DCGAN is initialized with random weights, so a random code plugged into the network would deliver a very random impression. Nonetheless, as you may think, the network has millions of parameters that we could tweak, as well as the purpose is to locate a setting of such parameters that makes samples generated from random codes seem like the training