Bonsai Image 4B

Bonsai Image 4B is a binary/ternary text-to-image diffusion transformer released by PrismML. It is designed to run locally in a browser using WebGPU, with a model size of only 3GB. This makes high-quality image generation accessible without requiring powerful cloud GPUs or large downloads. The model uses efficient quantization techniques to reduce memory and compute requirements while maintaining image quality. It is part of a growing trend toward on-device AI that prioritizes privacy and offline capability. Based on community signals so far, the model has been shared on Reddit and is gaining attention for its practicality and low barrier to entry.

PrismML's release of a binary/ternary diffusion model that runs in-browser at 3GB sparked interest on Reddit's LocalLLaMA community, highlighting a shift toward efficient, local image generation.