DCGAN is initialized with random weights, so a random code plugged into your network would generate a very random impression. Nevertheless, as you might imagine, the network has millions of parameters that we can tweak, as well as the target is to find a placing of such parameters that makes samples generated from random codes look like the coachin