This paper introduces MesoNet, a compact deep learning network designed to detect face tampering in videos, specifically focusing on two recent techniques: Deepfake and Face2Face. Traditional image forensics methods are often ineffective for videos due to the degradation caused by compression. MesoNet employs a deep learning approach with a low number of layers to focus on mesoscopic properties of images. The paper evaluates two network architectures, Meso-4 and MesoInception-4, on both an existing dataset and a newly created dataset from online videos. The results demonstrate high detection rates, with over 98% for Deepfake and 95% for Face2Face. The paper also discusses the robustness of the method to video compression and provides insights into how the network identifies forged faces, highlighting the importance of detailed features like eyes and mouth. The authors conclude by emphasizing the practical and theoretical contributions of their work, including the creation of a dataset for Deepfake detection and the visualization of network layers to enhance understanding and future improvements.This paper introduces MesoNet, a compact deep learning network designed to detect face tampering in videos, specifically focusing on two recent techniques: Deepfake and Face2Face. Traditional image forensics methods are often ineffective for videos due to the degradation caused by compression. MesoNet employs a deep learning approach with a low number of layers to focus on mesoscopic properties of images. The paper evaluates two network architectures, Meso-4 and MesoInception-4, on both an existing dataset and a newly created dataset from online videos. The results demonstrate high detection rates, with over 98% for Deepfake and 95% for Face2Face. The paper also discusses the robustness of the method to video compression and provides insights into how the network identifies forged faces, highlighting the importance of detailed features like eyes and mouth. The authors conclude by emphasizing the practical and theoretical contributions of their work, including the creation of a dataset for Deepfake detection and the visualization of network layers to enhance understanding and future improvements.