DCGAN is initialized with random weights, so a random code plugged to the network would generate a completely random picture. However, while you may think, the network has countless parameters that we could tweak, and also the purpose is to locate a environment of those parameters which makes samples generated from random codes seem like the teachi