01
Choose a generation mode
Choose text-to-video, first/last-frame image-to-video, or reference generation for the task at hand.
MiniMax H3 · Open general-purpose multimodal video model
MiniMax H3 AI Video Model
MiniMax H3 is MiniMax’s open general-purpose multimodal video model. It understands text, images, video, and audio in one context, supporting 768P/2K, 24 fps, 4–15 second video generation, reference generation, and video editing.
MiniMax H3 Ready to create
Supports text-to-video, image-to-video, first/last-frame control, and multimodal reference generation.
Prompt examples
Model overview
What is MiniMax H3?
MiniMax H3 is MiniMax’s open general-purpose multimodal video model. It places text, images, video, and audio in one context, supporting text-to-video, image-to-video, first/last-frame control, reference generation, and video editing.
According to MiniMax’s official materials, H3 generates 768P or up to 2K video at 24 fps for 4–15 seconds, with native stereo sound. It is designed for commercial content creation across advertising, branding, e-commerce, product design, UI/UX, gaming, and more.
With image, video, and audio references, H3 can understand relationships between a subject, motion, camera, style, and voice. Describe how the references relate to the target shot in natural language, and let one generation handle the multimodal context.
API model IDMiniMax-H3
Model typeOpen general-purpose multimodal video model
Output768P / 2K · 24 fps
Clip length4–15s, whole seconds
How it works
How to generate videos with MiniMax H3
Start with a prompt, first/last frame, or reference media, then create and refine a multimodal video.
02
Add reference media
Add a first or last frame, or upload image, video, and audio references; mixed input supports up to 12 files.
03
Write the prompt
Describe the subject, camera, action, sound, and the relationship between the references and the target shot.
04
Generate and refine
Submit the asynchronous task and review the result, then keep refining with video editing or regeneration.
Official technical specifications
Built for multimodal video creation
Native stereoAudio
Video generation can produce native stereo audio with the picture
768P / 2KResolution
Supports 768P and up to 2K output at 24 fps
4–15sClip length
Set the duration in whole seconds for each generation
Open modelModel type
MiniMax describes H3 as an open general-purpose multimodal video model
Text + image + video + audioUnified context
Understand multiple input modalities and their relationships in one task
9 + 3 + 3Reference inputs
Up to 9 images, 3 videos, and 3 audio clips; mixed input is capped at 12 files
First + lastFrame control
Use 0, 1, or 2 images to control the opening and ending frames
7,000Prompt length
The API prompt limit is 7,000 characters
Core capabilities
One model for more video-making tasks
01Native stereo sound
Generate video and sound together
H3 puts video and native stereo audio in the same generation flow, making it useful for complete shots with dialogue, sound effects, music, and ambience.
Start creating02Unified multimodal context
Understand text, images, video, and audio together
Describe relationships between references in natural language. H3 can condition on subject, motion, camera, style, and voice for multimodal reference generation and video editing.
Start creating03Commercial content creation
From advertising to product design
MiniMax’s official materials highlight instruction following, accurate text and brand rendering, and V2V motion transfer for advertising, branding, e-commerce, product design, UI/UX, gaming, and more.
Start creatingGeneration modes
MiniMax H3 generation modes
According to MiniMax’s official API documentation, H3 creates and edits video from text, first/last frames, and multimodal reference media.
| Mode | Input | Typical use |
|---|---|---|
| Text-to-video | Prompt | Generate video from a text description |
| First/last-frame image-to-video | Prompt + first and/or last frame image | Control the opening or ending frame and bring a still image to life |
| Reference generation | Prompt + reference image, video, or audio | Reference a subject, motion, camera, style, voice, or editing rhythm |
| Video editing and regeneration | Prompt + existing video | Edit a video or regenerate a qualifying 768P result at 2K |
API
MiniMax H3 input limits
Reference images≤9 · 256–5760 px · ≤30 MB each
Reference videos≤3 · 2–15s each · ≤15s total · ≤50 MB each
Reference audio≤3 · 2–15s each · ≤15s total · ≤15 MB each
Mixed references≤12 image/video/audio files; audio must accompany image or video
First / last frame0–2 images · 256–5760 px · aspect ratio 2:5–5:2
Input formatsVideo H.264/H.265 · image JPG/JPEG/PNG/WEBP/HEIC/HEIF
Audio formatsWAV, MP3
Request body≤64 MB · prompt ≤7,000 characters
Prompt examples
Prompts that move
Use these prompts as starting points for cinematic clips. The locally hosted demo videos are illustrative stock footage; the prompts can guide advertising, branding, and narrative work.
City Rail in Motion
Prompt
A commuter train glides through an elevated city track, cinematic telephoto shot.
Sunset Over Open Water
Prompt
A flock of seabirds crosses a calm ocean at sunset, golden reflections on the water.
Rainy Mountain Drive
Prompt
A car follows a winding mountain road beneath heavy clouds, quiet cinematic travel footage.
Through the Wormhole
Prompt
A luminous portal opens in an abstract tunnel, light rushing toward the camera.
A Quiet Moment
Venice After Dark
Prompt
A slow boat glides through a historic canal at night, warm city lights reflected in the water.
MiniMax H3 prompt gallery
Copy a curated prompt and start creating multimodal AI video.
A commuter train glides through an elevated city track, cinematic telephoto shot.
A flock of seabirds crosses a calm ocean at sunset, golden reflections on the water.
A car follows a winding mountain road beneath heavy clouds, quiet cinematic travel footage.
A luminous portal opens in an abstract tunnel, light rushing toward the camera.
A woman pauses outdoors at dusk, soft bokeh and a contemplative mood.
A slow boat glides through a historic canal at night, warm city lights reflected in the water.
A commuter train glides through an elevated city track, cinematic telephoto shot.
A flock of seabirds crosses a calm ocean at sunset, golden reflections on the water.
FAQ
MiniMax H3 frequently asked questions
What is MiniMax H3?
MiniMax H3 is MiniMax’s open general-purpose multimodal video model. It understands text, image, video, and audio inputs in a unified context and supports video generation, reference generation, and video editing.
Which inputs does MiniMax H3 support?
H3 supports text prompts, images, video, and audio for text-to-video, first/last-frame image-to-video, and multimodal reference generation.
How long and how large can MiniMax H3 videos be?
The official API documentation lists 768P/2K output at 24 fps, with a duration of 4–15 seconds per generation specified in whole seconds.
Does H3 generate sound?
Yes. MiniMax describes H3 as a multimodal video model that generates video with native stereo sound in the same generation flow.
What is MiniMax H3 reference generation?
Reference generation uses reference images, videos, or audio together with a prompt. It helps the model understand a subject, motion, camera, style, voice, or editing rhythm and create the target video.
How many reference files does MiniMax H3 support?
You can use up to 9 reference images, 3 reference videos, and 3 reference audio clips, with mixed input capped at 12 files. Audio references must be accompanied by an image or video input.
Is MiniMax H3 open source?
MiniMax describes H3 as an open general-purpose multimodal video model and provides open weights. Open weights do not necessarily mean that the training data, all code, and every component are released under open-source licenses.
How do I call the MiniMax H3 API?
Submit an asynchronous video generation task with model ID MiniMax-H3, poll the task using its task_id, and retrieve the finished video from content.url when the task succeeds.
MiniMax H3
Turn a multimodal idea into usable video.
Describe an idea with text, images, video, or audio. MiniMax H3 creates 768P/2K video with native stereo sound, ready to edit and refine.
Try MiniMax H3 nowOfficial sources and references
The specifications and capability descriptions on this page are based on MiniMax’s official H3 launch article, video API guide, and model overview.
Example clips use locally hosted free stock footage from Mixkit. This is an independent resource and is not affiliated with MiniMax.
Specifications checked August 4, 2026