As artificial intelligence (AI) and cloud computing have developed, video strategies, some of which are used to manipulate by altering people's face identities, called “deep fakes,” have become more sophisticated and faster. Spotting these fakes is not easy. In recent history, fakes have been employed as influencers to plan terrorist attacks, produce revenge porn, extort individuals and so forth, foment political unrest and much more. Hence, discovering fakes is crucial to allowing their widespread propagating on social media networks to come to a stop. This research article looks into how well Efficient-Net performs at identifying deepfake videos. In this paper, we propose an Efficient Net based efficient deep fake-detection system. To improve the quality of video inputs provided by the user frame extraction and preprocessing are performed. Efficient Net looks at the gathered frames and identifies even the slightest inconsistencies and abnormalities associated with deep fakes. Aggregation of these frame-level predictions leads to a classifier output predicting whether or not the video is fake. As per this study's result, 78% precise overall outcome is observed for the Efficient Net models. The findings of the experiment provide insight into its ability to address the challenges of detecting deepfake videos, making Efficient-Net a strong candidate for implementation in real life settings where rapid and accurate detection is critical for distinguishing between altered images and real ones.