Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Editor de video: reels con subtítulos y cortes

    1

    An autonomous video editor for Claude Code built on Hyperframes 0.8.

    $15

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Editor de video: reels con subtítulos y cortes

    Editor de video: reels con subtítulos y cortes

    Example session with this skill installed

    /editor-video Here's a 40-second talking-head clip shot horizontally on my phone, plus a cumbia loop. Make it a reel for Instagram: cut the silences, vertical, karaoke captions, and put the music low under my voice.

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

    This is the flow the skill runs on a talking-head + music request. The numbers marked measured come from our own validation project (a ~40 s talking-head and a music loop); the rest describes what each step does.

    1. Analysis

    • analizar.py reads the clip (orientation, HDR, fps, loudness, silences, scene changes, face position) and builds a contact sheet that Claude looks at before planning.
    • Mode chosen: A. talking-head → reel. No questions needed when destination and tone are in the request.

    2. Storyboard (STORYBOARD.md)

    MomentBlockEffect
    First secondHook: strongest sentence rewritten to ≤6 wordstitulo slam + impact SFX
    BodyTalking head with jump cutsjumpZoom on the cuts
    Whole videoKaraoke captions, 3-4 words per line at ~66% heightactive word in accent color, emojis on keywords
    Hard dataCountercontador outside the Reels UI zone
    Last 2 sClosing line / CTArise title

    3. Processing

    • preparar.py --formato 9:16 --modo cara: vertical reframe following the face (YuNet, smoothed, with dead zone).
    • Local transcription with Parakeet: measured 40 s of audio in 17 s.
    • cortar.py: measured about -14% duration with no real words lost.
    • audio_pro.py: voice to -16 LUFS, music bed to -20 LUFS, ducking under speech.

    4. Render + QA

    • render.sh final: lint, GPU render with 5 workers, master at -14 LUFS / TP ≤ -1, plus a lighter WhatsApp copy.
    • Measured output: 34.4 s, 1080×1920, rendered in about 2 minutes on an RTX 4060 laptop.
    • qa.py: black frames, freezes, silences, loudness, true peak and a safe-zone sheet. If the verdict is REVISAR, Claude fixes and re-renders.

    Delivered

    • out/<name>.mp4 and out/<name>-whatsapp.mp4
    • A short note of what was decided without asking: hook text, cut thresholds, caption style, music level.

    Next steps

    • Tell it what to change ("bigger hook", "less zoom") and it re-renders.
    • If you love the style, it packages it as /editor-<name> so the next video comes out the same way with no questions.

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

    About this skill

    An autonomous video editor for Claude Code built on Hyperframes 0.8. You drop in raw material (talking-head clips, phone videos including iPhone HDR, photos, WhatsApp voice notes, music) and say what you want. Claude analyzes the footage, decides the edit like a senior editor, renders it, and looks at its own work before showing it to you.

    What it does

    • Analysis first: metadata (HDR, rotation, fps), loudness, silences, scene changes and face position, plus a contact sheet Claude actually looks at before writing anything.
    • Jump cuts at word level: combines the transcript with the real audio silences, removes filler words without eating real words, with 50/80 ms margins and 30 ms fades. Measured: about 14% shorter with no lost words.
    • Reframe 16:9 → 9:16: smoothed face tracking (YuNet) with a dead zone, or blurred background when there are two people.
    • Pro audio: highpass, denoise, EQ, de-esser, compressor, two-pass loudnorm. Voice at -16 LUFS, music bed at -20 with automatic ducking, final master at -14 LUFS / TP ≤ -1.
    • Composition kit (hf-kit.mjs): karaoke captions with emphasis and emojis, jump-zooms on cuts, kinetic titles, counters, lower thirds, stickers, progress bar, beat-synced pulses, SFX aligned by their peak.
    • Beat sync: BPM, beats, downbeats, drop and sections from the music to place cuts and flashes.
    • QA loop: lint, snapshots, then a final check for black frames, freezes, silences, loudness, true peak and a safe-zone sheet for Reels / TikTok / Shorts. If it fails, it fixes and re-renders (max 3 passes).

    Edit modes

    Talking-head → reel · photo/video/voice-note montage (birthdays, tributes, memes) · music → beat-synced video · faceless explainer · product promo without footage · highlights from a long video. It also packages a style you approve as your own reusable skill (/editor-<name>).

    Requirements

    ffmpeg, Node 20+ (Hyperframes runs via npx), Python 3.10+ with opencv-python, numpy, librosa. Built and tested on Windows 11 with an NVIDIA GPU (a 34 s reel renders in about 2 min). On macOS/Linux the Windows-only GPU preference step is skipped; without a GPU it works but slower. Transcription is local (Parakeet), no API key needed.

    Known limitations

    • No music or SFX included: bring your own licensed assets (the skill indexes your library and aligns SFX by peak).
    • Word-level cutting is very good, not perfect; for swallowed consonants raise the padding or cut by hand.
    • Face tracking follows the largest face; with people taking turns, use blur mode or cut by segments.
    • Skill instructions are written in Spanish; Claude answers in your language.

    Word-level cutting may require manual padding for fast speech. Face tracking follows the largest subject only. High-performance rendering requires an NVIDIA GPU and Windows 11/WSL.

    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 any payment.

    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 today

    Requires

    hyperframes
    Node.js
    Python

    Permissions

    Terminal / Shell
    Write Files
    Browser
    Network Access
    Read Files
    Environment Variables
    Listedtoday

    What's inside

    Frequently Asked Questions

    This skill is specifically optimized for Claude Code, leveraging its ability to execute local scripts to orchestrate complex FFmpeg, Python, and Node.js workflows directly on your machine.

    The skill handles the heavy lifting of professional editing, including silence removal, 9:16 face-tracking reformatting, audio normalization (LUFS), and the generation of beat-synced motion graphics and captions.

    While the skill runs via Claude Code, optimal performance requires a local GPU for rendering, along with Node.js 20+, Python 3.10, and FFmpeg installed on your system.

    The skill folder: SKILL.md with the full editing playbook, the Python and Node scripts (analysis, face-tracking reframe, word-level cutting, audio mastering, beat analysis, composition kit, render and QA), the YuNet face-detection model and three reference guides. Hyperframes itself is not bundled: it is an open-source tool that runs via npx. No music or SFX are included; you use your own licensed assets.

    The skill uses YuNet face-tracking to automatically center the primary subject when converting 16:9 footage to vertical 9:16, or it can apply an aesthetic background blur for wide-angle shots.

    The skill was built and verified against Hyperframes 0.8 (v0.8.140). Hyperframes is maintained by its own authors, so a future major version could change its CLI; the skill documents the exact version it was tested with so you can pin it if needed.

    New in Image & Video