Skip to Main Content

PlateProof

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

PlateProof — Demo video

PlateProof makes honest menu photos for small restaurants. Small restaurants need good photos to sell on delivery apps, WhatsApp and Instagram, and a photoshoot costs Rs 4,000-5,000. AI images are cheap but they lie: Just Eat paused its AI menu images in October 2026 after one showed bacon floating in cola, and Zomato banned AI dish images in 2024. Nobody was checking the generated photo against the real dish. How it works: the owner uploads a photo or a 5-second clip of the real plate (and can speak the request in Telugu, Hindi or English). SAM 3.1 and Muse Spark build an inventory of what is really on the plate: components, counts, colours and container. Muse Image generates a candidate, which goes through the same eyes. An independent auditor on open weights (Muse Spark Open) compares the two inventories without ever seeing the generator's prompt and rejects anything invented, missing or miscounted; its reason becomes the next prompt (up to 3 rounds). On a pass, the owner gets the photo, a short report in English and Telugu, and a hash of the source photo. If nothing passes, PlateProof ships nothing. One click also makes Instagram story, Swiggy tile and WhatsApp catalogue versions, each audited again. Live: https://plateproof-1037019152152.asia-south1.run.app

Team