TapLinko

VisualForge AI is a verified visual-difference engine, built from scratch during the hackathon, for a problem I live with: my spot-the-difference game Diferia ships with sixty hand-made levels, and when a player finishes them the game is over. VisualForge rebuilds it as an engine, so a level exists the moment you ask for one. Give it a photo, or just describe a scene (typed or dictated), and it produces a modified version with realistic, localized changes. Every change is measured from the pixels (mask, bounding box, centroid) and verified before it is published; outside the edited regions, every pixel is the original. Annotations stay on the server, clicks are validated there, and each difference comes with progressive hints and a difficulty rating. The same engine serves manufacturing: the QC Inspection Trainer adds realistic defects (scratches, dents, corrosion, missing parts) to a product photo or to a part it generates from a description, at the inspector level you choose, and exports COCO / YOLO datasets with masks, defect classification and confidence, with low-confidence samples split out for review. In both experiences you can challenge the AI itself to find the differences, validated against the real boxes. Live demo: https://visualforge-ai.taplinko.store (gallery free to play; generating new levels is gated by an access code as a spending guard; runs locally with your own key in five minutes). Code, MIT: https://github.com/RemotingCoders/VisualForge-AI. Everything in the repository was written during the event, starting Oct 28; Diferia existed as a game and none of its previous process was reused. Synthetic data: validate before industrial use.