VLMs typically connect an image encoder to a language model, aligning visual features with words. Research such as CLIP and LLaVA showed how to train them efficiently.
Applications include visual assistance for blind users, document processing and robots that follow spoken instructions. See LLaVA explained.