Ouyang Ziyu 欧阳子虞 - LoRa

This is the LoRa version of my previous Dreambooth model, which uses essentially the same dataset with some tweaks and better tagging.

You get better results with this LoRa than with the Dreambooth model.

It was tested with various popular checkpoints, optimal Lora strengths range from 0.5 to 0.9.