MiniMax Open-Sources H3 Image, Bringing 4K Image Generation Down to the Ten-Second Range
MiniMax (Xiyu Technology) has open-sourced its MiniMax H3 Image model and released open weights. It supports text-to-image, image reference, and image editing, generating 4096×4096 (~16MP) 4K images in about 10 seconds and accepting up to 16 reference images. Standard mode is about 1.5–2× faster than mainstream tools; high-speed mode is 3–4× faster. It is available on Hugging Face and via API, and integrated into MiniMax Code, Hailuo, and other products.
MiniMax Open-Sources MiniMax H3 Image, Bringing 4K Image Generation to the Ten-Second Range
MiniMax has released the MiniMax H3 Image model and opened its weights. The model supports text-to-image, image reference, and image editing, and can generate 4K images at 4096×4096 resolution—about 16 megapixels—in roughly 10 seconds. It accepts up to 16 reference images as input and focuses on improving complex instruction understanding, multi-image reference, and visual detail.
The company also launched a Highspeed edition. Generation speed in standard mode is about 1.5 to 2 times that of mainstream tools, while Highspeed mode reaches 3 to 4 times. The model is now available on Hugging Face and via the open platform API, and has been integrated into products including MiniMax Code and Hailuo. The standard version costs about $0.069 per 4K image, with a promotional price of about $0.034 during the first week after launch.
Hugging Face: https://huggingface.co/MiniMaxAI/MiniMax-H3-Image