P-H3-Image

Open-weight image model for text-to-image and editing, built on MiniMax-H3, with Diffusers and ComfyUI files.

Also known as Pruna P-H3-Image, P-H3 Image

What it does

P-H3-Image is Pruna AI's open-weights image model, released November 3, 2026, for text-to-image generation and image editing. The post says it is built on the MiniMax-H3 video model, renders images from 2 to 16 megapixels, and edits with up to 9 input images. A later section of the same post says it can combine up to 8 references. The Hugging Face model card says editing takes 1 to 8 input images, that text-to-image is 8 megapixels and edits are 4 or 8 megapixels, and that sampling is 8 steps with no classifier-free guidance. It lists a 33.1 billion-parameter transformer and a 4 billion-parameter text encoder. The post says the smallest version is 20 GB instead of 143 GB for H3 and that 8 megapixel images run in about 21 GB of VRAM. The ComfyUI card says ComfyUI supports the model natively, with no custom nodes, that this support is on the master branch and ships with the next release, and that the smallest set is a 20 GB download peaking at 21 GB of VRAM at 8 megapixels. The post also says the transformer backbone goes from 33 billion to 20 billion parameters. Neither page lists a price. The weights use the MiniMax H3 Community License. Both pages say that license excludes the European Union, the United Kingdom, the Republic of Korea, and the United States unless you have a separate license from MiniMax, and that products built on it must show "Powered by MiniMax H3". The text encoder's Qwen3-VL-4B parts and the prompt upsampler, based on Qwen3.5-9B, are under Apache License 2.0. The post says the ComfyUI nodes are MIT-licensed.

Who is P-H3-Image for?

People who want an open-weight image model they can run in Diffusers or ComfyUI.

Tags

• open-weight
• images
• comfyui