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?

Files with SHA256: 12e7e5b0fbf0c0af43e944a48a7444a632f3e855b9751cd4caa6eccef55b5c8e