The proliferation of fraudulent images and movies on social media is a quickly growing problem. Using commercial media editing software, anyone may make phony images by cropping, adding, or cloning people and objects. Fake content creation is not ethical and detrimental to society. These days, deepfakes are frequently employed in financial fraud, fake news, cyber extortion and identity theft videos with false celebrity obscenities for blackmail, and many other cybercrimes. Every day, new attempts are made to identify conventional counterfeits, despite the fact that numerous ways have been presented for doing so. Image-to-image translation, which makes use of generative adversarial networks (GANs) and allows for extremely realistic contextual and semantic image change, is one of the riskiest methods. Unprecedented achievements in picture production have been made possible by the generative adversarial networks’ (GANs) quick development. The advancement of BigGAN, StyleGAN, and other sophisticated GANs creates more deceptive and realistic generated images, putting personal privacy, society stability, and national security at risk. Creating a useful methodology for identifying deepfake content is our primary objective. The deepfake dataset, which contains both genuine and fake faces, is used to create neural network algorithms. Xception and Dense Net are the two deep learning architectures that are utilized in comparison. The recall, precision and F1 score metrics are used to assess the performance of the suggested algorithms.