Literature Survey on AI-Driven Frameworks for Deepfake Identification and Temporal Consistency Analysis

DOI

No DOI

Published In

Volume 4, Issue 2 (2025)

Published Online

May-August 2025

Pages

65-68

Article Type

Review Article

Check for updates

No DOI available for cross-reference

Deep Fakes are highly realistic synthetic media objects whose face or voice is substituted for another person's using deep learning. Deep Fakes pose daunting challenges to the authenticity and credibility of digital media, and although they have potential applications in entertainment, education, and accessibility, they also prompt serious issues regarding misinformation, identity theft, political manipulation, and privacy violation. This paper introduces a Deep Fake Detection System that is able to detect fake video content with high accuracy and reliability. The system uses a hybrid deep learning model with Res NeXt for spatial feature extraction and Long Short-Term Memory (LSTM) networks for modeling temporal sequences. The videos are analyzed frame by frame to record both static visual hints and dynamic facial inconsistencies along a time axis. The backend is developed on Flask to deploy the trained model, and the frontend, which is developed using React, provides smooth video uploading and real-time user feedback. The model gives a binary prediction—real or fake—and confidence score, and the results are interpretable to end-users. This system uses spatial and temporal features together, improving detection accuracy and generalizing across different Deep Fake datasets. The solution strives to make contributions to digital forensics and media integrity as the world becomes more saturated with synthetic content.