Introduction: Epileptic spasms (ES) detection remains challenging due to subtle ictal manifestations and limited specialized expertise, particularly in resource-limited settings. Automated diagnostic tools could address these critical gaps in pediatric epilepsy care. Objectives: To develop and validate an AI diagnostic tool using hybrid EEG-video signals for automatic ES detection. Methods: This retrospective cohort study with internal cross-validation and multi-center external validation was conducted from July 2022 to May 2025. We included 252 patients with ES from Chinese PLA General Hospital and 60 from three other centers. A multimodal fusion approach combining video and electrophysiological signals was developed. Internal cohort comprised 212 patients (723.4 hours video-EEG, 7 ,348 ES and 643,215 non-ES segments). Clinical validation included 100 patients across four datasets (212.2 hours, 5,709 ES and 185,207 non-ES segments) plus 78 controls without ES for false alarm analysis. Primary outcomes included sensitivity , specificity , accuracy , precision and F1-score versus electroencephalographer interpretation. Results: In internal cross-validation, the hybrid model achieved superior performance (PRAUC= 0.7334, ROC-AUC=0.9820). In clinical validation, the model demonstrated sensitivity (0.735) and specificity (0.995) comparable to experienced electroencephalographers. With AIassistance, electroencephalographers showed improved sensitivity trends, with one rater improving from 0.620 to 0.792. For subtle ES detection, machine-assisted recognition significantly improved sensitivity by 16–21% (p