AI-generated images have reached the level of realism that detecting them apart from real ones is a challenging task, posing various risks in crucial domains such as security, forensics, and medicine. To address this, we have proposed a novel approach RvAI to detect AI-generated faces using the 140K Real and Fake Faces dataset (70K real images from Flickr and 70K fake images from StyleGAN). Our method leverages a DenseNet-121 model amplified with handpicked features, including anti-aliasing, edge detection, DCT frequency speed, DCT on Contrastive Language-Image Pre-Training (CLIP) embeddings, lighting and perspective errors, object co-occurrence, contextual clues, and pixel noise analysis. The proposed model achieves a detection accuracy of 92%, surpassing state-of-the-art methods by approximately 5%. However, the model faces challenges in generalizing to cartoonish or overtly AI-generated images, which differ significantly from the highly realistic samples in the training dataset. Future directions include enhancing adaptability to diverse generative styles. This proposed methodology gives a good base for further improvements in the field of AI-generated images.