In case of solving many classification problems, various machine learning models are used such as: KNN classifier, Decision Tree, Random Forest or even SVM but not only. In classification problems neural networks also have found broad applications, these networks help recognize and analyze photos. Additionally, they are utilized likewise to anomaly detection, such as detecting objects that differ significantly from the rest of the data in the data set. Generative Adversarial Networks deal with processes like: improving photo resolution, changing their colors, or creating typically new images. In this paper we will look at three different configurations of DCGAN models, especially at various types of generators and we will analyze the facial photos that were generated by each creator. We will show that one of the models that produced the most realistic images included downsampling and upsampling layers in the generator. Its architecture resembles that of a simple autoencoder. In addition, we will compare how every DCGAN architecture coped with producing a selected photo that was only in our collection of real photos.