soumatheusgomes/editonny

editonny – AI‑powered video‑editing skill for Claude Code & Codex

What it is

An open‑source skill (plugin) that runs on your computer and lets you edit videos by simply describing what you want in plain language.
Works with the Claude Code and Codex agents, so you type a request like “cut the pauses, add captions and turn this part of my speech into a motion scene”, and the skill does the heavy lifting.
All transcription, rendering and visual processing happen locally – no external LLM API keys are required (the agent’s subscription covers the LLM calls).

Key features

Feature	What it does
Pause & mistake cutting	Detects silences, breaths and mis‑starts and removes them automatically.
Caption generation	Generates subtitles from the audio and can style them in several fonts.
Audio fixing & reframing	Normalises volume, can re‑frame footage to different aspect ratios while keeping the subject centred.
Motion scenes	Turns spoken statements into pre‑built animated “catalog” scenes (145 templates, 79 bundled clips).
Local transcription & OCR	Uses a local speech‑to‑text model and Apple Vision or Tesseract OCR for caption checks.
Full‑pipeline verification	Shows an 8‑15 s sample for user approval before final export.
Cross‑platform install	One‑click install request works for both Claude Code and Codex; also supports a manual terminal install.

How it works

Install – Paste the provided install request into Claude Code or Codex. The agent clones the repo, sets up a Python environment (via uv), installs FFmpeg, Node, OCR tools, and verifies everything.
Describe your edit – Send a short command (e.g., $editonny Edit a 45‑second vertical video from /videos, cut pauses, add captions).
Planning – The skill scans the footage, proposes a full set of editing parameters, and waits for your confirmation.
Processing – It cuts the video, adds captions, applies any reframing, and optionally inserts a motion‑scene clip based on spoken text.
Review – An 8‑15 s preview is saved; you approve or tweak before the final MP4 is exported to ~/Downloads/editonny/<date‑title>/exports/.

Installation options

Agent‑driven – Copy the multi‑line request from the README and paste it into Claude Code (/editonny) or Codex ($editonny). The agent handles cloning, dependency installation, and registration.

Terminal –

git clone https://github.com/soumatheusgomes/editonny.git
cd editonny
python3 install.py --target both   # or --target codex / --target claude

Afterwards run the provided doctor and media status scripts to verify the setup.

Running the guide locally

npm ci               # installs the Next.js guide and design‑system
python3 editonny.py guide   # starts dev server at http://localhost:4173

The guide is a multilingual (EN/PT‑BR/ES) web UI that helps you build edit requests and browse the motion‑scene catalog.

Typical use cases

Quick social‑media video polishing: remove “ums”, add subtitles, and re‑frame for TikTok, Instagram Reels, etc.
Turning a spoken product demo into a short explainer with animated highlight scenes.
Batch‑processing interview footage where you only want the clean takes and captions.

Project structure (high‑level)

skills/editonny/ – the portable skill and CLI entry point.
skills/editonny/engine/ – core Python code for audio analysis, transcription, and video stitching.
skills/editonny/visuals/ – bundled motion‑scene assets and recipes.
web/ – Next.js website (the guide) with localized request builder.
design-system/ – reusable React component library used by the guide.
docs/ – detailed architecture, installation, integration and validation docs.

License

First‑party code and documentation are MIT‑licensed. Third‑party assets retain their original licenses (see THIRD_PARTY_NOTICES.md).

Where to get help

Open an issue on GitHub, star the repo, or contact the author via Instagram or email (links in the README).

editonny lets you talk to your AI agent and get a fully edited video back, with no cloud‑only services required.