From what I've seen LoRAs tend to focus on styles, characters, hand fixes, face fixes, clothing. They're quite adaptable.
lora has been on radar. i tried to run locally via confyui, but i don't have a very good GPU available. it takes AGES.
I tried a couple of agents and stuck with open art consistent character. gonna try higgsfield as well. the hardest part is her body. i generate a lot of junk until i get something that looks like her. for consistency in video, i ask veo 3 to change the angle and screenshot for a new frame to video prompt, so it looks like a doc with two cameras.