Deepfake generation technology has advanced rapidly with the development of artificial intelligence, generative adversarial networks (GANs), diffusion models, and neural rendering systems. Modern tools can create highly realistic manipulated images, videos, and speech that are often difficult for viewers to distinguish from authentic content. While these technologies have valuable applications in entertainment, filmmaking, virtual reality, and education, they also pose significant risks, including misinformation, identity theft, cybercrime, political propaganda, and societal manipulation. As deepfake creation becomes easier, faster, and more accessible, the need for reliable and scalable detection methods has become increasingly urgent. This work presents Beyond the Pixel, an end-to-end deepfake detection framework that combines neural-network-based face analysis with modern web deployment technologies. The proposed approach employs a face-focused workflow in which faces are detected and cropped using Multi-task Cascaded Convolutional Networks (MTCNN), followed by classification through an EfficientNet-B7-based model, DeepFake Classifier, trained on the Deepfake Detection Challenge (DFDC) dataset. By concentrating on facial regions rather than full frames, the method targets manipulation artifacts commonly found in skin textures, boundary regions, eye behavior, and facial expressions. The framework supports both image and video analysis. Images undergo face cropping, normalization, and EfficientNet-B7 inference to generate confidence-based predictions. For videos, every fifth frame is sampled and analyzed individually, with frame-level predictions aggregated into a final confidence score. The system is implemented using FastAPI and PyTorch on the backend and React with TypeScript on the frontend. Results indicate that face-oriented pre-processing improves inference robustness by reducing background noise and aligning inputs with DFDC-style training data. The study also demonstrates practical deployment through scalable APIs and discusses implementation details, limitations, and future enhancements, including interpretable AI, live-stream analysis, uncertainty quantification, and cloud deployment.

{{conclusion}}

References [1] Tan, M., & Le, Q. (2019). EfficientNet: Rethinking model scaling for convolutional neural networks. Proceedings of the International Conference on Machine Learning (ICML). [2] Rossler, A., Cozzolino, D., Verdoliva, L., Riess, C., Thies, J., & Nießner, M. (2019). FaceForensics++: Learning to detect manipulated facial images. Proceedings of the IEEE International Conference on Computer Vision (ICCV). [3] Dolhansky, B., Bitton, J., Pflaum, B., Lu, J., Howes, R., Wang, M., & Canton Ferrer, C. (2020). The DeepFake Detection Challenge (DFDC) dataset. arXiv preprint arXiv:2006.07397. [4] Seferbekov, S. (2020). DeepFake detection challenge solution [GitHub repository]. [5] Zhang, K., Zhang, Z., Li, Z., & Qiao, Y. (2016). Joint face detection and alignment using multi-task cascaded convolutional networks. IEEE Signal Processing Letters, 23(10), 1499–1503. [6] Buslaev, A., et al. (2020). Albumentations: Fast and flexible image augmentations. Information, 11(2). [7] Paszke, A., et al. (2019). PyTorch: An imperative style, high-performance deep learning library. Advances in Neural Information Processing Systems (NeurIPS). [8] Bradski, G. (2000). The OpenCV library. Dr. Dobb’s Journal of Software Tools. [9] Facebook AI. (n.d.). DeepFake Detection Challenge dataset. https://ai.facebook.com/datasets/dfdc/ [10] React. (n.d.). React documentation. https://react.dev/ [11] FastAPI. (n.d.). FastAPI documentation. https://fastapi.tiangolo.com/ [12] Timesler. (n.d.). facenet-pytorch documentation. https://github.com/timesler/facenet-pytorch

© 2025 International Journal of Engineering and Techniques (IJET).

Digital Object Identifier (DOI)

IJET assigns a unique DOI to every accepted article via Zenodo for persistent linking and citation. Your article’s DOI: https://doi.org/{{doi}}

Open DOI About Zenodo DOI

Close