Manipulation of images, videos and audios using face edit apps and web services have long been in use, since decades but recent advances in deep learning has led to rising AI generated fake images and videos with swapped faces, lip synced audios and puppet masters, popularly known as Deepfakes. Generated primarily using one of the following two approaches namely, Autoencoders and Generator Adversarial Networks which rests on trained deep neural networks, deepfakes offer unprecedented challenges. The degree of realism achieved by deep learning powered deepfakes increases with increasing amounts of data i.e, fake images and videos readily available on the internet at disposal to train GANs. Deepfake algorithms create media leaving a bare margin of difference between the authentic or original source and the forged or deepfaked targets. Thus, new mechanisms and techniques to detect and filter out such deepfakes is the need of the hour.This paper exploits two powerful deep learning based CNN architectures namely, Inception-Resnet-v2 and XceptionNet for detecting the deepfakes. The proposed approach not only outshines the existing approaches in terms of efficiency and accuracy but also offers the best in terms of the given space and time complexity. © 2021 IEEE.