LoRA Preprocess Master

LoRA Preprocess Master is a simple and tiny GUI tool for one-click preprocessing and automatic captioning of LoRA model datasets. It uses very light weighted captioning models moondream and BLIP, which can easily run on your local machine.

Features

• Batch Image Preprocessing

  • Batch resize images to common LoRA training sizes (512 for SD1.5, 1024 for SDXL, or custom).

  • Option to crop to square.

  • Output format conversion (JPEG / PNG).

• Automatic Image Captioning

  • Two integrated models:

    • BLIP - CPU-friendly option

    • Moondream - Requires CUDA and GPU (~5GB VRAM)

  • Batch adding of trigger words

  • Cache directory support for efficiency

• One-click Workflow

  • Simply set input and output folders, enable the features you want, and click Start.

The goal is simple: fully local, super lightweight, and absolutely minimal. Give it a try and let me know how it runs, or if you think I should add more features.

Github link: https://github.com/jiaqi404/LoRA-Preprocess-Master