Oh man…

That is the point, to show how AI image generators easily fail to produce something that rarely occurs out there in reality (i.e. is absent from training data), even though intuitively (from the viewpoint of human intelligence) it seems like it should be trivial to portray.