Background and Motivation: The rapid advancement of Deep Learning (DL) models has enabled the generation of extremely realistic synthetic multi-media content, commonly referred to as Deep fakes. While these technologies offer applications in education, entertainment and research, their misuse poses important threats to public trust, security, and privacy. Key Contribution of the Survey: This review focuses on state-of-the-art DL based DeepFake detection models across images, video, audio and hybrid modalities. In this review, the analysis critically evaluates the detection techniques, benchmark datasets and their performances, highlighting current limitations such as high computational cost and generalization gaps. It summarizes the state-of-the-art models, including Convolutional Neural Network (CNN), Generative Adversarial Network (GAN), transformers, and forensic signal analysis. Challenges and Future Directions: Despite important progress, challenges such as generalization across datasets, robustness to emerging deepfake generation techniques, model interpretability, and real-time deployment persist. This review highlights the need for self-supervised learning, adversarial robustness, cross-modal fusion, lightweight models for mobile devices, and enhanced explainability. Review Contribution: By offering a structured synthesis of current DeepFake detection research. This review delivers a detailed evaluation of state-of-the-art deepfake detection techniques, including architecture insights, practical detection tools, and benchmark performance analysis. It aims to provide researchers with a comprehensive foundation to enhance deepfake detection systems and build more secure and trustworthy digital ecosystems.
Singh et al. (Sat,) studied this question.