A multimodal AI model integrating skin images with structured clinical information improved skin lesion classification across three datasets, achieving balanced accuracy of up to 99.6% in an institutional dataset, according to a retrospective diagnostic study. The model was assessed on two publicly available datasets and one institutional dataset: PAD-UFES-20, comprising 2,298 clinical images with 21 structured clinical variables across six lesion classes; ISIC 2019, containing 25,331 dermoscopic images spanning eight diagnostic categories; and the institutional PKU-Skin dataset, which included 800 clinical images from 643 patients alongside corresponding clinical metadata and focused on binary classification of psoriasis and basal cell carcinoma. On PAD-UFES-20, the model achieved a balanced accuracy of 85.6%, outperforming a previous multimodal baseline model, which achieved 75.1%, with sensitivity of 76.1%, specificity of 95.2%, an F1 score of 78.1%, and overall accuracy of 78.3%. On ISIC 2019, it achieved 86.0% balanced accuracy, an F1 score of 84.3%, and overall accuracy of 84.5%. The highest results were observed in the PKU-Skin dataset, with balanced accuracy of 99.6% and overall accuracy of 99.2%, though this dataset represented a separate binary classification task, providing supportive evidence from an institutional setting rather than formal external validation of a single fixed