TrueScene

THE PROBLEM Building sites, roads, hotel rooms, restaurants and shops are checked every day. The manager wants to know if the job was done right; the owner wants to know if anything is missing or damaged. They can't be everywhere, so they ask someone, wait, and hope the answer is true. And "looks fine" means something different to every person. THE IDEA Take one photo when a place looks right. Every new photo or walk-through video is compared with it by Meta's models. TrueScene marks each difference on the photo with a reason, and a person confirms or dismisses it. The AI does the looking; people make the decisions. FEATURES - Board: every scene sorted by what it needs next (Not set up, Ready, Needs review, Issues, Reviewed), grouped by site or client, assigned to people, with filters. - Set the standard: Muse Spark finds the objects in the initial photo, SAM 3.1 outlines and counts them. Remove one with a click, tap to add one. - Checklist: Muse Wonder drafts a short, standard or thorough checklist; every line is editable. - Photo inspections: same view, what changed and object counts are checked in parallel. Each finding is marked on the photo with a reason. - Trustworthy checks: retake hints like "turn a little right"; SAM's count decides what is missing; Muse Wonder double-checks uncertain findings; the AI never closes an issue by itself. - Review: confirm or dismiss each finding, add notes, mark reviewed, mark issues fixed. Full history of who decided what. - Video walk-throughs: one video covers many scenes; skipped places are marked "not seen"; labels stay on objects as the camera moves (optical flow, no model cost). - Reports: each scene's history in plain words with a Muse Wonder summary, plus daily, weekly or monthly summaries sent automatically. - Photos in from anywhere: QR code upload link (no app, no account), fixed cameras, or any tool through the API. - Results out: signed webhooks to any system. - Operations: live progress, cost of every model call, spending caps, location data removed from photos. META MODELS SAM 3.1 measures, Muse Spark looks and compares, Muse Wonder writes and double-checks. Our own code combines their answers with clear rules, so every result can be explained. TRY IT https://truescene.fly.dev → "Try TrueScene" (no sign-up). Open a card in Needs review, decide each finding, then play a video inspection. Important! GitHub repo is public! https://github.com/nodarik/TrueScene