MiniMax has open-sourced MiniMax-H3-Image, a unified image generation and editing model that directly produces images up to 16 megapixels without requiring a separate super-resolution step. A single set of weights handles both text-to-image and image editing tasks, accepting up to 16 reference images as input. The diffusion Transformer architecture contains 33.1 billion parameters with approximately 68.65GB of weights, supporting resolutions from 4096×4096 to 6272×2688 and distilled sampling steps reduced to just 6 iterations. The model can be deployed via SGLang, vLLM-Omni, diffusers, and ComfyUI frameworks. MiniMax claims H3-Image offers the strongest instruction-following capability among open-source image models. The model is also accessible through the MiniMax API and Hailuo AI platform, released under a community license agreement.