Avant-Garde-AI/video-annotator ? reverse-engineered prompt

Build me a React video annotation toolkit that I can drop into an app and use to draw on top of a video, mark objects and motion, and export those annotations in a few useful formats.

I want a clean demo page where I can load a video, play and pause it, scrub through time, and use simple tools like freehand drawing, arrows, rectangles, ellipses, and text. It should save annotations with undo and redo, let me edit or remove them, and show a timeline and export button. Please make it work as an embeddable component with an imperative ref API so I can control playback and export from my own UI.

The exports should include a normal JSON format plus versions that are ready for Veo 3.1, Sora 2, Runway Gen 3, LangChain, COCO, MOT, and CSV. If you need to look up current docs online for the video model export formats, do that. Keep the code typed, modular, and easy to extend.