Core Insights - The article discusses the challenges of distinguishing between real and AI-generated content in the AIGC era, emphasizing the need for explainable detection systems that can provide diagnostic reports on the authenticity of images and videos [1][6][7]. Group 1: Background and Motivation - The current AIGC detection methods are often "black box" systems that can classify content as real or fake but lack the ability to explain their reasoning, which limits their transparency and trustworthiness [6][10]. - Existing research typically treats image and video detection as separate fields, creating complexity in research and application [10][11]. Group 2: Key Contributions of IVY-FAKE - IVY-FAKE is introduced as a unified framework aimed at enhancing the explainability and effectiveness of AIGC detection [7][34]. - The IVY-FAKE dataset includes over 150,000 labeled training samples (94,781 images and 54,967 videos) and approximately 18,700 evaluation samples, covering diverse categories such as animals, objects, and DeepFake [13][12]. - Each sample in the IVY-FAKE dataset is accompanied by detailed natural language reasoning, explaining why it is classified as real or AI-generated, which is a significant advancement over existing datasets [13][12]. Group 3: Method Overview - The IVY-XDETECTOR model is designed for robust and explainable AIGC detection, utilizing a multi-modal large language model architecture [18][19]. - The model incorporates a progressive multimodal training framework, which enhances its ability to detect and explain AIGC content through a phased approach [20][22]. Group 4: Experimental Results - IVY-XDETECTOR achieved state-of-the-art (SOTA) performance in various benchmarks, with an average accuracy of 98.36% in image content classification, significantly outperforming previous models [24][25]. - In the GenVideo dataset, IVY-Det and IVY-xDet achieved over 99% accuracy, with IVY-Det's recall rate reaching 99.57% in the challenging "HotShot" subset [28][29]. Group 5: Conclusions and Implications - The introduction of IVY-FAKE marks a significant milestone in the field of AIGC detection, providing a large-scale, unified dataset for explainable AIGC detection [34][35]. - The performance of IVY-XDETECTOR demonstrates the feasibility of a unified framework for image and video AIGC detection, paving the way for more reliable and understandable content review tools [37][38].