I tried to train LoRa on the same picture, but on XL model

It shows interesting results on different XL Checkpoints (in my opinion), so I hope this LoRa will help you create beautiful pictures. I'm planning improve this LoRa in the future))

This LoRa showed good results with next settings:

Sampling Method: I more often used Eulear a and DPM++ 3M SDE Exponentional, but you can try different variants.

Steps: 30-50

CFG: 3-7

Clip Skip: 1-3

Weight: on average 0.8