My first try, not exactly what I wanted, but it can do something with proper prompting.

White skin seems to be very unstable, but sometimes it can work if you play with strength and the consistency lora. Pale skin might give better results even if it's not in tagged in the training data. But if it fails too much, might be better without either of them.

You might also need to prompt the expression when editing, to keep it more accurate.

Lora strength: 0.6 - 1.4

Main Trigger: g0thmkup,

Helpers: gothic makeup, eyeliner, white skin,

Lipstick: black_lipstick | dark_purple_lipstick | dark_red_lipstick

Hair: long_black_hair | long_black_hair_with_bangs

Piercings: septum_piercing, lower_lip_piercings, nose_piercing,

I did have few pictures in the data set tagged with runny_makeup but that became quite messy.. use at you own risk.

ADD FOR EDITING: keep her face and expression accurate,

This consistency lora helps a lot with editing! (Use it with 0.2 - 0.8 strength)