props on including training data.

It looks like you used multiple copies of the same images. That probably won't help your results? Someone can correct me if I'm wrong.

But we can see, a character LoRA made with AI generated images with plastic-y skin will result in a model that generates just that.