This exemplify one of the biggest challenges with those type of models. Consistency over multiple outputs (look at the robot... it's the same character but looks different in every image)