Ratnesh Saxena

Cutline Studio is a continuity checker for video producers. Give it two takes of the same scene, the script or brief for that scene, and the object you care about. It compares the takes, flags where they disagree, and shows the exact frame behind every flag. A human editor accepts or rejects each flag. Nothing is auto-fixed, and nothing is exported without the editor. The problem is the 2 AM problem on real shoots and on AI-generated work alike: the presenter's jacket changes between takes, the sponsor logo drifts out of frame, a prop moves between setups, and catching it late costs a reshoot or a shipped mistake. Scrubbing two takes side by side by eye is slow, and it misses exactly the small changes that matter on delivery. Cutline works in three stages, each one inspectable by the next and by the human at the end. First, Muse Wonder is the designed ledger role: it reads the brief and each take and writes a structured scene ledger: the objects on set, wardrobe, props, screen positions, and the beats the brief calls for. In the fixture path, ledgers are authored files validated against the schema. The ledger is a fixed schema, not free text, because the rest of the pipeline has to diff two takes field by field. Second, SAM 3.1 segments the object you care about in stills, and the mask is the evidence surface: when Cutline says a prop moved or a logo is cropped, it shows the mask on the frame it came from. Clips are not yet a scored path: sampled frames from short clips are the intended next step, and Cutline does not claim tracking across frames. Third, Muse Spark is the designed judge role: it judges what measurement cannot settle: whether the jacket in take two is the same jacket in different light, or a different jacket. Every Spark judgment has to point at the ledger lines and the frames it rests on, and that citation contract is enforced in the fixture judge. Plain code catches the provable differences first: an object present in one take and missing in the other, a mask that has moved or changed size. Whatever is uncertain is raised to the human, never passed silently. Muse Image is designed for one narrow job inside the loop: when an editor wants a corrected still, the flagged frame goes out as a targeted edit on an uploaded still, and the result comes back through the same checks before the editor sees a pass. The demo will use a fixture mode, so a review never depends on a live call. On a frozen-label fixture suite of 14 checks on synthetic fixture scenes, the pipeline caught 4 of 6 planted continuity errors outright, sent 2 to human review, missed 0, and raised 0 false alarms across 4 clean controls and 2 near misses, with 1 soft alarm by design (a ledger-claimed plant raised to review rather than passed). That suite is a fixture evaluation, not a claim about real shoots; it is how the checks are tested before any real take is trusted to them. I am building Cutline Studio solo, for the Indie Developer Track, under Build for Your World: a tool built for the world I already work in. I have spent 10+ years in video production and lead video at a corporate-events agency in Mumbai, and I built it for the producer who has to prove, frame by frame, that the shot kept what it was told to keep.