Deepfakes have recently become a powerful but also dangerous tool if used in the wrong way. In this project, which represents my Bachelor Degree's thesis, we propose the implementation of a variational autoencoder (ResNetOCVAE) based on the ResNet-18 architecture to perform a task of Deepfake Anomaly Detection on a dataset of images. This model will let us obtain better results with respect to this paper. More details here