Ashna Manzoor, Dr. Mohammad Pasha 02 09 Recent developments in artificial intelligence and other synthetic picture generating techniques have produced incredibly lifelike visuals that are nearly identical to actual photographs This poses serious problems for the legitimacy and dependability of data particularly in fields where image integrity is crucial like journalism social media and scientific study Using a deep learning network based on ResNet50 this study suggests a method for efficiently differentiating between actual and artificial intelligencegenerated photos Images are divided into two categories as part of the categorization task real and AIgenerated Even while artificial photos can mimic intricate visual elements like lighting reflections and textures they frequently differ from real photographs due to minor visual flaws The study examines these variations concentrating on small irregularities and artifacts that are commonly found in AIgenerated material like distorted backgrounds strange lighting and strange textures Machine learning algorithms can accurately identify these artifacts even if they are not always visible to the human eye These visual signals are learned and classified using the ResNet50 model which gives the system a high degree of accuracy when separating actual photos from fakes The algorithm finds important image characteristics that act as authenticity markers by training on a sizable dataset of both actual and artificial intelligencegenerated photos In order to determine which features of the photos are most instructive for categorization the study also investigates the interpretability of the models judgments Copyright (c) Journal of Science Engineering Technology and Management Science. All rights reserved 2026