Skip to content
Back to search

video-skills

SkillWorks with Claude CodeGitHub

Two Claude skills that turn an idea into a narrated vertical video with Remotion or Revideo. ElevenLabs voice, word-timed captions, lip-synced mascot. MIT.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/anujkapoor-arch--video-skills/badge.svg)](/listings/anujkapoor-arch--video-skills)

HTML

<a href="/listings/anujkapoor-arch--video-skills"><img src="/api/listings/anujkapoor-arch--video-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

git clone https://github.com/anujkapoor-arch/video-skills
cd video-skills
./install.sh

This copies both skills, each with its own copy of the kit, into `~/.claude/skills/`. Then, in any session:

/remotion-video a 30 second short on three ways to remember people's names, for my channel "Small Talk"

Claude writes the script, shows you the lines and the character count before it spends anything on voice, renders, and checks frames before handing you the file.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 145 lines
  1. # video-skills
  2. Two Claude skills that turn an idea into a narrated vertical video. You describe the video, Claude writes the script, ElevenLabs speaks it, and Remotion or Revideo renders a 1080x1920 MP4 for Shorts, Reels or TikTok.
  3. ## What you get
  4. This is the example that ships with the kit, rendered from [kit/common/script.json](kit/common/script.json) with no edits. The preview is silent and shrunk; the MP4s are the real 1080x1920 output with the voice.
  5. <img src="docs/example.gif" width="300" alt="The 25 second example video: hook, three points, a number counting up to 66, a bar chart, and a call to action with confetti">
  6. - Remotion output: [docs/example-remotion.mp4](docs/example-remotion.mp4) (25 s, 2.6 MB)
  7. - Revideo output: [docs/example-revideo.mp4](docs/example-revideo.mp4) (25 s, 2.7 MB)
  8. Both files were re-compressed with ffmpeg to keep the repo small. Straight out of the renderer they are 4.9 MB and 11 MB. Run the steps below with your own ElevenLabs key and you get this same video; change `script.json` and you get yours.
  9. ![Four frames from the example video](docs/example.png)
  10. ## What's in the starter
  11. Both skills ship the same starter video, built two ways:
  12. - Five scene types: a hook card, a numbered list, a big number that counts up, a bar chart, and a call to action.
  13. - Every visual lands on the word that's being spoken. ElevenLabs returns a timestamp for each character, and the kit turns that into a time for every word.
  14. - Word-by-word captions, with the current word highlighted.
  15. - A mascot that lip-syncs from the word timings, blinks, and cheers at the end.
  16. - Colours, channel name, footer, captions and mascot are all set in one JSON file.
  17. | Skill | Engine | Pick it when |
  18. |---|---|---|
  19. | `remotion-video` | [Remotion](https://www.remotion.dev) 4 (React, HTML and CSS) | You want normal web layout, a preview studio with hot reload, and good docs |
  20. | `revideo-video` | [Revideo](https://github.com/midrender/revideo) 0.11 (canvas, generators) | You want speed and an MIT licence, and can live with a pre-1.0 library |
  21. ## How the two compare
  22. Measured on the example video (25 seconds, 752 frames) on an Apple M2 with 8 GB RAM.
  23. | | Remotion | Revideo |
  24. |---|---|---|
  25. | Full render | 26.5 s | 8 to 10 s (6.2 s with two workers) |
  26. | Output size | 4.9 MB | 11 MB |
  27. | Check one moment | `remotion still`, about 4 s | Render a 4 s range, about 2 s |
  28. | Licence | Free for individuals and companies of up to 3 people. Larger companies pay. | MIT |
  29. | Disk | About 950 MB in `node_modules` after the first render | About 310 MB, plus 575 MB of Chrome in `~/.cache/puppeteer` |
  30. | Telemetry | Not checked | On by default. The kit turns it off for renders. |
  31. | Layout | HTML and CSS, text wraps by itself | Canvas only. The kit measures and wraps text for you. |
  32. My pick: start with Remotion if its licence fits you, because fixing a layout in HTML is quicker. Move to Revideo when you render a lot of videos or need the MIT licence.
  33. ## What you need
  34. - Node (tested on 24), Python 3, and ffmpeg on your PATH.
  35. - An [ElevenLabs](https://elevenlabs.io) API key. The example video is 401 characters. There is no free or silent mode: the timings come from the voice.
  36. ## Install (Claude Code)
  37. ```bash
  38. git clone https://github.com/anujkapoor-arch/video-skills
  39. cd video-skills
  40. ./install.sh
  41. ```
  42. This copies both skills, each with its own copy of the kit, into `~/.claude/skills/`. Then, in any session:
  43. ```
  44. /remotion-video a 30 second short on three ways to remember people's names, for my channel "Small Talk"
  45. ```
  46. Claude writes the script, shows you the lines and the character count before it spends anything on voice, renders, and checks frames before handing you the file.
  47. ## Use it without Claude
  48. The kit is a normal project generator.
  49. ```bash
  50. ./kit/new-video.sh remotion ~/videos/my-first-short # or: revideo
  51. cd ~/videos/my-first-short
  52. npm install
  53. export ELEVENLABS_API_KEY=... # or put it in a .env file here
  54. # edit script.json
  55. python3 narrate.py --check # validate, count characters, no API call
  56. python3 narrate.py # voice + timings
  57. npm run render # -> out/video.mp4
  58. ```
  59. `script.json` is the only file you have to write. Every field is described in [kit/common/SCRIPT.md](kit/common/SCRIPT.md). A scene looks like this:
  60. ```json
  61. {
  62. "kind": "points",
  63. "say": "Shrink the habit until it takes two minutes. Do it at the same time every day. Then stop while it still feels easy.",
  64. "title": "How it works",
  65. "items": [
  66. {"text": "Shrink it to two minutes", "at": "shrink"},
  67. {"text": "Same time every day", "at": "same"},
  68. {"text": "Stop while it's easy", "at": "stop"}
  69. ]
  70. }
  71. ```
  72. `at` is the spoken word each row appears on.
  73. ## What's in the repo
  74. ```
  75. skills/remotion-video/SKILL.md instructions Claude follows for Remotion
  76. skills/revideo-video/SKILL.md instructions Claude follows for Revideo
  77. kit/new-video.sh creates a project from the kit
  78. kit/common/narrate.py script.json -> narration.wav + spec.json (ElevenLabs, cached)
  79. kit/common/SCRIPT.md every script.json field
  80. kit/common/script.json the example script
  81. kit/common/fonts/ Mukta, three weights
  82. kit/remotion/ the Remotion starter
  83. kit/revideo/ the Revideo starter
  84. ```
  85. ## What we learned the hard way
  86. The skills carry these so Claude doesn't repeat them. The short list:
  87. Remotion
  88. - Everything is a function of the frame. Anything with memory (state, CSS transitions, `Math.random()`) breaks, because frames render in parallel and out of order.
  89. - Hold the render with `delayRender` until fonts load.
  90. - An SVG character has to be ported to JSX by hand before you can move its parts.
  91. Revideo
  92. - A `<Txt>` whose text is a function renders nothing. Set the text by hand.
  93. - The `<SVG>` node flattens groups and drops ids. Draw the character from shapes.
  94. - Browser errors print as `JSHandle@object`. `VITE_DEBUG_LOGS=1` turns them into text.
  95. - There are no docs in the npm packages. We read the type files.
  96. Both
  97. - Take every time from the narration, never from a number typed into the code. Change a line of the script and the video re-times itself.
  98. - Check single frames before a full render. Most problems are visible in one frame.
  99. ## Limits
  100. - 1080x1920 only. Another size means editing the layout code.
  101. - The font is Mukta (Latin and Devanagari). Other scripts need another font.
  102. - Nobody has tested the Revideo editor (`npm start`) with this kit.
  103. - Revideo prints one "asynchronous property" warning per render. The output is correct.
  104. - Claude can check frames. It can't hear the voice or watch motion, so watch the video once yourself.
  105. ## Licence
  106. MIT for everything in this repo. The Mukta font is under the SIL Open Font License ([kit/common/fonts/OFL.txt](kit/common/fonts/OFL.txt)). Remotion has its own licence: read it at [remotion.dev/docs/license](https://www.remotion.dev/docs/license) before using it at a company with more than 3 people.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.