speechlab0210/video-production-skill

video-production-skill is a set of instructions and scripts that lets an AI coding agent, such as Claude Code, Codex, or Gemini CLI, make narrated educational slide videos on its own. It was written and shared by an AI agent that runs a YouTube channel, and it is the exact pipeline used to produce more than 40 videos on that channel, with only private account details swapped out for config fields. This is not a video generator you run and get a video out of directly. It is a written procedure meant to be read by an AI agent, paired with 9 runnable scripts covering slide generation, text to speech with verification, video assembly, subtitle alignment, and cover creation. A separate folder of reference material includes a teaching style guide, narration writing advice, a list of tricky Chinese pronunciation pitfalls, and a written record of lessons learned across all 40 plus videos. The videos it produces follow one format: slides paired with AI narration and subtitles that line up with the audio. Slides can be hand drawn style images or dark themed HTML pages captured as screenshots. Narration comes from ElevenLabs text to speech, and every sentence is checked by transcribing the audio back with Whisper and comparing it to the script, only passing if the match is close enough. Subtitles are built from the original script text combined with word level timestamps so they do not drift out of sync. Running it needs Node.js 18 or newer, Python 3.9 or newer, FFmpeg, and API keys for ElevenLabs and OpenAI. The project is tuned for Traditional Chinese content, though the underlying approach is said to carry over to other languages with some adjustment. It guarantees the technical pieces line up correctly, not that the writing itself is good, since that depends on the agent and topic. The project is released under the MIT license.