An Efficient CNN-Based Framework for Deepfake Face Image Detection

Deepfake images—photorealistic images or videos manipulated by AI—pose significant threats to information trust, privacy, and security. This study develops a convolutional neural network (CNN) approach to detect deepfake face images. We first review the surge of generative models (e.g. GANs) and the pressing need for automated detection. We then implement and train a CNN classifier (MobileNetV3-based) on a large dataset of real and fake face images. Our experiments demonstrate an accuracy of about 80.3% on a held-out validation set (using precision, recall, and F1 metrics, detailed below). We analyze the model’s strengths and weaknesses in distinguishing real versus AI-generated faces. Finally, we discuss implications for media forensics and outline future work to improve robustness. We compare our results with prior works (e.g. XceptionNet, MesoNet) and highlight the gap in generalization and adversarial resilience. Our detailed methodology, source code, and evaluation reinforce the utility of CNNs for deepfake detection and set a baseline for further research.