DCGAN is initialized with random weights, so a random code plugged in the network would produce a very random picture. However, as you may think, the network has countless parameters that we can easily tweak, and the goal is to locate a setting of those parameters which makes samples generated from random codes appear like the coaching facts.Mouser