Specific Layer Training at transformer.single_transformer_blocks.(7|9|20|25).,network_dim 128
This LoRA seems to be overfitted. Please adjust the weights accordingly. For generating patterns, 1000 steps‘ version is better.