MFLUX
A line-by-line MLX port of several state-of-the-art generative image and video models from the Hugging Face Diffusers and Transformers libraries, letting a Mac run models such as FLUX and Qwen-Image locally through native Apple Silicon acceleration.
MFLUX reimplements several state-of-the-art generative image and video models from scratch in MLX, Apple's array framework, rather than wrapping the original PyTorch implementations; only tokenizers are reused from Hugging Face Transformers. The project is deliberately kept minimal and explicit.
It installs as a command-line tool with a separate generation command for each supported model family, such as a Z-Image Turbo model, and downloads model weights automatically on first use. It also exposes a Python API, so a model can be loaded and run from a short script instead of the command line, with quantization options to reduce memory use.