Weight of around 0.7- 1 usually works well. It still struggles with men a bit, but it can do them. Note: If you are used to using my sliders, this is not that. If you slide this up to 5.0, you are going to have a bad time :) This is an experiment for a targeted training method I am working on where you edit 2 images in opposite directions and train on them. It does not overfit and targets only your specific changes. Eyeliner has always been difficult to do without changing the scene or person, so I thought it would be a perfect test case. This dataset was only a few manually edited images trained for 3k steps. It is not perfect, but does a pretty good job of adding eyeliner without altering the character or image much. If you appreciate my work, I am not asking for coffee, but a kind review is always appreciated.

Files with SHA256: 71ef2df5990e7670ba1ba44c0fb41df7c2f1a5c548db8516830ee939eee3399a