This is made from a total 148 images, with only 13 of those being cropped head shots. When organizing data I removed anything with mosaics, or other censors to avoid that being learned by the ai. I also divided the work into what I could identify as two generations of style, sticking to the newer (what I think is newer) style for training. This is about half of the data I was planning on putting into this and is getting an earlier initial release, due to patience. When using this with characters with a highly defined artstyle raise the lora value. I may update this in smaller chunks to mix in the rest of the data that was planned. I did not mix my half hand tagged with the other half auto tagged since I thought that would bring down the quality of the lora.
