MFLUX

Mac users get native local image generation through MFLUX, Filip Strand's port of modern diffusion models to Apple's MLX framework. What began as a FLUX.1 port now covers a wide roster including FLUX.2, the 20B parameter Qwen-Image, Z-Image, FIBO, Krea 2, ERNIE-Image, and the SeedVR2 upscaler, all running on Apple Silicon without CUDA or a discrete GPU. Beyond plain text-to-image it supports quantization to shrink memory use, LoRA loading, image-to-image, ControlNet conditioning, inpainting, upscaling, and metadata export for reproducible generations. Installation is a single command, uv tool install mflux, which provides a command line interface on Python 3.10 or newer; larger models simply need Macs with more unified memory, and quantized variants extend reach to smaller machines. The code is MIT licensed with roughly 2,300 stars and over 550 commits of steady development through 2026, and because it is the main way to run FLUX-class models natively on macOS, a small ecosystem of community tools has grown up around it.