Video · Text → video · Image → video
MiniMax H3
Up to 2K video with stereo sound — from text, frames, or references
About the model
MiniMax H3, also known as Hailuo 03, is MiniMax’s general-purpose multimodal video model. It reads text and images in a single context and produces clips up to 15 seconds at up to 2K. Stereo sound is created together with the picture, so dialogue, music, and effects line up with the action on screen.
The model is well suited to commercial work: it closely follows detailed directions on shot order and camera movement, and renders text, signage, and brand elements cleanly.
Our service offers three modes:
- Text to video — a whole clip from a description, with aspect ratios up to 21:9.
- Image to video — animate a single frame or a transition between a first and last frame.
- Reference to video — up to 5 images that define a character, product, or style.
Strengths
- 768p or 2K output
- Native stereo sound — dialogue, music, ambience, and effects with the video
- Crisp on-screen text, titles, signage, and interface elements
- Precise instruction following — shot order, transitions, camera moves
- Animate a single image or a transition between first and last frames
- Up to 5 reference images for characters and products
- Clips from 4 to 15 seconds, aspect ratios up to 21:9
Best for
- Ads and product videos with logos and text
- Animated posters and key art
- Interface and product demos
- Cinematic scenes with a soundtrack
Limitations
- Maximum length is 15 seconds
- Images — JPG, PNG, WEBP, HEIC, or HEIF up to 30 MB, sides 256–5760 px, aspect 0.4–2.5
- Adaptive aspect ratio isn't available in text-to-video mode
Prompting tips
- Describe shots in order and state which elements must stay unchanged
- Direct the sound in the same prompt — lines, music, noises
- To move between two states, set a first and a last frame
- In reference mode, explain each image's role — character, product, style