This paper presents a novel generative architecture that models and transforms literary styles between two eminent Bengali authors—Rabindranath Tagore and Kazi Nazrul Islam—using a hybrid BERT-GAN framework. Although prior research on text style transfer has shown success in high-resource languages through models such as CycleGAN-BERT, UGAN, and cross-alignment methods, the Bengali literature remains largely underexplored due to its low-resource nature and rich morphological complexity. Our approach leverages multilingual BERT to extract deep semantic and contextual embeddings, which are then refined via a GAN to produce stylistically aligned representations. Unlike traditional methods, the model does not require parallel corpora or manual rules, allowing an effective unsupervised style transformation. Extensive experiments on curated literary texts demonstrate that the discriminator effectively distinguishes authorial styles, while the generator learns to replicate nuanced stylistic traits, preserving semantic content. This work contributes to the growing body of research on neural style transfer in underrepresented languages, providing a scalable foundation for Bengali literary generation and authorial style modeling.