This is an updated version of Chinese Luka Realistic LORA model which previously made for PonyXL. Uses same dataset as before so no significant changes of it. Should be compatible with most of IllustriousXL models including its merged models. All image examples are generated with Stable Diffusion DirectML with Nvidia RTX 3060 12GB.

Sampler: Euler A (recommended), others can use as long as it doesn't generate garbage image

Sampling Steps: 30

CFG: 7

Clip Skip: 2

Strength: 0.7 - 1

Use Adetailer Extension when necessary to fix distorted image. For Comfy, refiner method is likely needed to improve and fix images.