MiniMax launches H3-Image model that makes 16MP images in 10 seconds

MiniMax has officially launched a new image generation model called H3-Image, with open weights. Built on H3, it can create images from text, use reference images, and edit pictures. Users can access it through Hugging Face and the API. The model can generate a 4096x4096 pixel image (about 16MP) in just 10 seconds, five times faster than most tools on the market. Compared with common 1MP or 4MP images, 16MP pictures show much richer details. MiniMax says the model was improved with better data, reinforcement learning, and deployment work. It offers strong realism, follows instructions well, and supports up to 16 reference images. For the first week, requests for 16MP and higher are half price.