This study explores the integration of vision and language models to automate image captioning, leveraging a Vision Encoder-Decoder framework that combines a Visual Transformer (ViT) as the encoder and GPT-J as the decoder. We develop a model capable of generating coherent and contextually relevant captions for images, which could have broad applications in accessibility and content management systems. Using the MS COCO dataset, we implemented and trained our model, optimizing various parameters to enhance the quality of generated text. Initial results, evaluated using the ROUGE metric, indicate promising performance with respect to precision, recall, and f-measure, though they also highlight areas for improvement in handling complex visual contexts. This study not only underscores the potential of advanced encoder-decoder architectures in bridging visual and textual domains but also paves the way for future enhancements in automated caption generation technologies.