This study addressed the early detection of tomato leaf diseases using a transformer-based transfer learning approach. Tomatoes, a globally essential crop, are prone to diseases that can significantly reduce yields if not identified early. Leveraging cross-domain knowledge, this study introduced three transfer learning models—ViT-ImageNet, ViT-Base, and ViT-Small—and tested them on both the widely-used PlantVillage dataset and a newly collected "Tomato Ebola" dataset from farms in Dikumari, Kukareta, and Kasaisa, reflecting varied environmental conditions. The ViT-Base model achieved the highest accuracy of 99.17%, significantly outperforming existing methods, which typically fall below 90%. This demonstrates the effectiveness of transformer models in enhancing tomato disease detection, offering a powerful tool for early diagnosis and crop management.