High-resolution images can now be generated in 2.6 seconds using a newly open-sourced model that converts video AI into an efficient image generator. The P-H3-Image model from Pruna was adapted from the MiniMax-H3 video AI and is 7x smaller than its predecessor, allowing it to produce 4 MP images in 8 generation steps rather than the 50 steps typically required.
The tool supports resolutions between 2 and 16 MP and accepts up to 8 input images for editing tasks. Pruna reworked the original transformer, diffusion schedule, and text encoder to enhance photorealism in documentary and film styles. The model’s open weights are currently available via Hugging Face with native support for ComfyUI.
