Monkeypox, an occasional virus, is often misdiagnosed as other skin diseases. It is important to identify it early to treat. This research presents a robust hybrid deep-learning model which combines the powerful feature extraction capabilities of Vision Transformer B-16 and the adaptability of Big Transfer BIT-M-R50x1. LIME and GradCAM++ help make the model’s decisions clear and understandable. Our proposed model was trained using a public dataset with 3,192 images of two types and shows an impressive accuracy of 99.07%. Besdes, It attains 99% precision, 98.87% recall, and a strong F1-score of 99.16%. This remarkable performance of the proposed hybrid approach offers the potential to improve early and accurate monkeypox diagnosis, essential for timely medical action and effective public health strategies. Future work will focus on expanding datasets, integrating multi-modal data, deploying in real-time clinical settings, monitoring performance, and conducting rigorous evaluations.