DCGAN is initialized with random weights, so a random code plugged in the network would generate a very random picture. On the other hand, while you might imagine, the network has millions of parameters that we will tweak, and also the purpose is to locate a placing of these parameters which makes samples generated from random codes appear to be th