Please click on the following links to view the results for the particular 3D scene. Note that in this visualization (fixed camera w/ varying seed), more flickering reflects higher diversity. We present two variants of our full method titled Ours and Ours_6 which are trained for 7 and 6 number of pyramidal stages respectively. Ours_6 has much higher variance at the cost of slight loss in quality, while Ours has high quality (realism) but slightly less variance.