Isnt this already possible with the tech that makes like those annotations for images on the web for blind people? Just extend that into images that you allow and boom, now you can use semantic search to give importance to words and properly search it. Does this person really need millions?