Mashed Mice - Local AI trainer for Stable Diffusion

Mashed Mice Fine-Tuner (for Stable Diffusion)

Version: 1.0.0

This tool lets you fine-tune your own Stable Diffusion model using your own image dataset.

The training is 100% local — no internet connection is required after setup.

────────────────────────────────────────────

HOW TO USE:

1. Download a compatible Stable Diffusion model  from https://huggingface.co/.

2. Prepare a dataset folder with square PNG images (e.g., 512x512, 640x640, etc).

3. Launch the app and:

   - Select the model folder (must contain the model files like `unet`, `vae`, `text_encoder`, etc).

   - Select the dataset folder (your images).

   - Select where you want to save the fine-tuned model.

4. Adjust training settings or use the Quick Setup button to auto-calculate safe defaults.

5. Click "Start Training" and follow the live progress.

Samples will be generated during training (if enabled), and logs will be saved along with your model.

────────────────────────────────────────────

HELPFUL STUFF:

If You're looking for a free and local dataset editor, here's a custom made tool designed specifically for this trainer, but it will work for any square-sized dataset:

https://mashed-mice.itch.io/mashed-mice-dataset-prepper

────────────────────────────────────────────

REQUIREMENTS:

- Windows PC with a CUDA-compatible GPU (e.g. RTX 3060/4060 or better)

- Tested on: GeForce RTX 4060 16GB VRAM GPU

- Recommended: 16GB system RAM or more

────────────────────────────────────────────

LICENSE INFORMATION:

This software is licensed under the MIT License.

Copyright (c) 2025 Joel Sandström

This app uses the `diffusers` library by Hugging Face,

licensed under the Apache License 2.0:

https://github.com/huggingface/diffusers/blob/main/LICENSE

────────────────────────────────────────────

DISCLAIMER:

This software is provided "as-is" without warranty.

Use at your own risk.

────────────────────────────────────────────