I tried to train it on different models but it turned out to be something in between, not very flexible LoRa, possibly due to the lack of high-quality images, although perhaps the problem is in the models themselves

Training on the yiffymix_v37 model

Recommended value from 1 - 0.8