ARCHIVES

Deep Fakes are persuasively real fake media wherein a human's face or voice is manipulated through deep learning-based techniques. As they provide pioneering opportunities in industries such as entertainment, education, and accessibility, Deep Fakes also pose noteworthy threats to information misinformation, identity deception, politics manipulation, and privacy violations. This paper provides a Deep Fake Detection System with the capability of detecting manipulated video content with precise accuracy and repeatability. The system utilizes a hybrid deep learning methodology that combines the extraction of spatial features using ResNeXt with the exploration of temporal dynamics through Long Short-Term Memory (LSTM) networks. It analyzes videos frame by frame, identifying both static visual signals and temporal facial aberrations. The backend is designed with Flask to deploy the trained model, while the React frontend facilitates smooth video uploads and real-time user engagement. The system produces a binary prediction—real or fake—and confidence score, with end-user interpretability. With the combination of spatial and temporal features, the model improves detection accuracy and generalizes well across different Deep Fake datasets. The solution is designed to assist digital forensics and maintain media authenticity in an age where synthetic content is becoming increasingly prevalent.