Bonsai Image Demo
Bonsai Image Demo is a command-line demo that generates images locally with the ternary or binary quantized Bonsai model, running on Apple Silicon via MLX or on NVIDIA GPUs on Linux and Windows through gemlite and HQQ kernels.
Bonsai Image Demo runs local image generation with the Bonsai model family. It targets three backends: Apple Silicon Macs through mflux and MLX, NVIDIA GPUs on Linux through gemlite and HQQ kernels, and NVIDIA GPUs on Windows using the same gemlite and HQQ stack via triton-windows, without needing WSL2.
A setup script downloads the chosen model variant, ternary (1.58-bit, the recommended default) or binary (1-bit, smaller), and a generate script produces an image from a text prompt. Configuration is controlled through environment variables such as BONSAI_VARIANT. By default, setup and serve scripts also refuse to install any Python or npm package version newer than a configurable minimum age, as a guard against fresh supply-chain compromises.