DCGAN is initialized with random weights, so a random code plugged in the network would produce a very random image. Nonetheless, as you might imagine, the network has millions of parameters that we could tweak, along with the goal is to find a placing of such parameters that makes samples generated from random codes appear like the instruction det