The generation of synthetic human faces using deep learning has advanced rapidly, enabling the creation of photorealistic images that closely resemble real human faces. In this work, we propose a novel framework that combines Deep Convolutional Generative Adversarial Networks (DCGANs) for synthetic face generation with a Convolutional Neural Network (CNN)-based classifier enhanced by facial landmark analysis for explainable fake image detection. Our system is trained on a subset of 50,000 images from the CelebA dataset and generates 10,000 synthetic faces. The classifier achieves an accuracy of 88% and a Frechet Inception Distance (FID) of 31.2 for the generated images. Unlike existing methods such as FakeNet, our approach fuses deep visual features with landmark-based structural cues, providing visual justifications through asymmetry and alignment analysis. A key limitation is reduced performance when processing faces with complex backgrounds, occlusions, or low resolution. This work highlights the potential of integrating explainability into deepfake detection for greater trust and transparency.