The dataset was highly edited before training due to having a heavily bland background and wacky artifacts/anatomy, which is regular stuff in Manhwas.
So the character won't be close to accurate nor going to have a decent background compared to anime characters.