Image forgery detection faces growing challenges due to advanced manipulation techniques like GAN-based deepfakes and widespread post-processing. Despite progress in deep learning and transformer-based methods, issues such as poor model generalization, low-quality image handling, limited datasets, and lack of standardized evaluation hinder real-world effectiveness. This chapter highlights these key obstacles and stresses the need for robust, adaptable detection models, realistic generalized datasets, and unified evaluation protocols to ensure reliability in practical applications.