Lu Xueqi - A Chinese Ghost Story

The training bottom model used is:v1-5-pruned.ckpt

The above image shows the comparison between the weight opened by Lora and its corresponding effect.

Recommended parameters:

1. Recommended Lora weight:0.6~0.8

2. Clip:2

3. VAE :vae-ft-mse-84000-ema

4. Recommended high-definition restoration and enlargement algorithm:R-ESRGAN 4x+ Anime6B; Recommended redraw amplitude:0.2~0.4

5. Recommended sampling method:DPM++2M Karras

6. Sampling steps:20-30

7. When generating a full-body image, it is recommended to enable Adetailer plugin for repairing the face and hands to ensure normalcy.

8. Welcome to provide feedback on the images!