My first LORA, Japanese idol Anju Kozuki (香月杏珠).

The training data consists of 19 recent images of Anju, captionned through deepdanbooru, then manually adjusted (Some captions removed, others corrected, and others added).

I think it's the best I'll manage with my current knowledge and experience, so I'm releasing it.

It was mainly tested with Asian-focused models, but should also work fine on anime models (See previews). I recommend using it with a weight of 0.7, as that seems to give the best results when changing clothes.