Repolode
Submit
mflux-community

mflux-community/mflux

Native Apple MLX implementations of recent image generation models (FLUX, Z-Image, Qwen Image, and more) for running locally on a Mac.

Open-source alternative to:Midjourney

View on GitHub
Stars
2,520
Forks
202
Open issues
13
Last commit
8 hours ago

What is it?

MFLUX is a Python project that reimplements several recent generative image models from the Hugging Face Diffusers and Transformers libraries in Apple's MLX framework. Models run locally on Apple Silicon Macs without a separate CUDA or PyTorch GPU stack, and weights are downloaded automatically on first use.

The project covers text-to-image, image-to-image, LoRA loading and training, quantization, and editing workflows such as in-context and multi-image editing. Supported families include FLUX.1, FLUX.2, Z-Image, Qwen Image, FIBO, SeedVR2 upscaling and Depth Pro depth estimation, with a command-line interface per model and a Python API.

MFLUX is distributed on PyPI and installed with uv. Its code is kept deliberately small and explicit, so each model can be read against its reference implementation.

Key features

  • 01Runs image models natively on Apple Silicon using MLX, with no CUDA required.
  • 02Supports FLUX.1, FLUX.2, Z-Image, Qwen Image, FIBO, SeedVR2 and other model families.
  • 03Offers quantization and LoRA loading, including multi-LoRA and LyCORIS LoKr.
  • 04Provides per-model CLI commands and a Python API.
  • 05Includes editing features such as in-context editing, inpainting, ControlNet and upscaling.

Use cases

  • Generating images locally on a Mac without sending prompts to a cloud service.
  • Experimenting with several open image models using one consistent toolchain.
  • Applying LoRA adapters to image models on Apple Silicon.
  • Upscaling images with SeedVR2 or estimating depth maps with Depth Pro on a Mac.

Quick start

  1. 1

    Install uv from the link in the README

  2. 2

    Install MFLUX as a uv tool

    uv tool install --upgrade mflux
  3. 3

    Generate your first image with the Z-Image Turbo model

    mflux-generate-z-image-turbo \ --prompt "A puffin standing on a cliff" \ --width 1280 \ --height 500 \ --seed 42 \ --steps 9 \ -q 8

Things to know

  • Designed for Apple Silicon Macs running MLX, so it is macOS-only.
  • Models download automatically on first run, which can take a long time and use a lot of disk space.
  • Some large models, such as Qwen Image at 20B, run slower.

Topics

#mlx#apple-silicon#image-generation#diffusion-models#flux#lora#python#ai#diffusers#fibo

FAQ

What is mflux?
Native Apple MLX implementations of recent image generation models (FLUX, Z-Image, Qwen Image, and more) for running locally on a Mac. MFLUX is a Python project that reimplements several recent generative image models from the Hugging Face Diffusers and Transformers libraries in Apple's MLX framework. Models run locally on Apple Silicon Macs without a separate CUDA or PyTorch GPU stack, and weights are downloaded automatically on first use.
Is mflux free to use? What license does it have?
mflux is open source under the MIT license. Check the license terms before using it in commercial products.
Is mflux still maintained?
Actively maintained. The last commit to mflux was on Oct 20, 2026.
What can I use mflux for?
Generating images locally on a Mac without sending prompts to a cloud service. Experimenting with several open image models using one consistent toolchain. Applying LoRA adapters to image models on Apple Silicon. Upscaling images with SeedVR2 or estimating depth maps with Depth Pro on a Mac.
How do I install mflux?
Install uv from the link in the README Install MFLUX as a uv tool: uv tool install --upgrade mflux Generate your first image with the Z-Image Turbo model: mflux-generate-z-image-turbo \ --prompt "A puffin standing on a cliff" \ --width 1280 \ --height 500 \ --seed 42 \ --steps 9 \ -q 8
Where can I get mflux?
The source code and installation instructions are on GitHub: https://github.com/mflux-community/mflux

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects