Flux Gym is a deliberately simple web interface for training FLUX LoRAs, available on GitHub with about 3,257 stars under the MIT license and most recently active in 2026. Its reason to exist is hardware: serious LoRA training tools often assume a high-end GPU, while Flux Gym is built to run comfortably on graphics cards with 12GB, 16GB, or 20GB of VRAM.
The project combines two proven pieces. The frontend is a Gradio web UI forked from AI-Toolkit, giving you a friendly form-based workflow instead of terminal commands. The backend is powered by Kohya’s sd-scripts, the flexible and widely used training toolkit, and Flux Gym exposes the complete feature set through an Advanced tab that stays hidden until you need it. Setup is designed to be painless: Docker support is included, base models download automatically, and you can train on custom base models, not just the default FLUX variants, by registering them in a configuration file. When training finishes, you can publish your LoRA straight to Hugging Face from the app. The result is a tool that lowers the entry barrier to personal model training without giving up the depth of the underlying scripts.
Flux Gym is free and open source under the MIT license. You only need your own GPU or a rented one.
Flux Gym is for artists, hobbyists, and creators who want to train their own FLUX LoRAs for characters, styles, or products on modest hardware, without learning a complex training stack first.