Good work! But I can see that the model often produces curtains in the background of the image since you have to use "breast curtains, pelvic curtain" in the prompt. You might be able to work around this by adding "curtains" to the negative prompt or otherwise prompting a background in which curtains would be unlikely to appear.

If you wanted to make another version of the LoRA to address this issue (and if you have enough training images), you could try deleting the tags "breast curtains" and "pelvic curtain" altogether from the training images. The resulting LoRA would probably have less control over the two individual parts of the outfit, but it would probably still produce the entire outfit (both pieces together) without needing to use prompts involving the word "curtain".