Name It

How-to presenters point at things: "add this", "about this much", "until it looks like this". Blind and low-vision viewers hear the words but miss what is being pointed at. Human describers cannot keep up, and AI describers stay vague or guess. Name It finds each "this" or "that" in a video and works out what it refers to with the frame and the object's outline. It writes a short description that fits the next pause, or pauses the video if there isn't one. Anything it isn't sure about goes to a volunteer describer for review. An accessible player then reads out the approved descriptions. On real narrated films it had never seen, it named the right thing 75% of the time, against 60% without grounding. It was confidently wrong about half as often (22% against 41%). It hasn't been tested with blind or low-vision users yet. Live demo: https://name-it-rho.vercel.app