Based off screenshots from the game She Will Punish Them. Training used the CivitAI defaults.

The tag "3d render" was used for training and you can turn down the weight on it or add it to the negative prompt to help make the character work better with other style loras. This works better for v1 than v2 where it will kill alot of the detail on the character, but it still works in some cases.

v1

use prompts: white hair, pony tail, glowing eyes, orange eyes, large breasts, pink-skinned female, toned female body, demon wings,

revealing armor, high heels,

I didn't remove the background during training so it has a tendency to want to recreate it.

Suggested negative tags that can help with preventing the background:

gothic stone archway, intricate stonework, dimly lit stone walls, dark and mysterious atmosphere, large arched window, purple light filtering through window, ethereal glow, dim ambient light, antique red velvet chair, ornate rug,

V2

v2 uses some additional training images and changed the tag "revealing armor" to "demonika armor"

This version didnt come out like Id hoped, its a little more refined and detailed but the background still comes through often and its seems less compatible with some style loras.

The swpt_red_succubus tag actually lowers the quality in many cases, Im not sure why thats happening. Ill leave it however since it helps in some cases.