Visual Turing test is not sufficient to evaluate the performance of medical generative models.