More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    AI Video Editor: Raw Footage to Final Cut

    1

    Editing a talking-head video is mostly slow, careful work: finding every false start and retake, cutting each one cleanly, then timing captions and graphics to the words.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    AI Video Editor: Raw Footage to Final Cut

    AI Video Editor: Raw Footage to Final Cut

    Example session with this skill installed

    /edit-video Edit my footage into a tight 3-4 minute YouTube video (16:9) with captions, camera moves, a title and
    lower-third, and a light music bed. Use ElevenLabs Scribe for transcription, my key is in .env. Drop the false starts
    and retakes, keep the pace natural, and show me the cut plan before you cut anything.

    • Read your context and instructions
    • Compiled the ai video editor

    Result: an 18:03 phone recording became a finished 4:02 YouTube video in about 2 hours 25 minutes, including
    approvals and renders.

    Fileedit/final_youtube.mp4, plus edit/captions.srt (137 captions) for YouTube
    Picture1920×1080, 30 fps, H.264, 262 MB
    Loudness−14.0 LUFS, true peak −1.4 dBTP
    Cut255 transcript pieces → 102 segments; every join at −54 dB or quieter
    Camera24 moves: slow pushes, punch-ins on key words, a close-up on the twist
    LookKeynote dark with a violet accent: title, lower-third, a two-point funding card, end card
    MusicComposed for this video, about 24 dB under the voice
    Checks0 blank frames in 2,423 checked

    The steps it stopped for:

    1. Five questions about the brief: who it's for, the language, the names said out loud, the title and lower-third, the look.
    2. The cut plan: a table of what it kept, section by section, a list of every retake and false start it dropped, and what
      it trimmed to fit 3–4 minutes, with the measured length (3:59) before cutting anything.
    3. A listen to the clean cut, with what it fixed and what it measured.
    4. The camera, graphics and look plan: 27 beats, with the style it picked and why.
    5. One finished file, or layered files for another editor.
    6. Which preferences to save for next time.

    What it caught by itself:

    • After the first cut, seven lines had lost their first word. It found the cause (a short word followed by a tiny pause),
      fixed it for every cut, put back a word it had left out of the plan, and checked again.
    • It measured the music 28 dB under the voice, quieter than intended, and raised it.
    • It corrected its own earlier note that I leaned left in some takes, after checking the frames again.

    What it told me at handover: what it measured, that it can't hear (so the joins and music level needed my ears),
    which caption wordings follow my speech exactly, and how long each kind of change would take: text about 11 minutes,
    a different take or cut about 30.

    A follow-up, in a later session with the same pack: I asked for more, and it added 14 hand-drawn diagrams that
    sketch themselves as I speak, sound effects on each graphic, and a 46-second vertical Short cut from the same footage.
    Both are in the demo video.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Editing a talking-head video is mostly slow, careful work: finding every false start and retake, cutting each one cleanly, then timing captions and graphics to the words. AI Video Editor gives your coding agent a complete editing pipeline, 14 skills it follows step by step, so it does that work for you and stops for your approval at each decision. Your original footage is never modified.

    In our test, it turned 18 minutes of raw phone footage, full of false starts and retakes, into a finished 4:02 YouTube video with captions, camera moves, titles, cards and music. The session took about 2 hours 25 minutes from the first message to the finished file, including my approvals, a 10-minute frame-rate conversion and an 11-minute graphics render, with the footage on an external USB drive.

    How it works

    1. Checks your machine and your footage. Phones record at a variable frame rate, which makes sound drift out of sync at every cut, so it makes a steady copy first.
    2. Asks for the brief: platform, length, the names you say out loud, the title and lower-third.
    3. Transcribes every word, with ElevenLabs Scribe (your key) or free local transcription, and marks false starts, retakes, stumbles and long pauses.
    4. Plans the cut as a table: what stays, what goes and why, with the measured length. Nothing is cut until you say yes.
    5. Cuts frame-exactly on word boundaries, with short audio fades at every join, then checks its own cut: every kept word heard, every join quiet.
    6. Plans the dynamics: camera moves, captions, titles, graphics and the look, beat by beat. It waits for your go-ahead again.
    7. Builds it: smooth punch-ins and slow pushes, word-timed captions with the spoken word highlighted, titles and cards in one of seven styles, hand-drawn diagrams that sketch themselves as you speak, original music levelled under your voice, and sound effects on the graphics.
    8. Renders, checks every frame for blanks, normalises loudness and exports for YouTube, Shorts, TikTok, Reels, LinkedIn or X. You choose one finished file, or layered files (clean cut, transparent graphics, music, SRT) for Premiere, Resolve, Final Cut or CapCut.
    9. Hands over honestly: what it measured, what it can't check (it can't hear), and how long each kind of change takes.

    What you get

    • A cut you can trust: frame-exact, no clipped words, joins measured for clicks.
    • Camera moves that hold until the cut, the way an editor does it, and a vertical 9:16 crop around the speaker for Shorts.
    • A Short from the same recording: pick the lines, and it cuts and lays out a vertical version.
    • Captions burned in and as an SRT file for YouTube.
    • Seven styles, including Keynote dark and a Tech drawing style for diagrams, plus your brand colours and fonts.
    • Music and sound effects made on your machine: nothing to license, no library to search.
    • Carry-over: your tools, brand, styles and preferences are saved in ~/.video-studio/, plain files you can read and edit, so your next video starts further ahead.

    Requirements

    • A coding agent that can run terminal commands and write files. Built and tested with Claude Code.
    • ffmpeg, Node.js 22+ and Python 3.10+. The video-doctor skill checks these and gives the install command for macOS, Linux or Windows.
    • Optional: an ElevenLabs API key for fast transcription (Scribe). Without one, transcription runs locally for free, more slowly.
    • About 1.6 GB of disk on your home drive, once, shared with AI Video Studio if you have both. Your footage can live on any drive.

    What it doesn't do

    It doesn't sync multiple cameras, colour grade, or generate footage. It's built for one speaker on one camera (talking heads, pitches, tutorials and screen recordings). It can't hear: it measures the audio, but the final listen is yours.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Video Skills.

    • One speaker, one camera: no multicam sync, colour grading or generated footage.
    • It's a long agent session, not a quick prompt: our 18-minute recording took about 2 hours 25 minutes end to end and
      uses a meaningful share of an agent plan's usage. The test used a top-tier model at high effort.
    • Renders are CPU-heavy: the graphics layer for 4 minutes took 11 minutes on our machine. On 8 GB of RAM, use 2 workers.
    • The agent can't hear. It measures joins, loudness and music level, but the final listen is yours.
    • Our full 18-minute test used ElevenLabs Scribe. The free local transcription is tested on short clips, and it's slower.
    • A vertical crop from 1080p footage is enlarged about 1.8×, so it's a little softer; 4K footage crops cleanly.
    • Tested most in English.
    • Built and tested end to end with Claude Code. Other coding agents that read SKILL.md should work, but haven't been tested end to end.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 7 days ago

    Accounts
    ElevenLabs
    Software
    PythonNode.jsFFmpeg
    Google Chrome
    orChromium (downloaded automatically if missing)

    ElevenLabs (Scribe) for fast transcription, with your own API key.

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables

    Allowed Hosts

    api.elevenlabs.io

    File Scopes

    `./**`
    (the
    project
    folder)
    `~/.video-studio/**`
    (shared
    tools
    saved
    brands
    styles
    and
    preferences)
    `~/.cache/hyperframes/**`
    (renderer's
    browser)
    `~/.cache/huggingface/**`
    (local
    speech
    model)

    Nothing is installed without asking. Your original footage is never modified: it works on copies in `edit/`. Network access is used at setup to download the renderer and models from npm, GitHub and Hugging Face, fonts from Google Fonts, and, if you choose Scribe, to send the audio to ElevenLabs for transcription. API keys are read from `.env` and never printed.

    Listed7 days ago

    What's inside

    Frequently Asked Questions

    No. It works on copies in your project's `edit/` folder, and your original files are never touched.

    One person on one camera: talking heads, pitches, lectures, tutorials and screen recordings, from a phone, webcam or camera. Variable frame rate phone footage is detected and converted first.

    Yes. It shows a table of what it keeps and drops, with the reasons and the measured length, and cuts nothing until you say yes. You can put pieces back or drop more first.

    Yes. Ask for layered files: the clean cut with levelled audio, the graphics as a transparent video, the music and the captions as an SRT.

    Yes, from vertical footage or by cropping horizontal footage to 9:16 around the speaker. It can also cut a Short from the same recording as your main video.

    No. It's optional, for faster and more verbatim transcription. Without it, transcription runs on your machine for free.

    The music and sound effects are synthesised by code for your video, not taken from a library, so there's no licence to buy and no stock track for a matching system to recognise. No one can promise what a platform's systems will do, but they're original.

    No. AI Video Editor works on its own. If you have both, they share the same tools, brand and styles.

    No. It makes files on your machine. Uploading is up to you.

    New in Image & Video