GitHub - veedstudio/open-edit: Open-source, agent-driven editing pipeline: create subtitles, motion graphics, slides, edit and render videos.

Open the original

Short summary

OpenEdit is an open-source tool that lets you describe video edits to AI coding agents (Claude, Gemini, Codex) and get back finished MP4s. It handles transcription, caption design, rendering, and edits—no GUI or timeline required.

  • Works as a skill you install into your coding agent; you ask it to add subtitles, cut footage, or generate motion graphics and it handles the full pipeline from transcription through render.
  • Transcription options include VEED's hosted service (free tier ~10 min/month), local WhisperX (2GB model download, nothing leaves your machine), or your own service with per-word timings.
  • Agents can generate footage from scratch (talking heads via VEED Fabric, pure motion graphics), reframe and layer existing video, pull in external AI generators or MCP servers, and iteratively refine based on feedback.

Full summary

OpenEdit is an open-source video editing tool designed to work with AI coding agents like Claude Code, Codex, or Gemini CLI. Instead of a traditional GUI, you describe what you want in plain language—"add subtitles to my video"—and the agent handles transcription, design, rendering, and outputs an MP4. Installation is straightforward: `npx skills add veedstudio/open-edit --skill open-edit` from your project folder, then you're ready to go. The tool handles its own setup on first run, checking for Node and ffmpeg, installing missing dependencies through Homebrew on Mac or printing commands for Linux and Windows, and asking once how you want speech transcribed. The system supports three transcription methods with real tradeoffs. VEED's hosted service works best but uploads your file and requires a free account (covers about ten minutes monthly). WhisperX runs entirely locally for free if you have `uv` or `pipx` installed—the model downloads on first use, taking about 2 GB for the fast version, and nothing leaves your machine. You can also bring your own Whisper-compatible service. The agent can do more than just captions: it cuts and reframes footage, layers motion graphics, turns slides into video, and integrates with AI generators or MCP servers. The work lives in a `runs/` folder with both the rendered file and the code that created it, making it reproducible and version-controlled. The editing happens through an HTML composition that the agent writes, then OpenEdit renders frame by frame in a browser, inspects the results, and fixes visible problems before re-rendering only what changed. You can iterate conversationally—"make the yellow darker" or "move the text up"—and the agent adjusts. It can generate talking-head clips from scripts using VEED Fabric if you have no source footage, though generation costs are quoted upfront in your AI credits. The tool runs on macOS, Windows 10+, or Linux, though Linux testing is lighter than the other platforms. It's licensed under Apache-2.0, and rough edges are expected since every output depends on the model and agent driving it.

Ask about this article