What the project is
Chris Lema’s open-source videoeditor repo on GitHub is a set of Claude Code skills, one per step. You run a single command on a raw recording and get back a finished file, with an optional 9:16 version for TikTok and Reels. It can also cut between two or three cameras.
We read it because it’s the clearest picture we’ve seen of what “edit it for me” turns into when you build it yourself.
The eight steps
Silence goes first. ffmpeg’s silencedetect filter finds gaps longer than half a second, and each is trimmed down to a 0.3-second pause. Then Whisper, a speech-to-text model, transcribes the video, and the transcript is cut into chunks of three to six seconds. Each chunk is labelled normal, emphasis or critical, and that label decides how far to zoom in.
Next, OpenCV’s face detector samples ten frames to find where your face sits, and the crop follows it. A colour grade warms the picture. An audio chain adds filtering, compression and loudness normalisation to −16 LUFS. A second transcription pass feeds captions burned in with ffmpeg’s drawtext filter. You review the result, and the last step deletes the leftovers.
Where the real work hides
The README is honest about what went wrong along the way, and that’s the useful part. Cuts are re-encoded, never stream-copied, because stream copy drifts and leaves audio blips at the joins. A standard ffmpeg install can’t draw text, so you need a fuller build. If the caption font isn’t resolved exactly right, ffmpeg quietly falls back to a generic one and the captions stop fitting. A font download that fails can save a web page with a .ttf name.
Then there’s the housekeeping. The pipeline transcribes twice, leaves eight intermediate files on disk until you approve the result, and was built and tested on a Mac with Apple Silicon and Homebrew. None of this is a flaw. It’s what one workflow honestly costs.
What it says about Claude Code
Claude Code is very good at the glue: writing commands, running ffmpeg, saving the steps as skills you can repeat. It doesn’t edit video itself. ffmpeg, Whisper and OpenCV do the editing, and Claude Code drives them.
It also means you become the maintainer. When a tool updates, a path changes or your footage looks different, the fix is yours. This pipeline is tuned for one person talking to a camera in a warm room, and other footage means retuning.
Build it or skip it
If you like tinkering, build it. You’ll learn ffmpeg, and the project is a good place to start. If what you want is a week of posts, that setup is the thing standing between you and posting. Cutleap does the cutting, captions and music in the browser, and changes are a sentence.
| Feature | DIY with Claude Code | Cutleap |
|---|---|---|
| Setup | ffmpeg, Whisper, Python, fonts, models | Open the app |
| Runs on | Your own computer | In your browser |
| Changing one thing | Edit a skill or re-run a step | Type the change |
| Tuned for | One person talking, indoors | Phone clips of you, your product and your space |
| Cost | Free code, plus Claude Code and your time | From $19 a month |
“Cut the ums and pauses. Make it 30 seconds.”
“Add captions in my brand colours.”
“Make 3 versions with different openings.”
Want a fuller request? Copy the prompt Turn one talking clip into five short videos.
Let Cutleap do the editing
Upload your clips, say what you want, get finished videos back. 7-day free trial, cancel anytime.
Get early accessQuestions
Can Claude Code edit video?
Not by itself. It writes and runs commands for tools like ffmpeg and Whisper, and those do the editing. Projects like this one save the commands as skills so you can repeat them.
Does this pipeline run on Windows?
It was built and tested on macOS with Apple Silicon. Other systems need changes, such as the font location.
Is it free?
The code is open source. You still need Claude Code, a computer that can run Whisper, and the time to set it up.