Add as a preferred source on GoogleMiniMax Open-Sources MiniMax-H3-Image for 16-Megapixel Output
The image model uses one set of weights for generation and editing, accepts up to 16 reference images and is available through MiniMax’s API and Hailuo AI.
MiniMax has open-sourced MiniMax-H3-Image, an image model that generates images from text and edits images using the same set of weights. It supports up to 16 reference images and is available under a community license agreement.
The model can produce images of about 16 megapixels without a separate super-resolution model. Supported dimensions include 4,096 × 4,096 and 6,272 × 2,688. Its sampling process has been distilled to six steps.
MiniMax-H3-Image’s diffusion Transformer has 33.1 billion parameters, and its weights total about 68.65 GB. Developers can deploy the model with SGLang, vLLM-Omni, diffusers or ComfyUI. It is also available through MiniMax’s API and Hailuo AI.
MiniMax described the model’s instruction-following ability as the strongest among open-source models.
Simon Yoon
ReporterSimon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.