talking-head-reel skill
A skill that turns a messy phone recording into a vertical reel: best takes, snap zooms on cuts and word-timed captions.
- 70 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill runs `ffmpeg`, `whisper` and `npm install` on your machine and downloads memes from the network.
- The agent reads the transcript of your speech.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/talking-head-reelDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill runs `ffmpeg`, `whisper` and `npm install` on your machine and downloads memes from the network.
- The agent reads the transcript of your speech.
This is third-party code. Review the repository files before installing.
What it does
talking-head-reel assembles a vertical clip for Instagram, TikTok or Shorts from a phone recording of a person talking to camera and repeating every line two or three times. The skill transcodes the video to 1080x1920 at 30 fps, transcribes the whole recording with word timestamps, picks the best take of each sentence and cuts on word boundaries. Every cut gets a snap zoom so the jump cuts read as a style. On top of the video come three-word captions, logos and callouts landing on the spoken word, numbered lists, quote cards, memes and an end card. Everything is built in a `Remotion` 4 project inside the skill folder, and the finished `.mp4` lands on the Desktop.
Who it is for. Short-video creators who record a monologue to camera on a phone and spend hours picking takes and editing.
Good fit when
- Five minutes were recorded and only ninety seconds of speech are usable.
- A reel needs word-timed captions and callouts that land on what is said, with no manual editor work.
- The author has the intended script and pastes it into the request so takes are picked more accurately.
- The clip goes to Instagram and the layout must clear the app interface.
Not a fit when
- The recording is landscape or has several people: the skill targets portrait video with one speaker.
- The computer does not run macOS: the instructions are written for macOS and `brew`.
- You need to edit other people's footage or cut interviews.
Example request
Use talking-head-reel to edit `~/Downloads/IMG_1234.MOV` for Instagram: I did many retakes, pick the best parts, remove the pauses, add zooms and animations.Limitations
Requires macOS, `ffmpeg`, the `openai-whisper` command (the `turbo` model downloads about 1.5 GB on first run), Node 20 or newer and Claude Code. A run takes 10 to 15 minutes, mostly rendering. Only what the speaker said appears on screen: the skill makes speech visible and adds nothing of its own. The example timeline in the repository fits a fictional recording, so your own recording has to go through the pipeline. Memes are fetched from the imgflip template API during editing and are not stored in the repository.
How to disable. Delete the `~/.claude/skills/talking-head-reel` folder. Remove the installed `ffmpeg`, `openai-whisper` and the downloaded model separately.
FAQ
How do I install talking-head-reel?
Run `git clone https://github.com/mariagorskikh/talking-head-reel ~/.claude/skills/talking-head-reel`, then `npm install` inside `remotion`. After that give Claude Code the path to the recording.
What does it need?
macOS, `ffmpeg` (`brew install ffmpeg`), `openai-whisper` (`pip install openai-whisper`), Node 20 or newer and Claude Code.
How long does editing take?
Ten to fifteen minutes of Claude working, most of it rendering.
Does a script help?
Yes. Paste the text the author meant to say into the request; it improves take selection.
Where do the memes come from?
The skill fetches them from the imgflip template API during editing; they are not part of the repository.
Discussion
No questions or reviews yet.
Related
SkillsClaude Code skill that turns raw talking-head clips into a finished Instagram reel, picks the best takes and cuts dead air
Eleven Claude Code skills turn a talking-head clip into reels: titles, graphics behind you and split screen
A skill that turns a topic into a narrated 4K video: research, script, local TTS, Remotion composition and an FFmpeg mix
The video-explainer skill turns a script into a vertical 9:16 explainer: narration, captions, Remotion and a pre-render review
Claude Code skill that edits a vertical reel from raw talking footage: cuts pauses and slips, adds animated interface pieces.
A skill that turns a coding agent into a studio for voiceover explainer videos: word-level sync, a motion library and Remotion assembly
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| talking-head-reel (this card) | no | 70 | |
| Video Editor Skills | no | 443 | editing skills where the agent remembers the author's notes; this one is a ready pipeline for talking to camera |
| HyperFrames Motion Reel | no | 185 | builds a showreel from a topic line; this one builds from your own recording |
README in short
The pipeline is a set of scripts for prep, transcription, cutting, check frames and render. The folder holds rules for take selection, vertical layout, Instagram safe zones and zooms, an example timeline and a minimal `Remotion` project with Geist fonts and a sound kit.
README badge
Original SKILL.md
---
name: talking-head-reel
description: Turn a long, messy phone recording of one person talking to camera (every line taken several times, false starts, pauses, "okay, again") into a vertical Instagram / TikTok / Shorts reel, about a minute long, with the best take of each sentence, snap zooms on the cuts, word-timed captions, and animated overlays that amplify what the speaker says (logo chips as they name things, numbers as big callouts, lists that get stamped, a quote card, a typed prompt, polaroids of the other takes, memes, an end card). Use whenever the user hands over a portrait phone recording (IMG_xxxx.MOV) and says "edit this for Instagram", "make a reel", "vertical", "I did a lot of retakes, pick the best parts", "remove the duplicates and pauses", "add animations and zoom in and out", "make it entertaining", or wants a re-cut of a previous reel ("swap the closing take", "change the callout at 0:40", "add music"). Not for landscape walkthroughs with a screen recording.
---
# Talking-head reel
The raw material is one person on a phone, portrait, talking to camera for
five minutes to get ninety seconds of usable speech: every sentence is said
two or three times, the best version is usually the last, and the file ends
with an outtake. The deliverable is a 1080x1920 reel of the best take of
each sentence, cut on word boundaries, with framing snaps on every cut so
the jump cuts read as a style, and overlays that land on the word they
illustrate. Nothing on screen says anything the speaker did not say; it
makes what they said visible.
`assets/Reel.example.tsx` and `assets/reel-segments.example.json` are a
placeholder timeline built on the timings of a fictional recording, so
they compile and show the wiring, the pacing and the density; read them
once before planning a new reel. `overlays.tsx` holds the shared `Card`,
`Big`, `LogoRow` and `Word`; `reel-overlays.tsx` holds the vertical
grammar.
The Remotion project is `remotion/` next to this file (`cd remotion &&
npm install` once). Every path below that says `src/`, `public/` or
`out/` is inside it, and the scripts are run from inside it.
## What "good" looks like
The speaker is centred and fills the width. Under the chin, a band of dark
cards that change every few seconds: a row of logo chips popping one per
spoken name, a big number, a numbered list that fills in as things are
listed and then gets a red stamp slammed across it when they are
dismissed, a quote card whose last line lands in the accent colour, a
claim struck through, a terminal typing a prompt, three polaroids of the
other takes, two big logos bouncing at the close. Next to the head,
top-right on the wall, a "reaction slot": a logo that appears big and
alone on its name and is gone a second later, a big background-free emoji
on a feeling, a meme popping in on a punchline. Under the band, three-word
captions with the spoken word in the accent colour. The framing snaps
between 1.0 and 1.1 at every cut and at most sentence starts, and pushes
in slowly on the line that matters. The last frame freezes under an end
card with the thing the speaker wants people to click. All of it above
the Instagram UI.