Oh yeah, this feels right.This was trained with Hassaku instead of the usual Anime model (if I programmed the download model thing right) but it uses a rather unique thing that I noticed where instead of tweaking the Repeats, I instead tweaked the number of steps so that it guaranteed 1000 despite the small data size of 12 images. This will explain why you'll see the weird epoch size of 17.However, I think it will work all the same with the same weights (4-8 range), the same baseline activation tags (though I felt like splitting the Nurse Cap and Nurse Outfit in case you want her in something else), and same Negatives that come with my usual LoRAs (or, well, I guess any LoRAs that use the Colab system).I don't think I will be using this "fixed step" thing for the Another Chimera LoRA due to me having come too far into the project to suddenly start doing Version 2s for literally the entire squad (the whole point was to give the minor characters their own LoRA where they end up getting mixed), but I will definitely do this for future LoRAs and even look back at Past LoRAs where I gave them more Epochs, then fix them up with that step method. Maybe even feed more Repeats?I'm rambling. Go ahead and download the nurse.

Nurse (from Pokemon Black & White)

Decided to tweak it with 20 Repeats after some chatting with folks in the Discord about whether it was overbaked.

Files with SHA256: 698175ee8030a8d25baf3e02c692102889e32fafa8921c461fd2dfa621c75af8