Your guide in. A finished video out.
A script from your material
Claude reads your guide or procedure and writes for the ear: one action per sentence, button names exactly as they appear on screen. You approve it before anything is recorded.
A voice made on your computer
Natural British and American voices, or clone a voice you have permission to use. Whisper listens to every sentence and catches any word that came out wrong.
Your screens, in sync
Phone apps in a hand or a clean device frame, desktop software in a browser window. Highlights, taps and zooms land on the word that names them.
An AI presenter and scenes
Optional: a presenter who opens and closes the video, lip-synced by HeyGen, and short animated scenes. Every image and clip is checked for AI glitches.
Motion graphics built in
Topic chips, numbered step titles with progress, a halfway recap for longer videos and a closing recap that ticks off what was covered.
Captions and music
Captions that light up word by word, plus an .srt file for your LMS. A calm music bed you own, mixed well under the voice.
Made with the skill.
A short how-to for Larkspur, a made-up company and app: claiming a work expense from your phone. The script, voice, edit, motion graphics, captions and music all came from the skill; the presenter and scenes from the optional AI steps.
Claude Code, and two optional keys.
- Claude Code Required
- On a Mac or a Windows PC. The skill runs tools on your computer, so it needs Claude Code rather than the chat app.
- Python and ffmpeg Required
- Free. If they aren’t installed, Claude walks you through it.
- OpenRouter key Optional
- For the presenter photo, scene images and animated scenes, and the automatic checks. A few cents to under a dollar each.
- HeyGen key Optional
- For the talking presenter. Needs HeyGen API credit, which is separate from a HeyGen plan.
No keys? You still get a complete video: the voice, your screens in a drawn phone or browser, highlights, motion graphics, captions and music. Add the AI presenter later if you want one.
Three steps.
SKILL.md sits directly inside it:Mac:
~/.claude/skills/ai-training-video/Windows:
C:\Users\<you>\.claude\skills\ai-training-video\You approve each stage. You pay only for the AI parts.
- 1Brief. Claude asks only for what it can't find: your material, screenshots, logo, brand colour, presenter and voice.Free
- 2Script. Written from your material, shown to you, changed until you're happy.Free
- 3Voice. Every sentence read locally and checked by Whisper; you listen to a preview.Free
- 4Edit. Each screen, highlight, tap and callout planned against the spoken words, with preview stills to check.Free
- 5Images. The presenter, scenes and phone-in-hand shots, each checked automatically.Cents each
- 6Video. Scenes animated, the presenter lip-synced, then rendered with captions and music.A few dollars
- 7Review. Give timestamps; Claude makes the fix and re-renders only what changed.Free
Every paid step keeps what it made, so running it again never pays twice. A typical video with a presenter and a few scenes costs a few dollars in total.
Does it work in the Claude app?
It needs Claude Code, because the voice, the checks and the rendering run on your computer. Claude Code is available on Mac and Windows.
Can it do an Australian (or any other) accent?
The built-in voices are British and American. For any other accent, or your own voice, record 15–60 seconds of clean speech from the person who’ll be the voice, with their consent, and the skill clones it.
What about desktop and web software?
Use browser mode: your screenshots sit in a clean browser window, with highlights, clicks and zooms. Phone apps can be shown in a drawn phone or in an AI photo of a hand holding one.
Can I use our brand?
Yes: your logo, your brand colour throughout, and your own font files if you have them.
What about private information on screen?
The skill can replace a test user’s name and blur anything private in screenshots and recordings. Check every screen before it goes in a video.
Who owns the result?
The scripts are MIT-licensed, the bundled font is open source, the local voices are free for commercial use and the music is generated for you. Content made with OpenRouter and HeyGen follows those services’ terms.
Turn a document into an interactive course.
CourseConverter turns a Word document into an interactive course with knowledge checks, flip cards and more, and exports SCORM for your LMS or HTML to host anywhere. Write it yourself in CourseTag, or let Claude draft it with the free CourseTag skill.