I have increased the image count in the training dataset to 6,000. To retain this information capacity, we have significantly expanded the Network_dim to 256 and Conv_dim to ??. Although the size of LoRA has been proportionally increased, the response to prompts has improved.