Towards Deepfake Detection for Everyone: A Lightweight Deepfake Detection Algorithm (LiDD)
Anoop Krishnan, Amit Ranjan Basu · 2025
Modern AI technologies such as Autoencoders, Generative Adversarial Networks (GANs), and Diffusion Models have advanced Deepfake content generation remarkably over the past few years. The risks associated with malicious deepfakes have motivated significant advances in methods for detecting deepfake content, but the most effective methods require substantial computational resources and processing time. Unfortunately, while deepfake generation does not typically face time constraints, deepfake detection methods are unlikely to be broadly adopted unless they are efficient even on modest computing devices. In this paper, we propose a lightweight algorithm for deepfake detection that achieves impressive performance on various reference deepfaking techniques, even though it utilizes relatively sophisticated deep learning models with over 4 M parameters, and even on modest CPU-based laptop computers. This approach paves the way for integration into public infrastructures like web browsers and multimedia players, advancing global efforts to combat digital disinformation and strengthen the security of online platforms.