What is auto-video-gen?

Auto Video Gen turns a news article link, a GitHub repo URL, or a plain text or markdown file into a finished vertical video, sized for TikTok, Reels, and Shorts, without any manual video editing. The pitch is that a task which normally takes about three hours (writing a script, choosing visuals, recording or finding a voice, then editing in CapCut or Premiere) can be done in around five minutes by pasting a single URL. The recommended way to use it is through Claude Code, either in a terminal or the VS Code extension. After installing the project and setting a text-to-speech API key, a user types a single command such as create-news-video followed by a URL, and the pipeline handles the rest. There is also a manual path for people without Claude Code, where a script file is edited by hand and run through the project's own pipeline command. Under the hood, Claude Code handles the creative part, generating a script and picking from six visual templates such as a hook, comparison, stat highlight, feature list, callout, and outro. The script is validated, then each scene is converted to speech using one of three text-to-speech providers, Vbee, LucyLab, or ElevenLabs, with LucyLab and Vbee aimed at Vietnamese voices and ElevenLabs supporting more than 30 languages. Sound effects are mixed in automatically. Frames are rendered with a tool called HyperFrames, built on Puppeteer and GSAP animations, and encoded into a final video at 1080 by 1920 resolution, 30 frames per second. The project is a fork of an earlier Auto-Create-Video project, adding the Vbee voice provider, a second lighter visual theme, and automatic generation of a TikTok caption with hashtags after each render. It is built with Node.js 22 and above, TypeScript, Zod for validating the script format, and Vitest for testing. The README does not state a license.