The rapid growth of generative models and deep learning has raised concerns about "deepfakes," which are synthetic media that closely resembles real-world information. These media assets are subject to restrictions in a variety of fields, including entertainment, politics, and safety. As a result, there has been a lot of interest in designing systems to detect deepfakes. Numerous experts have developed a wide range of two-classification-based algorithms for identifying Deepfakes. This article gives a thorough examination of the most current advances in deepfake detection techniques. It gives a thorough examination of the underlying theories, the datasets utilized for training and testing, and the current issues that this rapidly developing profession faces. Significant scholarly work has been committed to investigating alternative techniques to tackling the issue that Deepfake raises. To assist research, the methodologies are divided into four categories: multimodality-based DFDT, video-based DFDT, image-based DFDT, and audio-based DFDT. Deepfake photos and movies were shared on social media sites using cutting-edge techniques that we developed. Scholars and researchers specializing in this discipline will be captivated by these methods. Several datasets, such as FaceForensics++, the DeepFake Detection Challenge (DFDC), and Celeb-DF, are used to test and improve deepfake detection systems. Discuss the model's features, adjustments, and constraints, highlighting the importance of using a wide range of real-world datasets to ensure its applicability across multiple situations.