The advancements in AI-generated image technologies such as Generative Adversarial Networks and Diffusion Models have led to advancements in detection techniques of these AI-generated images as well. However, most of these techniques are resource intensive and expensive to use in real-time. Real-time detection of such images has become a need of the hour due to the misuse of image generation technology. In this paper, we use the CIFAKE dataset, containing both real and fake images, to test the efficacy of different deep learning techniques in detecting AI-generated images efficiently. ResNet, Inception, an Inception-CNN hybrid, and EfficientNetB5 are the four models we implement and evaluate. We assess these models using different evaluation metrics such as accuracy, AUC and computational time. Our fine-tuned EfficientNetB5 delivers state-of-the-art performance with an AUC of 0.98 and a validation accuracy of 98.78%. Additionally, it maintains an average epoch time of 118.2 s, which is computationally achievable. These results demonstrate how EfficientNetB5 can be used in real-world situations where great accuracy and computing efficiency are necessary.