filipstrand/mflux

MLX native implementations of state-of-the-art generative image models

79
/ 100
Verified

Implements multiple state-of-the-art diffusion models (FLUX, Z-Image, FIBO, Qwen) as line-by-line MLX ports optimized for Apple Silicon, with quantization support and minimal, explicit code following Karpathy-style principles. Supports advanced capabilities including LoRA fine-tuning, in-context editing, ControlNet, inpainting, and upscaling across different model families. Integrates with Hugging Face model hub and tokenizers while maintaining a pure MLX backend for efficient local inference on macOS.

1,882 stars and 31,305 monthly downloads. Used by 3 other packages. Actively maintained with 17 commits in the last 30 days. Available on PyPI.

Maintenance 20 / 25
Adoption 23 / 25
Maturity 18 / 25
Community 18 / 25

How are scores calculated?

Stars

1,882

Forks

127

Language

Python

License

MIT

Last pushed

Mar 09, 2026

Monthly downloads

31,305

Commits (30d)

17

Dependencies

22

Reverse dependents

3

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/filipstrand/mflux"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.