MiniMax H3-Image is an open-weight model for text-to-image generation and multi-reference editing. It produces native 16-megapixel images with strong prompt adherence in roughly ten seconds.