gnipbao/codex-whiteboard-video-skill

codex-whiteboard-video-skill is an adapter that lets the Codex AI coding assistant call a separate whiteboard video generation engine. The underlying engine, called whiteboard-video-engine, takes images (photos, SVG files, sketches, or scripts) and renders them as videos that look like a hand drawing the image on a whiteboard. This repository contains only the skill instructions and a thin wrapper script, the actual rendering code and AI models live in the companion engine repository. Once installed, you can invoke the skill by referencing its SKILL.md file inside a Codex session and describing what you want. The engine traces the image through a line-art extraction step (converting a photo or illustration to sketch lines), animates the drawing stroke by stroke with configurable stroke style and timing, and outputs an mp4 video. The line-art extraction uses one of two AI models: Informative Drawings (for contour and sketch-style output) or Anime2Sketch (for anime-style output). Both models require downloading their full upstream repositories along with their pre-trained weight files, not just the weights alone, because the engine imports Python modules from those repositories directly. Setting up the models is the most involved part of the installation. Installation has two steps: installing the Python engine package from GitHub, then cloning this skill repository into the Codex skills directory. A doctor command verifies the setup. No API keys are required since all processing runs locally. The README is written primarily in Chinese. This project is aimed at developers who want whiteboard video generation inside an AI coding assistant workflow. It is MIT licensed, though the upstream model code and weights follow their own separate licenses.