The rapid advancements in deep learning and computer vision technologies have enabled the creation of highly realistic AI-generated content, popularly named as deepfakes. These false media, often faces and expressions have been a major ethical concern leading to seepage of security since many could be influenced which in turn will affect public opinion. Deepfakes have been used in identity theft, misinformation, and even political manipulation. To combat these issues, substantial research has been directed toward the generation and detection of deepfakes. This paper offers a thorough review of the existing deepfake generation techniques, including face swapping and expression swapping, primarily relying on deep learning models such as generative adversarial networks (GANs). It also explores publicly available datasets used to train detection models and evaluates various deepfake detection approaches. The discussion highlights the major research challenges in deepfake detection techniques and proposes future research directions aimed at developing more robust and accurate detection systems. The existing approaches result in promising performance in case of unknown or generalizations attacks. Further, the detectors face challenges due to sparse, incomplete, and noisy training data. By examining the intricate balance between the generation and detection of deepfakes, this study contributes to the ongoing efforts to safeguard digital content authenticity.