The Visual Counter Turing Test (VCT2): A Benchmark for Evaluating AI-Generated Image Detection and the Visual AI Index (VAI)

The introduction of the Visual Counter Turing Test (VCT2) marks a significant step in addressing the growing concerns surrounding the misuse of AI-generated visuals, particularly in misinformation campaigns. With the rapid advancement of text-to-image generative models, existing AI-generated image detection (AGID) methods have struggled, often overfitting to known generators and failing to accurately identify outputs from newer models. The VCT2 benchmark, comprising 166,000 images from six state-of-the-art T2I systems, reveals a concerning detection accuracy of only 58% on both COCOAI and TwitterAI subsets. This highlights the pressing need for more robust detection mechanisms. Additionally, the proposed Visual AI Index (VAI) aims to provide a more nuanced understanding of image realism, indicating a moderate inverse relationship between VAI scores and detection accuracy. As AI-generated images become increasingly realistic, the challenge of distinguishing them from real images intensi…