DCGAN is initialized with random weights, so a random code plugged in the network would create a very random impression. 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 that makes samples created from random codes look like the training data.T