A style LoRa trained on a dataset of cute drawings (mostly furry/kemono) I've collected over the years, with emphasis on outlines, expressiveness, and appealing colors.
This LoRa also serves as the foundation for uwumerge.
Description
Refactored the data set by removing tags which only had one use (over 500) and then manually adding back in missing species and stuff. Also removed use of trigger words, and replaced images that I felt influenced the style too much. Trained over multiple resolutions from 512 to 1024.
FAQ
Comments (2)
These have been amazing!
Thanks! tbh I'm really just building these models to try out new training and merging strategies haha