Yeah +1 to this

People use text with LLMs but it's great to have a high fidelity "analyze this image"