So It seems the only way to get consistent AI transport interiors, is to train a Lora with alot of images of basically the same angle (different Loras for different angles, I guess) The more you know... -Anyway, here is one of my tests, The Iconic Japanese 'Box Seat' booth, you can prompt the upholserty to the color of your choice, and optional 'wall has a wood finish' prompt works too :P. Most importantl;y you can prompt who ever sitting in them (Scaling seems about right) and prompt what ever you want to be outside the window, Even at Strength:1 it will still help to pompt "Side view" and 'opposite facing seats', to remind Flux not to give you AI transport interior salad. It Seems to work ok with Anime styles but its not trained for it. Best rendered to Landscape or Square orientation, portrait starts to squish those seats uncomortably close! Enjoy.