Back to directory

claude-code-video-toolkit

A Claude Code workspace of skills, commands and templates that write a video script, generate voiceover, music and visuals with open-source models, and render the finished MP4, running on the user's own cloud GPU account at near-cost prices.

claude-code-video-toolkit

claude-code-video-toolkit turns Claude Code into an AI-native video production tool: describing a video lets it write the script, generate the voiceover, music and visuals, and render the final MP4, using open-source models such as Qwen3-TTS for voice, FLUX.2 for images and ACE-Step for music rather than a paid hosted video-generation API. Because those models run on the user's own cloud GPU account, per-video generation cost is reported in cents, for example roughly a cent for a voiceover and about twenty-three cents for an AI video clip.

An interactive setup command configures a cloud GPU provider, file transfer and voice options in a few minutes, and a second command scaffolds a new video project from a template and walks through the whole workflow; a bundled example can render a short video immediately with no API keys or setup at all.

It requires Node.js 18 or newer and Claude Code, with uv recommended for installing the Python-based AI tooling and ffmpeg optional. It is released under the MIT license.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.