Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{195639,
        author = {G. Divya and Karra Saketh Reddy and Kunooru Rahul and Kampati Nithin Kumar},
        title = {Detecting Deepfakes Using Advanced Deep Neural Networks},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {997-1004},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=195639},
        abstract = {DeepFake videos appear on a steady platform even more rapidly than one might have anticipated and such are nearly surreal in their realistic appearance. I have even seen a few of them unwind further through time and it becomes a little hard to believe your eyes that what you think is true is really true. These are synthesized videos that are based on deep generative techniques, GANs, autoencoders, diffusion-based manipulations, and faces and voices are reshaped with an accuracy that is uncomfortable. Conventional detectors constructed on stationary design or old filters explode soon. They mark the prevalent ones but overlook those that are cleverly coded and manipulate light, movement and talk to the extent that older systems do not notice. This paper uses multilevel detection architecture with deep learning components that neither operate independently of one another. A ResNeXt backbone pecks at spatial textures - small details such as misshapen pores or an imbalanced light which DeepFake models tend to flounder to cause them. It is used together with a frequency block that captures small ripples that are not noticed by the human eye, signals that slip into Fourier space with GANs when upscaling or it carries out frame merging. The step of Vision Transformers comes next, which scan relationships across remotely located parts of the face, and identifies mismatches in context which recurs in CNNs but are not caught. Then the LSTM unit observes an array of frames, and they determine that motion of expressions is natural or making a jerky rush like stitched tape. The combination of these portions acts not so much like one classifier, as like a stratified observer.},
        keywords = {DeepFake Detection, ResNeXt CNN, Vision Transformer, LSTM, Frequency Analysis, Hybrid Deep Learning, Media Forensics.},
        month = {April},
        }