LEGION introduces a groundbreaking approach to detect AI-generated images with artifact-level explanations, addressing growing security concerns in digital media.

• Combines grounding and explanation techniques to identify synthetic images
• Created SynthScars, a comprehensive dataset featuring modern generators and fine-grained annotations
• Provides artifact-level textual interpretability of detection decisions
• Significantly improves upon existing detection methods with explainable results

This research is critical for security professionals as AI-generated deceptive imagery creates unprecedented challenges for content authentication. LEGION's explainable approach provides defenders with both detection capabilities and understanding of the specific artifacts that reveal synthetic content.

LEGION: Learning to Ground and Explain for Synthetic Image Detection