From manim-reel
Turn text content into a polished 9:16 Manim explainer reel (MP4) with a Deepgram voiceover and audio-synced captions. Use when the user wants a Reel, Short, TikTok, or vertical explainer video made from text, an article, a topic, or a script.
How this skill is triggered — by the user, by Claude, or both
Slash command
/manim-reel:manim-reelThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Turn any text into a vertical 1080x1920 explainer video: Manim visuals, a
Turn any text into a vertical 1080x1920 explainer video: Manim visuals, a Deepgram Aura voiceover, and captions synced to the narration.
You do the creative work — turning text into a storyboard JSON. The
reelgen engine deterministically renders that JSON into an MP4. You never
write Manim code; you only write JSON.
Get the source. Use the text / article / script the user gave. If they gave only a topic, first write a tight ~100-140 word explainer script.
Plan 5-8 scenes. One idea per scene. Scene 1 is always title; the last
is always outro. Pick a layout per scene (see the table below).
Write narration for each scene — conversational, 2-4 short sentences. Each sentence becomes one on-screen caption line. ~100-140 words total gives a ~40-50s reel (ideal for Reels/Shorts).
Write the storyboard JSON following reference/schema.md exactly. Save
it to storyboards/<slug>.json.
Validate, then build:
python -m reelgen validate storyboards/<slug>.json
python -m reelgen build storyboards/<slug>.json --out output/<slug>.mp4 --preview
--preview renders at half resolution, fast. When it looks right, run again
without --preview for the final 1080x1920 render.
Report the output path, duration, and that audio is present.
| layout | use for | visual fields |
|---|---|---|
title | opening hook (scene 1) | title, subtitle |
concept | one big idea / keyword | keyword (1-3 words), support |
diagram | how parts relate | nodes[], edges[] |
steps | a process or list | heading, items[] |
comparison | A vs B | left, right (each heading + points[]) |
outro | recap + call to action | title, cta |
steps / diagram / comparison, the
k-th caption line reveals the k-th element. Write roughly one narration
sentence per steps item, per diagram node, and per comparison point.keyword for concept scenes: 1-3 words — it renders very large.diagram: 3-6 nodes. Give each a row (higher = nearer the top) and col
(0 = center, negative = left). Every edge must reference real node ids.items / points short — 2-5 words each.The reelgen engine must be installed once (from the project or plugin root):
pip install -e .
.env as DEEPGRAM_API_KEY=... for the voiceover.
Without a key the reel still renders, with silent, correctly-timed narration.See reference/schema.md for the full JSON schema and a complete example.
See examples/harness-ai-agents.json for a working 7-scene storyboard.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub yash-kavaiya/manim-reels-youtube --plugin manim-reel