Synthetic Video Integrity Analyzer using Vision Transformer (ViT)

This project presents a web-based application for deepfake video detection using the Vision Transformer (ViT) model. The system, developed with Flask and TensorFlow, allows users to upload videos, which are then processed to extract frames using motion detection techniques. Each extracted frame is analyzed using a pretrained ViT model to classify whether it is real or fake. The application incorporates temporal smoothing to improve prediction accuracy by averaging predictions over several frames. Confidence scores are used to weight predictions, ensuring reliable results. Users can track their detection history, including video verdicts and probabilities, in a secure user dashboard. The system ensures smooth user authentication, with the option to sign up, log in, and manage their profiles. This solution provides an efficient and user- friendly approach to combat the growing concern of deepfake videos.