More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Explainer Director: script to narrated, animated video

    1

    Turn a script into a narrated, animated explainer (5-20 min) or a 45-second promo, rendered on your own machine. Beat map and approval gates first, 17 scene templates, word-synced voice and music.

    Free

    Security scanned
    Explainer Director: script to narrated, animated video

    Explainer Director: script to narrated, animated video

    Example session with this skill installed

    Make a 70-second explainer from this script. Narrator voice, plus a short dialogue between an engineer and their lead.

    "An AI answer can sound confident, and still be wrong. Ask a data agent how many orders there were. It says twenty thousand orders. Fluent, fast, and sure. But the truth was five times bigger: almost one hundred thousand. The model was fine. Five of nine tables were only partly imported. The problem was the data, not the model.
    LEAD: How do you know the answer is right?
    DEV: I count the rows, and compare them with the totals I already know.
    So do not just trust the answer. Check it. Three checks: read only queries, a log of every answer, and known totals to test against. Measure the agent, before you trust it."

    • Read your context and instructions
    • Compiled the explainer director

    Summary

    Split the script into 8 beats over 4 chapters (121 words, about 1:10). Every beat uses a built-in template, so no custom scene code is needed.

    Gate 1: please approve the beat map before I pick the look and render anything.

    Beat map: CHECK THE ANSWER

    beatatlenchaptertemplateon screenvoice
    b010:005sCold openkineticthe hook: "CONFIDENT" / "AND WRONG" stampNarrator
    b020:108sThe questionbigNumbercounts up to 20,666 "ORDERS, SAYS THE AGENT"Narrator
    b030:179sThe questionbarsagent 20,666 vs truth 99,441Narrator
    b040:329sThe causeprocessEXPORT → IMPORT → TABLES → QUERY → ANSWER, import cut "5 OF 9 PARTIAL"Narrator
    b050:409sThe causedialogueengineer and lead in an officeLEAD, DEV
    b060:554sThe fixsplitTRUST vs CHECKNarrator
    b070:597sThe fixiconsread-only SQL, audit log, known totalsNarrator
    b081:063sThe fixhero"MEASURE BEFORE YOU TRUST"Narrator

    Voices (Kokoro, offline): Narrator am_michael, DEV am_echo, LEAD af_heart.
    Check: every cue word is spoken, every beat has a template (8 of 8).

    Next steps

    1. You approve or edit the beat map (Gate 1).
    2. I render still sheets of each beat so you can pick the look (Gate 2).
    3. I generate the voices and music, then a half-resolution draft (Gate 3): out/example-draft.mp4, about 1 minute to render on a 4 GB GPU.
    4. After your OK, the final 1080p MP4.

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

    About this skill

    Hand your agent a script and get an upload-ready MP4: a long explainer (5-20 min) or a 45-second promo.

    How it works

    • Beat map first. The script is split into beats with a length estimate before anything renders.
    • Three approval gates. You approve the beat map, then the look (still sheets), then a half-resolution draft. Nothing slow runs before you agree.
    • 17 scene templates. Kinetic text, big numbers, bar charts, price comparisons, icons, world-map pins, split screens, process flows, quotes, dialogue, terminal, montage, glitch, script reveal and more. Most beats need no code.
    • Word-synced voice and music. Visuals land on the spoken word. Free offline voices (Kokoro) by default; ElevenLabs optional with your own key.
    • Runs locally. Renders on a 4 GB GPU. No cloud render, no telemetry.

    Requirements

    Node 18+, Python 3.10+, ffmpeg. The first setup downloads the Kokoro voice model (~330 MB) and asks before it does. All network calls are listed in SKILL.md under "Network access".

    About

    Built and maintained by me, Mohammad Hassan. Open source (MIT): https://github.com/mdhassan91/explainer-director

    16:9 only (1920x1080); no vertical 9:16 yet. English narration (Kokoro voices use US-English pronunciation). Tested on Linux (Ubuntu); macOS untested; on Windows use WSL. First setup downloads ~330 MB (voice model). Flat motion graphics, no photoreal faces.

    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

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 12 days ago

    • Free with an Agensi account

    Requires

    GitHub
    ElevenLabs
    Npmjs
    Pypi
    Node.js 18+
    Python 3.10+
    ffmpeg
    Remotion (installed by setup.sh)
    Python
    remotion
    Node.js

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables

    Allowed Hosts

    github.com
    api.elevenlabs.io
    registry.npmjs.org
    pypi.org

    Terminal: runs setup.sh and the render tools. Files: writes the video project in the user's folder. Network: one-time setup downloads from registry.npmjs.org, pypi.org and github.com (Kokoro voice model, SHA-256 checked). ElevenLabs is OPTIONAL: only used when the user sets ELEVENLABS_API_KEY (environment variable); the default Kokoro voices run fully offline.

    Listed13 days ago

    What's inside

    Frequently Asked Questions

    It is not a text-to-video model. Your agent plans the video as a beat map, you approve each step, and a local Remotion engine renders motion-graphics templates synced to the narration. You keep every file (script, beats, scenes) and can edit any of them.

    Node.js 18+, Python 3.10+ and ffmpeg. Tested on Linux (Ubuntu); on Windows use WSL. The 71-second example drafts in about 1-2 minutes on a laptop with a 4 GB GPU.

    No. The default voices (Kokoro) run on your machine after a one-time ~330 MB download, with no API costs. ElevenLabs voices are optional and use your own API key.

    One skill folder: SKILL.md (the director workflow with its approval gates), the engine (Remotion project with 17 scene templates, voice, music mix and render tools, plus a selftest), a flat-explainer style guide and a story-structure reference. MIT licensed, free.

    Yes, at three gates: you approve the beat map, then still sheets of every beat (the look), then a half-resolution draft. The full 1080p render runs only after your OK.

    New in Workflow Automation