MiniMax H3 image generation model officially open-sourced: 6 steps to directly output 16.8 million pixels, community has already been running ahead for a week.

Beating AI News Flash: MiniMax has officially released H3-Image and open-sourced the model weights. It can directly generate 4096×4096 images at approximately 16.8 million pixels without additional upscaling. According to official data, generation takes less than 10 seconds, up to 5 times faster than mainstream products. It also integrates text-to-image and image editing into a single model, supporting up to 16 reference images.

H3-Image follows the architecture of the H3 video model, removing the audio component, retaining the original text encoder, and redesigning the decoder for images. The core image generation network has 33.1 billion parameters. After distillation, generating a single image requires only 6 sampling steps, significantly reducing computational steps.

This model was actually teased back in August. On August 7, MiniMax researcher Kiro revealed in a Reddit AMA that the team was developing a dedicated image model from the same architecture as H3 and planned to open-source the weights. A workflow was also proposed at the time: first use the image model to generate the first frame, then hand it off to H3 to continue generating video.

The community even beat MiniMax to it. On November 3, Pruna AI launched P-H3-Image based on the open-source H3 video model, also supporting image generation and editing. Pruna compressed the generation network from 33.1 billion to approximately 20 billion parameters, requiring only 8 sampling steps. The smallest quantized version is about 20GB, and when generating 8-megapixel images, peak VRAM usage is only about 21GB.

The official MiniMax version retains the 33.1 billion parameter generation network, further compresses sampling to 6 steps, and supports higher resolution and more reference images. However, the original weights plus the reused text encoder total about 135GB, which is noticeably heavier.

The open-source weight license for H3-Image has restrictions. The United States, United Kingdom, EU, and South Korea are not within the authorized regions of the community license. Enterprises with annualized revenue exceeding $1 million, or those providing model hosting services to third parties, must obtain separate authorization in advance.