Scaling items in text-to-image

Is the text-to-image model aware of the size and scale of objects within the resulting image? For example, my request: "Create a modern room with casual furnishings, on the primary wall create an empty rectangle to-scale, 20 inches wide and 16 inches high." The results were all pretty good, except for the rectangle. In each response the overall size of the rectangle was not close to 20x16 inches, and in some cases it wasn't even a rectangle (I remember one result clipped off a corner at the point when light through a window crossed the rectangle). For my own education, is there a place where I learn how to best phrase what I want in the result? Thanks!