Thank you. Maybe I could interest you in this free dataset prepper tool instead. It's usable for any similar trainer using square-image datasets: https://mashed-mice.itch.io/mashed-mice-dataset-prepper
A member registered May 11, 2025 · View creator page →
Creator of
A local AI dataset preparation tool for square-format dataset
This program lets you train your SD model with local files - No internet connection needed!
Recent community posts
itch.io Community » General » Release Announcements · Created a new topic Launch: Local AI trainer for Stable Diffusion
Announcement:
Mashed Mice Fine-Tuner for Stable Diffusion is now live!
This is an SD fine-tune trainer that runs local on your own device, training whatever dataset you've collected.
I've added loads of settings to it, and you'll not have to write a single line of code running it - its functions are fully integrated in the user interface!
You'll only need Windows as your OS, a not to slow processor and a CUDA compatible GPU (16 BG VRAM is good enough for a batch size of 3 running a 512x512 dataset).
The SD model will gradually adapt to the data You feed it with.
Link: https://mashed-mice.itch.io/mashed-mice-ai-trainer-for-stable-diffusion/devlog/9...
----------------------------------------------------------------------------
Finally version 1.0 is live!
Please take your time reviewing this tool - it is the first tool I openly publish to the world, so I'm quite exited about it.
The launch is the resault of over a month of late nights after my regular job, and it's been taking alot of my focus to crack some of the technichal nuts associated with this type of projects.
However, it's now live, and as far as I can see the trainer now seem to cover every aspect of what a SD trainer may need; advanced memory handling, VAE settings, attention slicing, visual and live training tracking with live updates on the loss graph, batch info and a collection of samples along the way.
The model also save in set intervals, so it's fully possible to backtrack in case it starts to overfit, making logs with data needed to pick up the training from where it left of in case its aborted by whatever cause.
Yes, I'm quite happy to announce the launch of the trainer.
I'm also working on a Dataset prepper tool for big batch making with the need of industy-scale cropping, naming, sorting and clearing of datasets, but that is for the future.
Anyway, I hope You all enjoy this first version of Mashed Mice Fine-Tuner for Stable Diffusion.
Over and out!
Joel Sandström, Developer
