Usage tips:

Set clip skip to 2

Resolution of 768x768

Use ADetailer instead of face restoration.

It has been trained using the base SD 1.5 model, so it should be compatible with all checkpoints. This is a photorealistic model, but anime checkpoints also work.

Works best with AbsoluteReality, Dreamshaper and Photon.

Samples using AbsoluteReality at different LoRa weights:

This model is not filtered and may produce NSFW.

Description

This is the LoRa version of my first checkpoint ever trained, Yu Lexuan:

This Lora has been optimized to reproduce his most recognizable clothing, a nice blue hanfu. It does a much better job than the checkpoint.

It works with most checkpoints very well. Keep the strength between 0.5 and 0.9 for best results depending on the base model you are using.