The rapid advancement of Deepfake technology poses considerable challenges to the preservation of digital media integrity. Several deep neural network architectures have been proposed. In this paper, a performance of well-known deep neural network architectures used in Deepfake detection is described. Models compared are ResNet, VGG, Xception, EfficientNet, MobileNet, and DenseNet. The efficiency of each model was assessed using performance metrics, including recall, accuracy, and precision. Additionally, the processing time was evaluated to determine the speed at which each model can detect Deepfakes, as this is crucial for real-time detection scenarios. Our findings suggest that although all architectures perform well in identifying Deepfakes, certain models outperform others in terms of some metrics.