This research explores how multimodal Large Language Models can be enhanced to detect and explain AI-generated images, addressing growing concerns about synthetic media manipulation.

• Introduces a novel framework for image authenticity evaluation and forgery localization
• Enhances multimodal LLMs with specialized prompting strategies to identify manipulated content
• Creates new forensic benchmarks to test and validate model performance
• Demonstrates significant improvements over existing methods in detecting AI-generated images

For security professionals, this research provides crucial tools to combat the rising threat of sophisticated AI-generated content that could be used for misinformation or fraud, helping maintain digital trust in an era of advanced generative AI.

Can GPT tell us why these images are synthesized? Empowering Multimodal Large Language Models for Forensics