This is my first attempt at training a lora, and its probably pretty bad. I took about 2000 images that I felt were the best quality representations of 'cute' and slapped them into a trainer. The results were actually surprisingly not the worst thing I have ever seen. This seems to work best at nudging outcomes to the cuter side at around half strength. Closer to full strength often pushes outcomes to a more 2D art style, as well as sometimes a bit more oddness.