Multi-modal medical imaging is a cornerstone of modern diagnostics, yet efficiently generating detailed and clinically accurate captions remains a challenge. The time-consuming nature of report generation coupled with increase in workload of radiologists, and the intervariability due to the reliance on the expertise of radiologists, highlights the need for a smart, consistent, and automated image captioning model to streamline radiologist workflow. However, most of the current research is focused on perfecting single modality or single organ medical image captioning, leaving the applicability of these methods to multi-modal image captioning largely unexplored. This paper aims to bridge this gap by evaluating the transferability of existing methods on single-modal multi-organ image captioning to the multi-modal image captioning model using the accessible subset of the MedTrinity-25M dataset. A multi-modal medical image captioning system is designed to enhance diagnostic interpretation across a variety of imaging modalities, encompassing MRI, histopathology, and CT scan modalities. Eight variants: three specialists and one generalized model trained on batch sizes 64 and 128 were evaluated using natural language generation metrics such as BLEU-[1–4], METEOR, ROUGE-L, and CIDEr. Results show that the generalized model, trained on diverse multi-modal datasets, achieving superior performance on average, particularly for batch size 128. This outcome culminates in the development of M3DICAPS, a web application developed using open-source Streamlit and deployed via Google Cloud Platform, allowing users to interactively explore and evaluate the models.