Deepfakes are purposely used to spread false information or have some malicious intent in their use. Detection of deepfakes has become increasingly challenging nowadays because of evolving technology that is used in generation in deepfake. This paper presents neural architecture search (NAS)-based deep learning model that uses the You Only Look Once (YOLO) model for image segmentation and data augmentation for diversifying the dataset, which aims to improve the accuracy in detection of deepfakes compared to existing models. Celeb-DF v2 dataset was used that contained 590 real videos and 5639 deepfake videos from which 100 deepfake and 100 real videos were selected; frames were extracted, and after augmentation, dataset formed contained 2000 real and 2000 deepfake images. The proposed model achieved testing accuracy of 98.81%, and it excelled in other metrics used for evaluation like f1 score, precision, recall. The proposed model can be used for real-time deepfake detection, and potentially for applications aimed at reducing the spread of misinformation, enhancing digital safety, and building trust in online platforms.