Unlike the subjects of images, backgrounds are almost never captioned, so models have to do what they can with alignment, depth, and perspective all on their own.