Forensics-Bench introduces the first comprehensive benchmark suite for evaluating Large Vision Language Models (LVLMs) on forgery detection tasks.

• Addresses the growing threat of AI-generated fake media across multiple domains
• Evaluates 9 LVLMs across 7 forgery detection tasks with 10,000+ test samples
• Provides detailed analysis of model capabilities and limitations in identifying manipulated media
• Establishes a foundation for developing more robust forgery detection systems

This research is crucial for security applications as AI-generated forgeries become increasingly sophisticated and harder to detect, posing significant risks to information integrity, privacy, and social trust.

Forensics-Bench: A Comprehensive Forgery Detection Benchmark Suite for Large Vision Language Models