DCGAN is initialized with random weights, so a random code plugged to the network would generate a totally random graphic. Nevertheless, when you might imagine, the network has millions of parameters that we could tweak, and also the target is to find a placing of those parameters that makes samples produced from random codes appear to be the coach