Prostate cancer stands as a pressing contemporary health challenge, urging the development of sophisticated diagnostic support systems. In response, digital pathology has emerged as a crucial field, offering novel methods for diagnosis enhancement. Within this discipline lie the methods of Deep Learning applied in medical imaging. Leveraging these advancements, our work uses a Generative Artificial Intelligence model to augment the performance of content-based image retrieval (CBIR) systems in prostate cancer diagnosis. In particular, this work aims to replace the traditional method of generating augmented views as positive views with synthetic samples created by our generative model. Through extensive experimentation, notable enhancements in key evaluation metrics such as top-k accuracy, majority vote, and precision, with k representing the number of retrievals for computing each metric, are showcased. These findings highlight the substantial refinement of latent vector representations within the feature space. This improvement underscores the potential of our method to revolutionize medical image analysis and diagnosis in the realm of prostate cancer.