Main goal of this LoRa is to enable generation of multiple yordle characters at once, while improving adherence to their designs, simplifying prompts and remaining as stylistically neutral as possible.
As a side effect can be used to enhance the proportions of any shortstack character.
v1 has some minor issues that I'll probably fix in future versions.
Also planning to add male yordles as well.
Trained on 2500+ manually picked, manually tagged images. (including regularization)
Enjoy the yordles ;)
Description
Basic version with 4 yordle girls and regularization.
Also trained on alt versions (like Noxus Poppy or Star guardian/human Lulu) and with some associated tags (like "pix" for Lulu's fairy), but they don't work well.
Tends to give earring for Lulu, tattoos for Vex and some other random shenanigans that are easily fixed with negatives. ...
Sometimes adds a fifth finger to yordles.. (YUCK!)
FAQ
Comments (3)
"Trained on 2500+ manually picked, manually tagged images"
You must really love yordles :)
Been maniacally "building the dataset" since way before there was a use for it :)
Wow, I'm impressed. This is awesome good job, this can even make OCs.