Exposing Deepfakes: A Comprehensive Approach to Video Authenticity Verification via Uploads and URLs

Deepfakes are becoming more lifelike and realisticin terms of synthetic videos created using deep learning architectures like Generative Adversarial Networks (GANs) and autoencoders. Whiletheycanhavepositiveapplicationsinmedia production and education, the potential misuse of deepfakes raises serious risks regarding misinformation, identity fraud,anddigitaltrust. In this work, we present a deep learning based framework for the detection of deepfake videos using an integrated ResNeXt-50 Convolutional Neural Network (CNN) for spatial features, and a Long Short-Term Memory (LSTM) network for temporal sequence modelling. The proposed system employs video uploads or forensics based on URLs to allow flexibility for real world deployment. Experiments on benchmark datasets (FaceForensics++, Deepfake Detection Challenge (DFDC), and Celeb-DF) demonstrate the effectiveness of the proposed approach which achieves high detection accuracy and robustness to diverse manipulations. The proposed approach also generalizes well to unseen forgeries and offers confidence scores for interpretability. This work contributes to developing scalable and transparent approaches to mitigate the spread of synthetic media on online platforms.