mflux

Why we included this project

On a Mac with Apple silicon, MFLUX runs current generative image models locally instead of routing every prompt through a paid cloud API. It ports Z-Image, FLUX.2, Ideogram 4, Qwen Image, and SeedVR2 into Apple's MLX framework, so they execute natively without a GPU cluster. The code is deliberately minimal and explicit, which makes it a good place to see how these diffusion transformers actually work. A CLI and Python API cover text-to-image, image-to-image, in-context editing, depth conditioning, and upscaling, and quantization with LoRA fine-tuning lets you customize the output. That makes it a practical fit for creative iteration and prototyping on Apple hardware, without assembling a separate Python diffusion stack.