DCGAN is initialized with random weights, so a random code plugged to the network would generate a totally random graphic. On the other hand, as you might imagine, the network has millions of parameters that we will tweak, and the purpose is to locate a environment of those parameters which makes samples generated from random codes appear like the