Fluxgym: Simple FLUX LoRA Training UI with Low VRAM Support

Introduction

Fluxgym is an intuitive web user interface designed to simplify the process of training FLUX LoRA models, especially for users with limited VRAM. It provides robust support for GPUs with 12GB, 16GB, and 20GB of VRAM, making advanced model training more accessible. The project integrates a user-friendly Gradio frontend, inspired by AI-Toolkit, with the highly flexible and powerful backend of Kohya sd-scripts, ensuring full compatibility with all Kohya features.

Installation

Fluxgym offers multiple convenient installation methods:

One-Click Install

For the easiest setup, you can use the Pinokio 1-click launcher:
https://pinokio.computer/item?uri=https://github.com/cocktailpeanut/fluxgym

Manual Installation

1. Clone Repositories:

   git clone https://github.com/cocktailpeanut/fluxgym
   cd fluxgym
   git clone -b sd3 https://github.com/kohya-ss/sd-scripts
   

2. Create and Activate Virtual Environment:
   • Windows:

     python -m venv env
     env\Scripts\activate
     

   • Linux:

     python -m venv env
     source env/bin/activate
     

3. Install Dependencies:

   cd sd-scripts
   pip install -r requirements.txt
   cd ..
   pip install -r requirements.txt
   

4. Install PyTorch Nightly:
   • For most GPUs (cu121):

     pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
     

   • For NVIDIA RTX 50-series (cu128):

     pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128
     pip install -U bitsandbytes
     

5. Start Fluxgym:

   Ensure your virtual environment is activated, then run:

   python app.py
   

Docker Installation

1. Clone Repositories:

   git clone https://github.com/cocktailpeanut/fluxgym
   cd fluxgym
   git clone -b sd3 https://github.com/kohya-ss/sd-scripts
   

2. Build and Run with Docker Compose:

   Adjust PUID and PGID environment variables if your user ID and group ID are not 1000.

   docker compose up -d --build
   

   Access the UI at http://localhost:7860 in your web browser.

Examples

Using Fluxgym is designed to be straightforward:

1. Enter your LoRA information.
2. Upload your training images and caption them, including your chosen trigger word.
3. Click "start" to begin training.

Fluxgym also includes powerful configuration options:

• Automatic Sample Image Generation: Configure prompts and frequency to generate sample images during training, allowing you to monitor the LoRA's evolution. Advanced flags like --d (seed) can be used for consistent comparisons.
• Publishing to Hugging Face: Easily log in with your Hugging Face token and publish your trained LoRA models directly from the UI.
• Advanced Kohya Features: An "Advanced" tab, hidden by default, exposes all underlying Kohya sd-scripts launch flags, providing full control over the training process.
• Caption File Uploads: Supports uploading .txt caption files alongside corresponding image files.

Why Use Fluxgym?

Fluxgym addresses several key challenges in FLUX LoRA training:

• Low VRAM Accessibility: It democratizes FLUX LoRA training by supporting GPUs with as little as 12GB VRAM, which is a significant barrier for many users.
• Simplified User Experience: It provides a "dead simple" web UI, abstracting away the complexities of command-line arguments, making training accessible to a wider audience.
• Full Kohya Script Power: By leveraging Kohya sd-scripts as its backend, Fluxgym ensures users have access to all advanced features and flexibility, without sacrificing ease of use.
• Streamlined Workflow: Features like automatic model downloads, sample image generation, and direct Hugging Face publishing streamline the entire LoRA training and sharing process.

Links

• GitHub Repository: https://github.com/cocktailpeanut/fluxgym
• Pinokio One-Click Install: https://pinokio.computer/item?uri=https://github.com/cocktailpeanut/fluxgym
• More Info (X Thread): https://x.com/cocktailpeanut/status/1832084951115972653