Automated report generation from medical images is a promising step to improve clinical decision-making and reduce radiologists’ workload. In order to generate diagnostic tags from chest X-ray images, this work proposes a deep learning system that combines Vision Transformer (ViT) embeddings with a customized transformer-based text model. Preprocessed X-ray images are used at the beginning of the process to improve contrast and standardize visual characteristics. Spatial-aware image embeddings are generated using a pretrained ViT model. In parallel, a T5 encoder is used to encode diagnostic tags from radiology reports into rich semantic text embeddings. Given visual input, a specific T5-based encoder-decoder model is trained using the corresponding image-text embeddings to generate diagnostic phrases. A BLEU-4 score of 0.502 indicates superior agreement between expected and ground reality tags in training results, supporting the model’s capacity to identify complex cross-modal relationships in medical data. The technique demonstrates how AI-enabled radiology platforms can produce accurate and readable medical reports, significantly supporting clinical procedures. For example, the model replaced general phrases like “pulmonary markings are prominent” with more specific outputs such as “consistent with diffuse pulmonary fibrosis,” improving interpretability while preserving clinical meaning.