Welcome back!
October 7th 2025 - Continue using 22-may-2025's preset, it's the simplest thing to use, you'll want to of course like before and edit the incorrect settings from the preset, it's simple and quick, sorry about needing that but it's because it's dumb dumb doo-doo brains in how it loads and saves the settings. I'll add a new zip for october that has the appropriate screenshots and the .toml file
July 6th 2025 -The attached zip of Malebolgia preset 06-Jul-2025 this time around has an 'inaccurate' toml file attached dated back from 22-may-2025, but it's the best I could set it with it keeping things like the alpha dims, block dims and block weights set properly as that part often breaks when saving the preset, and that's the worst part to have to set up, so I'd say use that preset because from there it's just about 1 minute to set up the preset to follow the screenshots. The screenshots are how the file should be set, do not rely on any of the toml or .json files in any of my 'preset eras'. This Page will also likely be updated soon as I'll try to figure out how to make it work properly with characters, as I feel it's more useful for styles than characters.
The config files are plug and play with:
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts?tab=readme-ov-file#installation
Assuming it loads them properly, which on my PC it seems to be reluctant to do so for every value, so you'll likely have to manually plug in some values.
You can also open those toml and json files with notepad or some variant of such word processor to see the data and preset information contained in them. And a warning to anyone who chooses to use EasyTrainingScripts, for some reason whenever I close the thing and open it later and load my presets parts, while the ARGS parts load fine, some stuff doesn't seem to want to load and must be manually edited, thankfully you can open the toml and view in notepad if you can't remember or don't know what goes where.
Here we go
So for number of repeats, I'd say try to aim for 15-25, I'm comfortable with 15-20 right now for accuracy for datasets around 50-60 images, but fewer images I plug it a bit higher to make sure things stick.
