The growth of e-commerce has resulted in an increase of counterfeit images that are almost identical to authentic ones due to sophisticated manipulation methods. However, current solutions often lack generalization capabilities across diverse datasets and struggle to provide interpretability in their detection outcomes. This research proposes a novel approach using two Deep Convolutional GAN (DCGAN) models, applied to the CelebA dataset and an e-commerce dataset of product images, to detect fake images. Both models integrate Gradient-weighted Class Activation Mapping (Grad-CAM) to visually highlight the areas where real and fake images differ, enhancing model interpretability. The suggested approach attains a recall of 98.76% and an F1 score of 98.74% on the e-commerce dataset, showcasing notable advancements compared to current methods. This approach not only enhances detection accuracy but also promotes explainable AI by delivering clearer understanding of the reasoning behind decisions, providing a robust and scalable tool for verifying image authenticity in real-world applications.