Lofi LoRA Data Prep

A desktop tool (Electron + Python sidecar) for curating LoRA datasets: rename, shuffle, convert, and tag images with automated captioning.

Discord

https://discord.gg/zxRACHM Maybe I'll help you with your errors, maybe I wont. Depends on my coffee supply

Features

• GUI for dataset management (rename, shuffle, convert formats)
• Automated image captioning via JoyCaption (LLaVA-based) (currently janky, but works)
• Tag editing and prompt customization
• Works with CPU or NVIDIA GPU (INT8, NF4, BF16 quantization)
• Built-in GPU/VRAM monitor

Requirements

• Python 3.10+
• Node.js + npm (from https://nodejs.org)
• NVIDIA GPU (optional for acceleration — CPU mode works everywhere)

Attibutions

This project uses JoyCaption for automatic image captioning. https://github.com/fpgaminer/joycaption JoyCaption is licensed under the Apache License 2.0 https://github.com/fpgaminer/joycaption/blob/main/LICENSE

Quickstart

Windows

The tool works, but is currently having a few bugs on windows that it doesnt have on linux, i am working on patching them.

1. Clone or download this repo

   git clone https://github.com/JustLateNightAI/Lofi-Lora-Data-Prep.git
   cd Lofi-Lora-Data-Prep

2. Run setup

   • Double-click setup.bat
   • Creates a Python virtualenv under sidecar\\
   • Installs Python + npm dependencies
   • Prompts you to pick CPU or CUDA PyTorch

3. Run the app

   • Double-click start.bat
   • Electron launches and automatically spawns the sidecar

Linux

1. Clone the repo

   git clone https://github.com/JustLateNightAI/Lofi-Lora-Data-Prep.git
   cd Lofi-Lora-Data-Prep

2. Set up Python sidecar

   cd sidecar
   python3 -m venv venv
   source venv/bin/activate
   pip install --upgrade pip
   pip install -r requirements.txt
   # Optional: install torch with CUDA (NVIDIA only)
   # pip install --index-url https://download.pytorch.org/whl/cu124 torch torchvision torchaudio
   cd ..

3. Set up Electron app

   npm install

4. Run the app

   # Make sure your venv is active so the sidecar uses it
   source sidecar/venv/bin/activate
   npm run dev

⚙ Notes

• No NVIDIA? Choose CPU during setup (or install CPU Torch on Linux).
• If CUDA install fails on Windows, update NVIDIA drivers or rerun setup and choose CPU.
• Logs stay open in the console for debugging.

Screenshots

License

Creative Commons BY-NC 4.0 License