Toshu... on the weekend!

This LoRA is a collaboration between Ripped-Saurian, Deltaphile, and Saur11.

We wanted to research methods for character dataset creation that are efficient and ethical, to help artists who'd like to try something like this without resorting to image scraping. Every single image in the dataset is either drawn from scratch, or made using original line art/flats as a base with a light SD filter on top. Through our research, we also learned that careful 45-degree dataset rotation (manually or through a script) greatly helps the LoRA understand the subject far more than simply flipping each image.

In short, everything used to train this LoRA is completely original or altered with specific artistic intent. We hope the results speak for themselves! Stay tuned for more!