Hi.
SinisterSh0t initially made the dataset and I spent some time trying to make headway into it. Unfortunately, all I got is a CTF model that doesn't really like to behave that well.
Anyway, here are 2 models. 1 is called "AutoV2" which has automatic tagging with some additional pruning while "Manual" is original captions from SinisterSh0t.
I think the models are too undertrained? Will have to experiment more later.
N.B. Chisato is an inpaint image. I think it works for inpainting quite OK?
Description
This model was trained Noob-XL VPred 0.75.
Final Epoch. This I think is the best among the 3 I've checked.