Updated

P-H3-Image

Pruna AI released P-H3-Image, an open-weights text-to-image and image-editing model built from MiniMax-H3. The team reworked the prompt upsampler, text encoder, diffusion schedule, transformer backbone, and decoder, and says the model generates 4 MP images about five times faster than MiniMax-H3 on a B200.

By No Code MBA · Facts checked

Provider
Pruna AI
Announcement date
Status at last check
Generally available

What changed

P-H3-Image adapts MiniMax-H3, a video model, into an image generation and editing model. Pruna AI says it reworked the prompt upsampler, text encoder, diffusion schedule, transformer backbone, and decoder, enabling eight-step generation and image editing with up to eight input images in the model card. The launch post reports 2.6 seconds for a 4 MP image on a B200 versus 12.4 seconds for MiniMax-H3; this is the company's reported measurement.

Access and availability

Download the Diffusers pipeline from the PrunaAI/P-H3-Image repository on Hugging Face. Pruna AI also provides a separate ComfyUI repository with quantized weights. The official model card lists substantial GPU memory requirements for the Diffusers version.

Official access documentation

What you can build

Evaluate it on the same prompts and image-editing references you use today, then check the MiniMax H3 Community License and hardware requirements before deployment.

Limitations

The model is based on MiniMax-H3 and governed by the MiniMax H3 Community License, which excludes the United States, European Union, United Kingdom, and Republic of Korea unless separately licensed. Pruna AI notes that fine details, hands, and small or dense text may be unreliable. The Diffusers model card lists about 101 GB GPU memory at 8 MP with the prompt upsampler; quantized ComfyUI weights are separate.

Sources and release history