Butt Plug v1.1

I was surprised to find that there is no such thing on this site yet, so I decided to take a swing at it myself.

The results are a mixed bag. I am surprised that the model has learned anything at all, especially how the object in question should look and where it should be located relative to the body. Sometimes it does quite well, but it also tends to noticeably worsen overall quality of the picture (seems that the LoRA learned much more than what I intended). If you have a lot of patience you can fix its shortcomings to some extent by using inpainting.

This is my first experience with LoRA training, and I'm sure I've made a lot of stupid mistakes that could have been avoided. I look forward to your comments and suggestions on how this model can be improved.

I'll also be glad if my example inspires someone else to try to achieve better results.

I want to do some more experiments with this concept when I have time, in particular I think about trying to:

• improve the quality of the dataset some more

• manually verify and fix all generated image descriptions

• prepare a regularization dataset

• try using constant LR