FluxGym

FluxGym pairs a simple Gradio front end with the Kohya sd-scripts training backend so that FLUX LoRA training works on 12, 16, or 20 GB GPUs instead of the 24 GB and up that most trainers assumed. Created by cocktailpeanut, the author of the Pinokio launcher, it installs with one click through Pinokio or manually with a Python virtual environment, and a Docker deployment option exists as well. The interface covers the whole training loop: it downloads base models automatically, supports Flux1-dev, Flux1-dev2pro, and Flux1-schnell along with custom base models, captions datasets, generates sample images during training with advanced sampling flags, and can publish finished LoRAs straight to Hugging Face. An advanced tab exposes the full range of Kohya sd-scripts options for users who outgrow the defaults, so simplicity does not cap what the tool can do. The project is MIT licensed with about 3,200 stars, and its low barrier to entry made it a common first stop for hobbyists training their own FLUX LoRAs on consumer hardware.