MiniMax H3-Image generates still images from text and edits them using up to 16 reference images, producing native frames of about 16 megapixels, including 4096 by 4096, in under 10 seconds. It is aimed at creators and developers who need product shots, game scenes, posters, and other layout-heavy visuals, and who want generation and editing without a separate upscale step. The model is live on the MiniMax API, in MiniMax Code, and on Hailuo AI, with weights published on Hugging Face. The same checkpoint covers text-to-image generation and reference-based editing. MiniMax distilled the 33.1 billion parameter diffusion transformer to six sampling steps with no classifier-free guidance, and the company says a full-resolution 4K image can run at up to five times the speed of mainstream alternatives. Hosted requests at 4K and above are half the standard price during launch week. The community license excludes the United States, the European Union, the United Kingdom, and South Korea, and directs people in those places to contact MiniMax about a separate license. MiniMax is a Shanghai artificial intelligence company listed on the Hong Kong Stock Exchange. It builds multimodal models for text, video, speech, and music, and ships them through Hailuo AI, Talkie, and a developer API. H3-Image is built on the H3 model family, so high-resolution stills sit alongside the company's video tools rather than in a separate stack.