Summary

video-use is an open-source tool that automates video editing using an LLM (Claude Code). It targets users needing to quickly produce polished videos from raw footage via a chat interface, handling filler word removal, color grading, and subtitle generation without manual intervention.

How It Works

The system avoids direct video frame analysis, instead processing video via text. It generates detailed audio transcripts with word-level timestamps using ElevenLabs Scribe, condensed into a compact takes_packed.md file for the LLM. On-demand visual composites (timeline_view) provide context for complex decisions. This text-centric approach significantly reduces computational load and token usage compared to frame-based methods, enabling efficient LLM-driven editing.

Quick Start & Requirements

Installation requires cloning the repository, symlinking it into the Claude Code skills directory, and installing Python dependencies (pip install -e .). System prerequisites include ffmpeg (required) and yt-dlp (optional). Users must configure an ElevenLabs API key in a .env file. Setup necessitates an existing Claude Code environment and pointing it to a directory of raw video takes.

Highlighted Details

• Automated removal of filler words and dead space.
• Auto color grading with customizable FFmpeg chains.
• Burned-in subtitles with customizable formatting.
• Animation overlay generation via Manim, Remotion, or PIL using parallel sub-agents.
• Integrated self-evaluation loop at each cut boundary to catch rendering errors.
• Session memory persistence via project.md.

Maintenance & Community

The provided README does not detail specific maintenance contributors, community channels (e.g., Discord, Slack), or roadmap links.

Licensing & Compatibility

The project is described as "100% open source," but a specific license type is not explicitly stated. This lack of clarity may pose compatibility concerns for commercial use or integration into closed-source projects.

Limitations & Caveats

The tool is tightly coupled with the Claude Code environment and relies on ElevenLabs for transcription, introducing external dependencies and potential costs. The absence of a clearly defined license is a notable adoption caveat. The self-evaluation and re-rendering process is capped at three attempts per cut.