Trying to use tf.constant as loss value results in gradient returning None's, better way to do it?

Question

I am trying to make my own GAN for a simple game AI after some limited (but successful) experience using tf keras for image related functions with a more standard CNN. In ma