FakeShield introduces a novel approach to detect and localize image forgeries, addressing critical limitations of existing methods through multi-modal large language models.

• Solves the black-box problem by providing explainable detection rationales
• Achieves superior generalization across diverse tampering methods (Photoshop, DeepFake, AI-generated content)
• Leverages multi-modal LLMs to bridge visual forgery detection with natural language explanations
• Enhances security posture against increasingly sophisticated AI-generated fake images

This research is crucial for security professionals as it provides transparent, trustworthy tools to combat the growing threat of manipulated visual content in an era of advanced generative AI.

FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models