# Pruna AI Releases P-H3-Image With a 20 Billion-Parameter Network

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/54554
Published: 2026-11-11T22:16:31.000Z
Updated: 2026-11-11T22:16:31.000Z
Section: Technology

> The image model supports generation and editing. Its smallest quantized version is about 20 GB and uses roughly 21 GB of VRAM when generating an 8-megapixel image.

Pruna AI released P-H3-Image Nov. 3, an image-generation and editing model built on MiniMax’s H3 video model. Its generation network has about 20 billion parameters and uses eight sampling steps.

The smallest quantized version is about 20 GB. Generating an 8-megapixel image uses about 21 GB of VRAM at peak.

MiniMax’s [open-weight H3-Image release](<https://www.tokenpost.com/news/technology/54190>) uses a 33.1-billion-parameter image network and six sampling steps. It supports image editing and up to 16 reference images, and can generate images measuring 4,096 by 4,096 pixels, or about 16.8 million pixels.



## Links in this article

- [open-weight H3-Image release](https://www.tokenpost.com/news/technology/54190)