DCGAN is initialized with random weights, so a random code plugged in the network would deliver a very random image. Nonetheless, while you may think, the network has a lot of parameters that we can easily tweak, and the objective is to find a setting of these parameters which makes samples created from random codes appear to be the training data.f