New Framework Explains AI-Generated Image Detection

Researchers have developed a detection framework for AI-generated images that prioritizes human-understandable explanations. The study, published on arXiv (2605.06143v1), addresses the misuse of generative AI in online disinformation campaigns. The team created a suite of detectors with varied architectures and fine-tuning strategies, trained on a large-scale dataset called AIText2Image, which contains photorealistic fake images. Performance was assessed on state-of-the-art text-to-image AI generators. Sixteen different explainable AI (XAI) methods were integrated into the framework. Visual explanations were refined and evaluated using a novel approach that collected textual and visual responses from 100 survey participants. The framework provides insights into visual-language cues in fake images, aiming to make detection systems more transparent and effective.