parts of the scene without rebuilding everything from zero.
l Audio-Visual Integration
H3 supports integrated video and sound workflows. It can work with dialogue, human voice, audio references, sound atmosphere, and voice style. It is especially useful for scenes where the sound needs to match the character, action, and emotion.
l Voice Cloning and Voice Transfer
Use reference audio to guide or replace a character's voice. This makes H3 valuable for digital humans, character dialogue, brand videos, dubbing, short dramas, and localized content production.
l Stable Local Editing
When editing one part of a video, H3 focuses on keeping the unedited parts stable. This is useful for changing a product, replacing a person, adjusting dialogue, or modifying a background while preserving the rest of the scene.
l Multi-Shot and Story-Aware Creation
MiniMax H3 can understand shot logic, transitions, narrative rhythm, and camera movement. It is suitable for short trailers, product showcases, visual proposals, storyboards, and cinematic scene previews.
l Flexible Output Formats
H3 supports 5 to 15 second video generation and multiple aspect ratios, including 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. In reference mode, it can also automatically infer the best aspect ratio from the input materials.
Prompt Examples for SeaArt Creators
AI Character Dialogue Video
Use the uploaded character image as the main reference. Keep the face, hairstyle, and outfit consistent. The character stands in a neon city at night and says, "Welcome to my world," with a calm and confident voice. Use the uploaded audio as the voice reference. Slow camera push-in, soft rain, glowing reflections, cinematic anime style, 9:16 vertical video.
Product Showcase
Use the uploaded product image as the main reference. Create a premium commercial video where the product rotates on a glossy black surface, neon light moves across the edges, and the camera slowly circles around it. Add soft electronic ambience and a clean reveal sound. Keep the product shape and logo consistent. 16:9 landscape video.
Replace Object in a Video
Use the uploaded video as the base. Replace the phone in the actor's hand with the uploaded product image. Keep the actor, background, camera movement, lighting, and timing unchanged. Make the replacement look natural with matching shadows and reflections.
Voice Transfer Scene
Use the uploaded video as the base. Keep the character's movement and facial expression the same, but replace the voice with the uploaded audio reference. The line should be: "This is the future of AI video creation." Keep the tone warm, confident, and natural.
Game Visual Trailer
Create a short game trailer scene. A female warrior runs through a burning battlefield, dodges an arrow, then turns toward the camera. Use image 1 for the character design, video 1 for action movement, and audio 1 for epic battle atmosphere. Fast camera movement, sparks, smoke, dramatic lighting, 21:9 cinematic ratio.
Brand Film Preview
Create a luxury brand film shot. A model walks through a minimal white gallery while fabric moves in slow motion. Use the uploaded fashion image as the outfit reference and the uploaded music as the mood reference. Soft light, slow tracking shot, elegant pacing, premium campaign style.
Style Transfer Edit
Use the uploaded video as the base. Keep the same motion, camera movement, and scene structure, but transform the visual style into a hand-drawn anime film look. Keep the character identity recognizable and preserve the original timing.
How to Get Started
1. Access the Model
Open MiniMax H3 on SeaArt AI.
2. Choose Your Materials
Start from text, upload images, add video references, use audio references, or combine multiple assets.
3. Assign Each Reference a Role
Tell H3 what each asset should control: character, product, style, motion, camera, music, voice, or atmosphere.
4. Use the H3 Reference Formula
Write your prompt with Goal + Main Asset + Reference Role + Action + Camera + Sound + Edit Rule + Output Format.
5. Generate the First Version
Review character consistency, camera movement, motion quality, sound, dialogue, and visual style.
6. Edit Locally
Replace an object, adjust the lighting, change the background, revise the dialogue, or update the voice while keeping the rest stable.
7. Iterate Toward the Final Video
Use H3 as a repeatable creative workflow: generate, edit, refine, and adapt for different platforms.