Mashed Mice - Dataset Prepper

Mashed Mice - Dataset Prepper

version 1.0

A powerful, slimmed and user-friendly tool for organizing square-format image datasets for AI training — especially ideal for prompt-based systems like Stable Diffusion. It can auto-align resolutions to a preset target format, quickly crop images with the built-in tool, randomize filenames for prompt variety, check spelling and auto-cleanse low res content from the dataset, all with safe file-handling and auto export to lossless PNG with every intervention. There's also a tool to find and remove double or very similar images.

____________________________________________________________

This tool was initially made for this locally run Stable Diffusion trainer:

https://mashed-mice.itch.io/mashed-mice-ai-trainer-for-stable-diffusion

____________________________________________________________

   Features:

-  Smart spell checker with rename suggestions  

-  Corrupt file detection

-  Duplicate image detection using perceptual hashes  

-  Resolution alignment with batch resizing  

-  Manual cropping via built-in crop tool  

-  Theme support (light/dark mode)  

-  Progress tracking with `processed.json`  

-  Recent folder memory  

-  Filename randomizer for training diversity

- 100% local use, you can unplug your network cable

____________________________________________________________

  How to Use:

1. Launch the app — no installation required.

2. Click `File → Open Dataset Folder` to select your image folder.

3. Browse and clean your dataset:

   - Use `Dataset → Align Resolution` to resize all images

   - Use `Crop Tool` to manually prepare single images

   - Use `Spell Check`, `Cleanse`, or `Find Duplicates` for batch sanity checks

4. Processed images are saved to the `Prepared` folder inside your dataset

____________________________________________________________

   Requirements

- Not much more required than a PC with Winrar installed. 

- This app runs as a standalone `.exe` file once the rar file is unpacked. 

- Make sure to not move any files from the original folder.

____________________________________________________________

License

This software is free for personal and commercial use.  

Attribution is appreciated but not required.

Developed by Joel Sandström.

Built with: Python, Tkinter, Pillow, and imagehash.