<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> This project offers a video editing agent powered by Claude Code Skills, designed to automate the removal of verbal tics, filler words, and silences, alongside generating high-quality subtitles. It targets users seeking efficient, AI-assisted video post-production, streamlining the workflow from transcription to final edits with a self-learning component for personalized results.

How It Works

The agent utilizes Claude Code Skills to orchestrate a multi-stage video editing pipeline. It begins with transcription using Whisper and specialized FunASR models for precise identification of verbal tics, filler words, and silence segments. Users review a generated draft before the system executes edits via FFmpeg, iteratively refining the video until all detected errors are removed. Finally, it generates and burns-in subtitles, incorporating dictionary corrections for enhanced accuracy. The system aims to learn user preferences over time for personalized editing.

Quick Start & Requirements

• Primary install / run command: Installation involves cloning the repository into the Claude Code skills directory (~/.claude/skills/videocut). Environment setup and model download (approx. 5GB) are initiated via the /videocut:install command within Claude Code.
• Non-default prerequisites and dependencies: Python 3.8+, FFmpeg, FunASR (for verbal tic detection), and the Whisper large-v3 model are required.
• Estimated setup time or resource footprint: Requires ~5GB for model downloads. Setup is integrated into the Claude Code environment.
• Links: No direct links to official quick-start guides or demos are provided in the README.

Highlighted Details

• Automated detection and precise localization of verbal tics, filler words ("um", "uh"), and silence segments (>= 1s).
• Subtitle generation utilizes Whisper with dictionary correction, claiming superior quality over tools like Jianying.
• A self-updating mechanism allows the agent to learn and adapt to user editing preferences over time.
• Iterative editing process ensures zero verbal tics in the final output after user confirmation.

Maintenance & Community

The README does not provide details on notable contributors, community channels (like Discord/Slack), roadmaps, or sponsorships.

Licensing & Compatibility

The project is released under the MIT license, generally permitting broad commercial use and modification with attribution.

Limitations & Caveats

The project relies on the Claude Code environment and specific AI models (FunASR, Whisper large-v3), requiring significant disk space (~5GB) for models. The effectiveness and learning capacity of the "self-updating" feature are not detailed. It is presented as a skill for Claude Code, implying integration within that specific ecosystem is necessary.