I trained this LORA on a subset of his work and captioned this subset with <code>waterkuma</code>. I selected even smaller subset of artwork that felt closest to zenless zone zero style and captioned it with <code>zezoze</code>. I wanted `waterkuma` and `zezoze` tags to have slightly different style, check the test on Corin in the gallery below. In my opinion it's a very subtle difference. Some of the most popular tags are: twintails, smile, gloves, jacket, jewelry, shorts, socks, shirt, navel, hair ornament, earrings, motor vehicle

[Illustrious] Waterkuma style (Zenless Zone Zero)

Increased uniformity of the dataset. Changed training settings to eliminate overfitting and added more steps to the training.

Files with SHA256: 32e21d1b07170fd1a1cf4bedac8251988fae35d15b385c3f603b32a8efc75eba