LoRA Dataset Auto-Sorter

This skill utilizes Qwen-VL (Vision-Language LLM) to analyze, inspect, and automatically classify training images for LoRA models into structured sub-directories.

Standard Operating Procedure (SOP)

When activated, follow these steps to organize the dataset:

1. Verify Parameters:

   • source_folder: Directory containing raw image files (.jpg, .png, .webp).
   • categories: Target sub-categories (Default: full_body, half_body, close_up).

2. Classification Logic:

   • Inspect image framing and details.
   • Categorize into corresponding folders (full_body, half_body, close_up).
   • Flag corrupted/distorted images into bad_quality.

3. Execution:

   • Execute script scripts/sort_images.py to move files.