The rapid rise of generative AI has encouraged the spread of deepfake photographs, jeopardizing digital media credibility and security. Efficient detection algorithms are crucial to counter misinformation and sustain trust in visual content. This research proposes to construct a novel convolutional neural network (CNN) that achieves significant accuracy in deepfake photo recognition while minimizing computational and energy costs. We introduce a unique lightweight CNN architecture, LiteFakeNet that leverages depthwise separable convolutions, trained and tested on the CIFAKE dataset of 120,000 real and AI-generated images. Our model achieves an accuracy of 90.5%, precision of 91.2%, and recall of 89.8%, with a compact design of around less than a million parameters along with only 0.16 million FLOPs. This research extends deepfake detection by proposing a scalable, energy-efficient method that maintains robust performance. This efficiency of LiteFakeNet aligns with Industry 5.0 paradigm by enabling human-AI collaboration for media verification and supports sustainable green technology that reduce the energy consumption of the deep learning model.