Deepfake technology is becoming increasingly realistic as a result of recent advances in Generative Artificial Intelligence (GAI). The rapid advancement in the creation and manipulation of synthetic images has raised serious worries regarding the implications for society. This could result in a drop in trust in digital content, as well as the spread of incorrect or fake news. This paper proposes a deep learning approach to address this problem, comparing Convolutional Neural Networks (CNN) to the MobileNetV2 model. The benchmark dataset “CIFAKE: Real and AI-Generated Synthetic Images” consisting of 120,000 images (60,000 fake and 60,000 real), with 50,000 images for training and 10,000 for testing per class, is used for the proposed comparative analysis. The experimental results demonstrated that the proposed custom CNN architecture was specifically designed for this classification task. It achieved highly accurate results, with an accuracy of 95.48%, a precision of 0.94, a recall of 0.97, and an F1-score of 0.95. Additionally, the MobileNetV2 model obtained a 95% accuracy rate, a 0.96 precision rate, a 0.94 recall rate, and an F1-score of 0.95. The demonstrated results highlight the effectiveness of CNN and MobileNetV2 for Deepfake image detection. These results highlight the potential of deep learning techniques to improve the Deepfake image detection approach’s precision and accuracy.