With the increased search for deepfakes, society and its stakeholders, institutions, governments, and even democracy come under threat. It could be false political addresses to fake corporate announcements, and such reasons make deepfakes a threat in that they will reduce confidence in digital media. Thus, researchers apply AI and ML to create systems that may detect deepfakes as effectively as possible. These detection systems are critical as they act as data guardians in today’s complex environment, combat fraud, guard privacy, and counter fake news. This review provides an overview of several cutting-edge deep learning-based methods for the recognition of human actions on six different kinds of CNN models. This work presents the history and current state-of-the-art CNN architectures with a thorough investigation of the model’s performance in image classification tasks. This paper aims to provide an analysis of the prevailing AI-based approaches for detecting deepfake-generated data. Thus, within this research, various machine learning models with a focus on CNN are demonstrated to learn the difficulties of detecting deepfakes and the shortcomings of existing solutions. This work will provide the groundwork for future discussion and analysis of the findings and how they affect the choice of the best models for deepfake detection.