I've been dabbling a bit and I'm enjoying and improving each go-round. The limitation of similar face-shapes has been a bit of a thorn in my side however.
In my first attempt one face had a pronounced chin (kinda circular with certain facial expressions) and the other has a barely visible cleft. The chins were quite distinct and, because I had trained with only 75% coverage on the "Face", was not included in the swap. A decided to train again with 100% coverage on "Legacy" which usually included the chin and forehead. The results looked like they might be better but the chins still didn't really work.
I started to wonder if the chin was not on the border of the input images and I manually included the necks in the mask if the NN would resize the chin like it does other features. I have 20,000+ images to manually edit before I give it a go and I thought it might be best to ask if it might even work here before I tried.