Talking Explainer

A Claude Code skill that turns a topic into a narrated motion-design explainer. A cartoon presenter really talks: its mouth follows your voice, it changes pose on cue words, and it blinks. Short text and clean scenes animate in time with every word you say, and your brand opens, sits in the corner of, and closes every film.

1920x1080 at 60fps, with a 9:16 cut for TikTok, Reels and Shorts when you ask for one.

Created by Kenneth Villar.

Install  ·  Your first film  ·  How a film gets made  ·  Your own presenter  ·  Troubleshooting

See it talk

This GIF was made from the skill's own presenter art. The mouth shapes come from the same rules the film engine runs on your recording (here, a computer voice reading one line). He waves on "Give", presents on "your", points on "I" and gives a thumbs-up on "words", and every word appears on screen as it is said.

What you get

• A script in your voice. Claude researches the topic, then writes a hook-first script in the way you talk. You can ask for any language.
• A presenter that talks. Six mouth shapes per pose, picked frame by frame from your take. Poses change on the words you choose. Blinks are added automatically.
• Motion that lands on the word. Your take is transcribed to word timings, and every entrance, camera push and highlight is keyed to the word that triggers it.
• Your brand. Your logo or wordmark opens the film, sits in the corner and closes it on your call to action.
• Proof before you post. The finished render is transcribed again and checked against the script timings, so you know the picture lands on the words.
• Optional extras. A portrait cut for short-form video, your own character as the presenter, and a music bed you have the rights to.

What you need

Install

1. Install the tools. Skip anything you already have.

Windows (PowerShell):

winget install OpenJS.NodeJS.LTS Python.Python.3.12 Gyan.FFmpeg
py -3.12 -m pip install numpy pillow scipy faster-whisper

macOS (Terminal, needs Homebrew):

brew install node python@3.12 ffmpeg
python3.12 -m pip install numpy pillow scipy faster-whisper

Then the video engine's helper skills (both systems):

npx hyperframes skills update general-video hyperframes-animation media-use

2. Get the skill. Either:

• download talking-explainer-skill.zip from the Releases page and unzip it, or
• clone this repository: git clone it, then cd into the folder.

3. Install it. Open a terminal in the folder that has install.mjs and run:

node install.mjs

This copies three skills into ~/.claude/skills/:

Then it checks your machine. Every line should say OK; anything marked MISSING comes with the command that fixes it. Restart Claude Code once it is done.

Options: --dest <folder> installs somewhere other than ~/.claude/skills/; --force replaces folders with the same names that this installer did not create.

Your first film

In Claude Code, in any folder, say:

make a talking explainer about how email marketing works

The first time, Claude sets you up (once)

1. Missing tools - the setup check lists them with install commands. Claude never installs system tools without asking.
2. Your brand, in one message: brand name, logo (optional; without one you get a text wordmark), text colour and accent colour, an opener word ("explains"), a call to action ("Follow for more"), the narration language and where films are saved.
3. Your voiceover route - an ElevenLabs voice clone, or recording yourself. You need one of the two.
4. Your speaking style - answer five quick questions, or give Claude the transcript of a 5+ minute unscripted recording of you talking and it measures your style from that.
5. Your opener and closer lines, word for word, with your first name.
6. Your presenter - the bundled demo, or your own character (see below).
7. Music (optional) - a track you own, the HeyGen catalog (npx hyperframes auth login once), or none.

All of it is saved to ~/.talking-explainer/profile/ and every film after that uses it.

Every film, Claude asks you for

1. The topic.
2. One MP3. Claude gives you the script as plain text you can paste into ElevenLabs or read aloud, then waits for your recording.
3. A yes on the storyboard - the list of scenes, one per paragraph, before anything is built.
4. A privacy check, only when the film shows your own app or setup: a list of what will appear on screen, for you to approve first.

When it is done you get the MP4 and the credit line for your video description.

How a film gets made

1. Script - researched facts, a hook-first opener, one everyday analogy, about twelve short paragraphs.
2. Your take - one recording of the whole script. The visuals are built around your timing, never the other way round.
3. Timings - faster-whisper turns the take into word-by-word times. These are the cue sheet for everything else.
4. Storyboard - one frame per paragraph, each with a camera push and an emphasis beat on a chosen word.
5. Scenes - parallel AI workers build the scenes against one shared plan. Claude reviews each one from contact sheets and sends fixes back to the same worker.
6. Presenter, audio, brand - poses and lip sync from your take, the voice loudness-mastered with music ducked under it, your logo in the corner.
7. Render and proof - a 60fps render, a seam check between every scene, then the sync check on the rendered file.

Your own presenter

The demo presenter has seven talking poses (above), six mouth shapes for each, and blinks. To use your own character, say:

make my own talking character

Claude gives you three image prompts to run in an image tool with a reference picture of your character: one sheet of poses, one of faces and one of effects. It cuts the results into sprites, fits the blinks and the six mouth shapes, and shows you review sheets to approve before any film uses them.

More you can ask for

Where things live

Updating

Get the new version (a new release zip, or git pull) and run node install.mjs again. Your profile is kept.

To uninstall, delete the three skill folders from ~/.claude/skills/ and the ~/.talking-explainer/ folder.

Troubleshooting