I must emphasize that employing the same tags used during training may provoke the LoRA, at a weight of 1.0, to generate images strikingly similar to the training dataset. Nevertheless, it occasionally offers delightful results.

I recommend blending various LoRAs and checkpoints to identify maximum compatibility, and do play with the LoRA weight value; I find the results pleasing even at 0.45.