Skin and oral cancers are the deadliest malignancies across world, fulfilling the critical requirements for accurate and early detection methods to improve the patient’s needs. This research proposes a novel deep learning-based method to enhance the diagnosis of both skin and oral cancers, leveraging the power of Convolutional Neural Networks (CNNs) in combination with other state-of-art deep learning algorithms. The proposal integrates CNN algorithm for feature extraction, with two other algorithms like Vision Transformers (ViTs) for classification and Generative Adversarial Networks (GANs) for data augmentation. The framework utilizes a dataset containing various high-resolution dermoscopic images as well as oral lesion images, which are firstly preprocessed to eliminate issues like unbalancing and noise. The CNN-ViT based hybrid model incorporates both localized as well as global features, which results in achieving high performance in detecting malignant lesions. Afterwards, GANs are utilized to synthesize real-time dataset, avoiding the challenges of limited data and enhancing the framework generalizability. Experimental outcomes demonstrate that the proposed model exhibits better results than the traditional machine learning approaches, achieving an accuracy of 97.7%, sensitivity of 96.3%, and specificity of 94.9%. Comparative analysis exhibits the framework's potential in minimizing false negatives, a critical factor in detection of cancer. This research encompasses the transformative role of deep learning in cancer detection, offering a highly scalable, non-invasive, and highly reliable solution for the timely detection of skin and oral cancers.