Flux Gym is a web UI for training FLUX LoRA models with low VRAM support.
The tool addresses the challenge of fine-tuning FLUX models on consumer-grade GPUs. It combines a simple web interface forked from AI-Toolkit with the training backend powered by Kohya Scripts, enabling LoRA training on systems with 12GB, 16GB, or 20GB of VRAM. The interface hides advanced options by default while exposing the full feature set of Kohya Scripts through an Advanced tab for users who need fine-grained control.
Flux Gym suits developers and creators who want to train custom FLUX LoRAs locally without expensive hardware. It works with Flux1-dev, Flux1-dev2pro, and Flux1-schnell models, with automatic model downloading on first use. The tool is extensible through its models.yaml configuration file, allowing users to add custom base models beyond the defaults. Unlike AI-Toolkit, which requires 24GB VRAM, this project explicitly targets lower-memory systems. Unlike raw Kohya Scripts, it eliminates the need to work from the terminal.
The project shows active development with recent additions including Docker support, automatic model downloading, custom resolution support, automatic sample image generation, and Hugging Face publishing integration. The codebase integrates established training infrastructure rather than building training logic from scratch, which reduces maintenance burden while ensuring compatibility with the broader Kohya ecosystem.
