Abstract: Background　Cutaneous warts are highly prevalent infectious diseases among military personnel, severely disrupting their daily work and lives. However, primary medical staff lack the clinical proficiency to diagnose plantar warts. Objective　To construct an artificial intelligence diagnostic model for cutaneous warts to provide a simple and reliable method for improving the capacity of primary medical staff in diagnosing plantar warts. Methods　Patients with plantar warts, common warts, calluses, and corns who visited the Dermatology Department of the Ninth Medical Center of PLA General Hospital from November 2022 to December 2024 were included. Smartphone photographs and dermoscopic images of lesions, including cutaneous warts (plantar warts, common warts) and non-wart lesions (calluses, corns), were collected. Patients were randomly assigned to the training set or validation set at an 8: 2 ratio. Diagnostic models based on smartphone images and dermoscopic images were constructed using Swin Transformer-Base, ResNet, Vision Transformer, and ConvNeXt architectures, respectively. The optimal model was selected according to performance metrics, including accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC), in the validation set. Smartphone photographs and dermoscopic images of identical plantar warts and non-wart lesions were collected from three medical centers of the Chinese PLA General Hospital from January 2025 to January 2026 for clinical testing. With the diagnostic conclusions of senior dermatologists as the gold standard, the performance of the optimal model for diagnosing plantar warts was evaluated. Results　A total of 600 smartphone images and 417 dermoscopic images of cutaneous wart lesions, as well as 142 smartphone images and 96 dermoscopic images of non-wart lesions, were collected for diagnostic model construction. For clinical testing, 103 smartphone images and paired dermoscopic images of the same plantar wart lesions were obtained; meanwhile, 19 smartphone images and paired dermoscopic images of identical non-wart lesions were collected. The Swin Transformer-Base model achieved the optimal diagnostic performance for cutaneous warts. For the smartphone image-based model, the accuracy, sensitivity, specificity, and AUC on the validation set were 0.919, 0.950, 0.793, and 0.972, respectively. For the dermoscopy-based model, the corresponding indicators on the validation set were 0.952, 0.964, 0.905, and 0.993. In the clinical test set, the smartphone image-based model yielded an accuracy of 0.943, sensitivity of 0.961, specificity of 0.842, and AUC of 0.986; the dermoscopy-based model attained an accuracy of 0.975, sensitivity of 0.990, specificity of 0.895, and AUC of 0.995. No statistically significant difference was found between the diagnostic performance of the smartphone image-based model and the dermoscopy-based model (P＞0.05).Conclusion　The Swin Transformer-Base model established in this study for diagnosing cutaneous warts based on smartphone and dermoscopic images can accurately identify plantar warts and has potential auxiliary application to improve the diagnostic proficiency of primary medical staff for plantar warts.