Social anxiety disorder (SAD) is a prevalent anxiety disorder marked by strong fear and avoidance of social scenarios. Early detection of SAD is crucial for timely interventions. However, due to the inherent social avoidance in SAD, clinical screening remains challenging. Traditional questionnaires suffer from subjectivity and cultural bias. To address this, we developed the Social Artificial Intelligence Picture System (SAIPS) using generative multi-modal AI models, comprising 279 social and 118 control pictures. The system represents key SAD triggers—fear of negative evaluation, social interaction, and performance anxiety—to capture the multifaceted nature of social anxiety. Laboratory and online studies demonstrated that ratings on SAIPS robustly predict social anxiety traits both cross-sectionally and longitudinally. Machine learning analyses revealed reliable predictions even from a short version (<30 images). SAIPS offers a promising, objective, and accessible tool for early screening and long-term monitoring of social anxiety.