Deep Fake Detection: Using a web Based Convolutional Neural Network System

Abstract – The rapid proliferation of deepfake technology poses significant challenges to digital media authenticity, necessitating robust detection mechanisms. This paper presents a web-based deepfake detection system developed using Flask, TensorFlow, and OpenCV, designed to classify uploaded videos as "REAL" or "FAKE" based on a pre-trained convolutional neural network (CNN) model. The system preprocesses video frames to a standardized (None, 128, 128, 3) input shape, leveraging a single-frame analysis approach for real-time classification. Key challenges, including model compatibility, input shape mismatches, and prediction biases, were addressed during development. Preliminary results indicate successful deployment on a local server, though limitations in model generalization were observed, with all test videos classified as "FAKE." This work highlights the feasibility of web-integrated deepfake detection and identifies areas for future enhancement Deepfake detection has evolved alongside generative technologies. Early methods relied on visual artifacts, while modern approaches leverage deep learning. Li et al. proposed CNN-based detection using frame-level features, achieving high accuracy on datasets like Face Forensics++. Rossler et al. introduced the Face Forensics++ dataset, pairing real and manipulated videos, with Caption models expecting (128, 128, 3) inputs—similar to our system’s model. Key Words: Synthetic Media , Video Manipulation Detection Deep Fake , AI- generated Content , Deep Neural Network