Conventional and AI-Generated Image Forgery: A PRISMA-Based Systematic Review and Emerging Paradigms

The rapid proliferation of advanced editing tools and sophisticated generative architectures has escalated the threat of conventional image manipulations and AI-generated synthetic content, undermining digital media integrity and facilitating cybercrime. Ensuring the authenticity of visual evidence requires robust, multi-layered forensic detection mechanisms. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, this study synthesizes recent detection methodologies published across six major databases: IEEE Xplore, ScienceDirect, MDPI, Springer Nature, Taylor & Francis, and Wiley (2020–2026). This study identified 3,462 records, screened 3,418, and selected 1,199 studies. For conventional forgeries, detection relies on machine learning (ML), deep learning (DL), and physical/format artifacts including pixel distributions, lighting inconsistencies, camera traces, and geometric anomalies. For AI-generated content, the paradigm extends beyond traditional digital forensics to incorporate Data Fusion Architectures (DFAs), Multimodal Vision–Language (MVL) models, and Explainable AI (XAI) frameworks. From this classification, four emerging challenges or avenues or avenues for future research are established: (i) zero-shot generalization for out-of-distribution detection; (ii) robustness against social network re-encoding and lossy compression; (iii) passive detection embedded within emerging regulatory transparency standards; and (iv) mitigation of the "liar dividend", wherein authentic images are deceptively dismissed as synthetic. Ultimately, this research provides a structured forensic foundation that may assist in evaluating the admissibility of visual evidence in legal contexts, proposing a roadmap that connects technical detection, regulatory compliance, and courtroom evidentiary standards.