Ok, I’ve got to grips with things now but unless I’m missing something obvious the face recognition will make a good job of a rectangle, but leave hair and neck outside the rectangle untouched. I can sharpen all but it never matches the hair that’s been improved by the AI. Surely there must be a way of adding to the face recognition area? It’s a real shame because other than that I’m really pleased.
That is a common problem with all face enhancement programs I have tested, such as GPEN, GFPGAN, CodeFormer, and VQFR. They all suffer from the same issue due to their face enhancement algorithm. First, they detect the location of the faces, crop out a tiny square surrounding the face, resize it, enhance it, and then paste the face back.
I’m not sure how TPAI/Gigapixel works for face enhancement, but I assume they are heavily influenced by those open-source face enhancement programs.
