AI-Generated Image Detection with CNN and Interpretation Using Explainable AI

Abstract

The rapid growth of GAN-generated images has raised serious concerns regarding the reliability of digital media. This study presents a Convolutional Neural Network (CNN)-based method for detecting AIgenerated images, enhanced with Explainable AI techniques. The model utilizes CNN feature extraction to capture subtle artifacts present in synthetic images. To improve interpretability, Grad-CAM and SHAP are employed to provide both visual and quantitative insights, highlighting the key regions influencing the model’s decisions. Experimental results show an accuracy of 95.3% in differentiating real images from AI-generated ones. Grad-CAM visualizations further indicate that the model effectively focuses on meaningful patterns such as unnatural textures and generative inconsistencies. However, the approach has certain limitations, including sensitivity to adversarial attacks and difficulties in generalizing to unseen GAN architectures. The proposed system is implemented as a Django-based web application, allowing real-time image classification along with explainable outputs.