Open-weight image generation model derived from H3, supporting text-to-image, reference and editing with 16MP output in 10 secon
Moore Threads announced that its MTT S5000 AI card and MUSA software stack were adapted for MiniMax H3-Image on the same day the model was open-sourced, enabling single-card generation of 4096×4096 high-resolution images. The adaptation reuses SGLang, MATE and MCCL components and targets the heavy attention overhead of diffusion models under multi-reference inputs, laying groundwork for Day-0 support of future image and video generation models.
MiniMax launched H3-Image, an image generation model that outputs 16-megapixel (4K) images in roughly 10 seconds, supports up to 16 reference images, and offers precise prompt following and editing. It ships with open weights and is available via the MiniMax API, MiniMax Code, and Hailuo AI, while also entering Arena's battle mode for blind user voting.
MiniMax has released H3-Image with open weights, an image generation model derived from its H3 base that supports text-to-image, image reference and image editing. It can produce a 4096x4096 (about 16MP) 4K image in 10 seconds, which MiniMax claims is roughly 5x faster than mainstream alternatives, and accepts up to 16 reference images with what it calls the strongest instruction following among open models.