Open-Source Claude Video Editor: Turn a Raw Clip Into an Animated Short

OPEN VIDEO EDIT is an open-source video editor that runs inside Claude Code. You hand Claude a raw talking-head clip and it hands back a finished, fully animated short: motion graphics, kinetic captions, real logos, real faces, reaction GIFs, live b-roll cards, rolling counters and a hyper-edit cut, all timed to the word you said. There is no timeline and no template to drag around. Every frame is drawn in Python and rendered on your own machine, so there is no per-video generation cost.

The code is free at github.com/kevinbadi/open-edits, and the full walkthrough is on YouTube: watch the launch video.

  clip.mov
    |
    +- prep_talking_head.py --> face band + clean audio
    |                           + words.json (word timestamps)
    |
    +- fetch logos / faces / GIFs for everything you name
    |
    +- render.py: sections cut at spoken words,
    |             every beat starts ON its word
    |
    +- PIL draws each frame: graphics + band + caption
    |
    +- ffmpeg --> 1080x1920 H.264 with your original audio

How OPEN VIDEO EDIT turns a clip into a short

Why an AI video editor built on Claude Code

Most AI video tools are web apps with a fixed set of templates: auto-captions, a zoom here and there, maybe stock b-roll. They’re fast, and every video looks the same. OPEN VIDEO EDIT takes the opposite approach. Claude reads your transcript, decides what should be on screen for each line, fetches the real assets (the actual logo of the product you mention, the actual face of the founder you name), writes a render script for that one video, checks preview frames for collisions, and renders. Each edit is custom because the script is written fresh each time.

Because it is a set of Claude Code skills, you steer it in plain English: “make the intro darker”, “put a counter on the 10,000 users line”, “use my screen recording for the demo section”.

What’s inside

split-animated-talking-head: the flagship. A 1080×1920 vertical short with your talking head in a band at the bottom, a dense animated slot on top, and captions resting on the accent line.
longform-animated-talking-head: the same hyper-edit for 16:9 YouTube long-form, with a split view (animation slot plus head card) and chapter beats.
fireship-style-edit and fireship-style-short: faceless tech-news style over a voiceover, fast cuts, footage that literally shows what each line says, tweet and headline stacks, diagrams and memes.
split-animated-short: the split format for an AI avatar instead of your own footage.
reverse-engineer: point it at any reference video and it measures the editing style (cuts, shot lengths, how often something new lands, motion, layout, palette, loudness, words per minute), builds contact sheets of every shot, then scaffolds a new editing skill in that style and proves it with a demo render.

Behind the skills: face-aware cropping that never ships a black bar, Whisper word timestamps primed with your product names, logo and portrait fetchers, Giphy search, an iOS-style rolling number counter, punch-ins, whip cuts, light effects, and three compositors.

Install

You need Claude Code, Python 3.9+ and ffmpeg.

git clone https://github.com/kevinbadi/open-edits.git
cd open-edits
pip install -r requirements.txt
brew install ffmpeg            # or apt install ffmpeg
./install.sh                   # copies the skills into ~/.claude/skills

An optional Giphy key enables reaction GIFs.

Use it

In Claude Code:

edit this talking head: ~/Downloads/my-clip.mov
here's b-roll of what I'm talking about: ~/Downloads/screen-recording.mp4

For long-form: “edit this landscape talking head”. For faceless: “make a Fireship-style video about X”. A 30-second short renders in a few minutes, and long renders split into parallel chunks.

The rules that make it look good

Graphics land on the word. Never while you’re still winding up to say it, at most a tenth of a second early.
Literal visuals. If you say a company, its logo appears. If you name a person, their face appears.
One moving thing per area. No pile-ups of animations fighting for attention.
Restraint on effects. Punch-ins and impact effects are used on the one dramatic beat of a section, not every line.

Every rule in the skills was learned on real videos that went out on our channels.

From edit to every platform

Editing is half the job. Once the short is rendered, Claude can post it to TikTok, Instagram Reels and YouTube Shorts in one step with the Creator OS MCP server, with a custom cover, a first comment and a schedule. See posting one short video everywhere. That is the full loop: record, edit, publish, read the analytics, all from one Claude session.

FAQ

Is it really free?

Yes. The code is open source, rendering happens locally, and nothing is generated with paid image or video models. You use your own Claude Code plan.

Do I need to know Python?

No. Claude writes and runs the scripts. Reading the example render files helps if you want to customise the style.

Can it copy a specific creator’s editing style?

The reverse-engineer skill learns the grammar of a style (pacing, layout, text treatment) from a reference video and builds a new skill from it. It never copies the reference’s footage, music or branding.

Prefer to watch? The KevBuildsApps YouTube channel walks through builds like this one, start to finish.

Get the code

Star and clone it at github.com/kevinbadi/open-edits, watch the launch video, and when you’re ready to publish what you make, connect your socials to Creator OS.