I'm not good with realistic LoRa, this burns my credit too much.
I think this Lora overfitting
recommend use at 0.4 to 0.8
0a9acfb2-63ca-4cf4-914e-d3eec5dca734.TA_trained.safetensors