Layer_Zero

PanDrop — Point. Talk. Sell & Discover. PanDrop turns a casual phone video into a live marketplace listing in minutes. Point your camera at any product and talk about it naturally — "this matcha whisk, 20 dollars, barely used" — and the AI pipeline does the rest: it hears you, locates the item in the video, separates it from the frame, cross-checks the condition you claimed against what the camera actually sees, and composes a studio-quality showcase image with the price burned in. The pipeline is powered end-to-end by Meta's Model APIs: Muse Voice transcribes the seller's speech; Muse Spark extracts every item (name, price, condition claim) and, for clips without narration, detects sellable products directly from the video frames using a multi-layer vision pass (proposal scan → consolidation/verification); SAM 3.1 segments the real product out of the frame (with a GrabCut fallback for cleaner cutouts); Muse Wonder audits each item's actual visual condition and honestly flags any mismatch between the spoken claim and the footage; Muse Image composes the studio backdrop for the final showcase. Every extraction is editable before posting — tweak the price and title directly on each item card — then publish to the built-in P2P marketplace. One unified account both sells and buys: switch accounts in one click, browse the Discover feed with search, category, price and AI-condition filters, inspect any listing's SAM cutout and raw frame for authenticity, and purchase with wallet balance. The condition audit keeps sellers honest — if you claim "works perfectly" but the frames say otherwise, buyers see it before they pay. Built with React 18, TypeScript, Tailwind CSS and Framer Motion on a FastAPI backend, serving real per-job pipeline artifacts (frames, masks, cutouts, showcases) — verified end-to-end with live models.