Turns stylistic/anime/cartoon images into more realistic images. Definitely not realistic, but much more real than originally. The output style is very biased, I created the dataset myself so it was fairly limited.

I iterated through this working on the dataset a bunch of times and this one by far was the winner from all my other methods and attempts. It's incredibly hard to get flux to keep expressions, I was really trying to solve for that but failed. Maybe next iteration I'll try a higher weight.