SkillAgentSearch skills...

aigent-talking-head

Free, open-source talking-head video editor: turn one phone clip into a captioned reel for TikTok, Reels and Shorts. 51 effects, 29 motion graphics, 56 transitions, local Whisper captions, kinetic typography and an MCP server. Works with Claude Code, Cursor, Codex and Gemini CLI. Apache-2.0.

Install / Use

claude mcp add wrg32786 -- npx -y github:wrg32786/aigent-talking-head

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

74/100

Category

Automation

Supported Platforms

Claude Code
Claude Desktop
Cursor
Gemini CLI
OpenAI Codex

Our assessment of aigent-talking-head

aigent-talking-head scores 74/100 on our quality scale, 2529th of 2,758 Automation skills we index.

Its MCP Server is 51 KB long, well organised into 36 sections with 10 code examples: long enough that it reads more like full documentation than a focused instruction file, which agents can find harder to follow.

It has 3 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
21/30
Structure
20/20
Description
15/15
Adoption
3/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated today, so aigent-talking-head is actively maintained.
  • It is released under the Apache-2.0 license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

No issues found

Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands.

Automated pattern scan on 2026-11-22. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

aigent-talking-head compared with similar skills

All 4 of these similar skills score higher than aigent-talking-head; compare them before choosing.

SkillScoreStarsUpdatedFormat
aigent-talking-head (this skill)by wrg32786743todayMCP Server
Agent-Reachby Panniantong100114.0ktodayCLAUDE.md
hyperframesby heygen-com10088.2ktodayCLAUDE.md
rufloby ruvnet10082.3ktodayCLAUDE.md
headroomby headroomlabs-ai10080.5ktodayCLAUDE.md

Frequently asked questions

How do I install aigent-talking-head?
Run claude mcp add wrg32786 -- npx -y github:wrg32786/aigent-talking-head. The install tabs above show the steps for each supported agent.
Which AI agents does aigent-talking-head work with?
It is written for Claude Code, Claude Desktop, Cursor, Gemini CLI and OpenAI Codex, as a MCP Server file. Other agents that read the same format can often use it too.
Is aigent-talking-head safe to use?
Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. It is Apache-2.0-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is aigent-talking-head still maintained?
The repository was last updated today, so aigent-talking-head is actively maintained.

aigent-talking-head

aigent-talking-head is a free, open-source (Apache-2.0) talking-head video editor that turns one phone clip into a captioned vertical reel for TikTok, Instagram Reels and YouTube Shorts, run by you or your coding agent (Claude Code, Codex, Cursor or any MCP client) on your own machine with Whisper, FFmpeg and Chrome.

Made by Will Gwyn at The AIgent.

GitHub stars

If it saves you an edit, a ⭐ helps other creators find it.

Quick start · Studio · Install · Features · How it compares · FAQ · Show your reel · Agent skill and MCP · Guide · Full gallery · Get guidance

Before and after: a plain phone clip on the left, the finished reel with kinetic captions and graphics on the right

Phone clip on the left, finished reel on the right. Watch the 40-second demo reel (20 effects) or read the plan files behind it.

What it does

  • Transcribes your clip locally with Whisper and cuts it into clean sentences, keeping the best take of each.
  • Captions every word on time, with kinetic hero words and titles that can sit behind your head.
  • Plans graphics from what you say: a number becomes a stat callout, a list a checklist, the close a call-to-action card. The words on a graphic are words you said.
  • Keeps your face clear: a title or graphic set behind you goes behind only your hair and crown, never your face, and a layout pass moves captions out of the way of graphics.
  • Exports for each platform inside its safe area, then checks the finished file (loudness, overlaps, safe area) before you post.
  • Renders fast: overlay frames render about 5x faster than before (163 s down to 31 s for a 30 s reel in our benchmark), and a whole render is about 2.8x faster.
  • Works from your agent: a one-line skill install for Claude Code, Codex and Cursor, plus a local MCP server.

Quick start

One line for your agent. Paste this into Claude Code, Codex, Cursor, Gemini CLI, OpenCode or any agent that runs commands:

Read https://github.com/wrg32786/aigent-talking-head/blob/main/docs/install.md and follow it to install, set up and connect Talking Head Studio in this agent, then open Studio.

It detects your OS, clones the kit (or reuses your copy), installs everything, connects itself to Studio over MCP, opens Studio and tells you only what it can't do itself (docs/install.md).

Just the command-line kit and skills. Paste this instead:

Read https://raw.githubusercontent.com/wrg32786/aigent-talking-head/main/INSTALL.md and follow it to install aigent-talking-head in this agent.

Then ask it: "Make a reel from the newest clip in my inbox."

By hand. Install once (Install), then npm run smoke should end with SMOKE PASS. Record yourself upright on your phone, drop the file in inbox/, and:

npm run reel -- inbox/my-take.mov --name my-reel --draft     # 1. transcribe and cut (Whisper small.en)
npm run plan -- my-reel                                      # 2. graphics proposed from what you said -> jobs/my-reel/Reel.md
#                                                              3. review the Reel.md: keep, change or delete lines
npm run reel -- --recipe jobs/my-reel/Reel.md --clips jobs/my-reel/clips --name my-reel --force --export reels   # 4. render
npm run verify -- my-reel --export reels                     # 5. VERIFY PASS, then look at jobs/my-reel/verify/contact.png

You get jobs/my-reel/reel-v1.mp4 and reel-v1-reels.mp4 (inside Instagram's safe area). Exports: tiktok, reels, shorts, feed45 (4:5), square.

  • Just captions and titles, one command: npm run reel (newest video in inbox/).
  • Change something: edit jobs/<name>/build/captions.js or build/cut.json, then npm run reel -- --rebuild --name <name>.
  • Accuracy over speed: --model large-v3 (about 3 GB, downloaded once, a notice prints first).
  • Everything else (looks, edit rules, themes, manual steps, presets, the file map) is in the user guide. Every command and flag is in contract.json.

Made a reel? Star the repo and share it.

Studio: edit in a browser, with your agent in the chat

Studio is a local page for making a reel without memorising commands. It walks you through six steps, shown as numbered dots across the top: Footage (drop in clips, paste a link or film it right there), Cut (pick what to make and which lines to keep), Look (one Look for every layer, or your own picks), B-roll, Finish (draft, check and render) and Post. The chat on the page is answered by your own agent (Claude Code, Codex, Cursor or any MCP client), so it can change the edit for you while you watch. Everything runs on your computer. Full guide: docs/studio.md.

Set it up by telling your agent:

Read https://github.com/wrg32786/aigent-talking-head/blob/main/docs/install.md and follow it to install, set up and connect Talking Head Studio in this agent, then open Studio.

Your agent installs everything up front (Python, the downloaders, the Whisper model, the effects, the humanizer skill), registers the MCP server in its own config, opens Studio and listens, and tells you only what it can't do itself, like a password prompt. In Studio, Connect agent walks you through Claude Code, Codex, Cursor, Gemini CLI, OpenCode or any MCP client with your install's details filled in, and turns green when the agent connects. Step by step for each agent: docs/setup-with-ai.md and the Studio section of AGENTS.md.

No agent yet? Tell any agent that runs commands to "install Talking Head Studio and put the icon on my desktop", and it runs our one-line installer for your OS (docs/install.md). Then click the "Talking Head Studio" icon. Without an agent connected, the chat works on keywords, a local model or an API key.

<img src="docs/studio/desktop.png" width="720" alt="Studio in dark: the AIgent wordmark, the six step dots with Cut open, Wizards, Outputs and Claude Code in the top bar, Your edit across the top, the Pieces sorted into Hook, Story, Tip, Proof, List and CTA, Ask my agent next to Look at the bottom, and the chat with Claude Code on the right"> <img src="docs/studio/light.png" width="720" alt="The same Cut step in the light theme">

Your edit. A strip at the top of every step says what you are making in plain words: the format, where it comes from ("7 pieces, 25 s, in your order, picked by AI"), the Look, the b-roll and the sound, with one button to the next thing to do (Add footage, Draft it). Click any part to jump to the step that sets it.

Reference links. Paste one or more links (YouTube and Shorts, TikTok, X, Vimeo, a direct .mp4, or one public Instagram post) and Studio downloads each to a folder on your computer, shows the progress, and breaks it down: length, shots and cut times, pace, music BPM and energy, caption style, transitions, finishes and graphics. It then suggests matching items from the effects menu. The breakdown goes into the chat, so "use the transitions from the reference but calmer" works. Use it as an influence (soft defaults) or recreate its structure with your clip. Link downloads need yt-dlp, which the installer adds; Instagram uses instaloader anonymously, with no login. Only use videos you have the rights to use as a reference.

<img src="docs/studio/ref-url-desktop.png" width="720" alt="Studio Look step, Reference view: a pasted YouTube link broken down into 90 seconds, 17 shots, 123 BPM and a 5.29 second average shot, a shot timeline with a beat grid, tags for what was found, suggested picks as chips, and a contact sheet">

Cut. Tell Studio what to make from the take with big cards: a Short (Reels, Shorts, TikTok), a Long form YouTube video or a Square post, short, medium or long, one full edit, clips or both, organic or an ad. It picks smart defaults from the take, so you can just hit Next. Find cuts on the Cuts and lines tab lists the best parts with the words, times, a score and why (picked by your agent, a model, or simple rules, and it says which), as colored blocks on the take's timeline to trim, keep, drop or add to. Each cut lands on the quiet in the audio after the last word, so no word is clipped. For a long recording, Lines and bites lists every line with keep or retake and groups them into named bites in any order. On the Reference step, Pull script puts the reference's words in your Scripts with the hook, body and CTA marked, and Adapt to my topic rewrites it for yours. More.

Pieces. Once the take is transcribed, every line is a piece sorted by what it does (Hook, Story, Tip, Proof or quote, List, CTA; asides and retakes tucked away). Let AI choose builds your edit, hook first and CTA last with a reason each, or Build my own and drag to reorder. Pieces are whole lines, so a reel never cuts a word.

<img src="docs/studio/cut-pieces.png" width="720" alt="Studio Cut step on Pieces in dark: lines sorted into Hook, Story, Tip, Proof or quote, List and CTA, numbered in play order"> <img src="docs/studio/cut-pieces-light.png" width="720" alt="The same Pieces view in the light theme">

<img src="docs/studio/cut-desktop.png" width="420" alt="Studio Cut step: the summary line, cards for format, length, output and goal, Find cuts, and how many cuts you kept"> <img src="docs/studio/cut-timeline.png" width="420" alt="Cut candidates as colored blocks on lanes under the take's waveform, with filler words marked">

Record now. No clip yet? Film it in the Footage step with your computer's camera and mic. Pick the camera and mic, choose 9:16 (the default), 16:9 or 1:1, and record with a 3, 2, 1 countdown, pause and a timer. Framing guides show where your head goes so titles can sit behind it and captions below, and an optional teleprompter scrolls your script near the lens. Every take is listed with a thumbnail: keep it, drop it from the edit, or retake. A kept take is converted to an mp4 and joins your footage right away, and your agent gets a one-line note ("New take recorded: 42 s"). The recording is sent to Studio in small pieces as you film, so a long take is not lost to one failed upload, and nothing leaves your computer. Browsers only allow the camera on a secure page, which http://127.0.0.1 is. On a phone, open Studio on the phone itself or over https; Studio never opens itself to your network.

<img src="docs/studio/connect-desktop.png" width="520" alt="Connect agent in Studio: pick your agent from a grid, then numbered steps with the claude mcp add command and the hook command filled in with this install's paths, each with a copy button, and the status waiting for your agent"> <img src="docs/studio/connect-phone.png" width="190" alt="Connect agent on a phone, as a full-screen sheet">

Footage from a link. Paste a public Dropbox or Google Drive share link, or a plain video URL, and it downloads into your footage with its size and progress, past Drive's virus-scan page for big files. Private links say they need a login.

<img src="docs/studio/footage-link.png" width="520" alt="The Footage step with a link box: a Google Drive file downloaded into the footage, and a private link with the line This link

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars3
CategoryAutomation
Updated1h ago
Forks0

Languages

JavaScript

Trust signals

92/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 low