DCGAN is initialized with random weights, so a random code plugged to the network would produce a completely random impression. Nonetheless, when you might imagine, the network has a lot of parameters that we are able to tweak, plus the objective is to locate a setting of those parameters which makes samples generated from random codes appear like