If I were going to try for this, I would first try YiffyMix for the first pass, then see if I can get good results using a photorealistic model on a second pass with Hires fix or img2img. If that doesn't do it, I bet that with regional prompter could make it work. I'm not as good with that though, so I usually only resort to it if I can't get what I need another way.
e: I'm talking local stable diffusion here, usually A1111
When I was in college I told my girlfriend's dad I was a Marxist and he "quoted" someone on the topic (which I'm pretty sure was made up on the spot):