Enhanced deepfake detection through CNN and Deepfake Architecture

Artificial intelligence has gone crazy in such a way that any ordinary man with technical knowledge could, in a sense, be poised to create an artificial deepfake simply by observing facial movements. The more the videos appear, the more realistic they become with these technologies. This presently presents a significant challenge with the rise of fake news, which, although seemingly bad for social networks, is currently one big technological threat in the spread of deepfakes online. Deepfakes pose allsided danger such as manipulation, misinformation, humiliation, and defamation. In this research paper, an overview of mechanisms that go into this approach would be provided by specializing in its ability to detect subtle inconsistencies within the manipulated videos. This project would be taking up the hybrid approach that includes Hybrid CNN and RNN to nullify the threats of deepfakes. This method, therefore, captures spatial features within video frames by CNN and frame-to-frame inspects for temporal inconsistencies by RNN. Audio-video misalignment is also considered in it. This hybrid approach has yielded good accuracy at any one point in the time of 96% with respect to our model, which thus has the capability to very precisely detect manipulated content.