🎬 chalktalk

An open-source AI-agent Skill for creating 3Blue1Brown-style animated math, calculus, physics, and STEM explainer videos — straight from a prompt.

Give your AI agent the ability to turn "make me a video explaining why the derivative of x² is 2x" into an actual rendered, narrated-quality animation — the kind you'd see on 3Blue1Brown.

⚡ Paste this to your AI agent

Copy this into Claude Code (or any coding agent). It installs the skill if needed, then builds your video — just swap in your own topic on the last line:

Use the "chalktalk" skill to make a 3Blue1Brown-style animated math explainer video.

Skill repo: https://github.com/ahkamboh/chalktalk

If it isn't installed yet, clone it into your skills folder first:
git clone https://github.com/ahkamboh/chalktalk.git ~/.claude/skills/chalktalk

Then read its SKILL.md and follow it end-to-end to create a video explaining:
why the derivative of x² is 2x        ← replace with any topic you want

(Above: a frame sequence from the bundled example — the growing-square proof that d/dx[x²] = 2x.)

---

What this is

This repo is a Skill: a small, self-contained package of instructions + scripts + references that teaches an AI coding agent (Claude Code, and any agent that supports the SKILL.md format) how to do one specialized task really well — here, producing animated math explainer videos end to end:

• 🧰 Bootstraps the environment — a Python virtualenv, the animation engine, and the system libraries it needs (cairo / pango / ffmpeg), across macOS & Linux (scripts/setup_env.sh).
• ✍️ Teaches good scene design — 3b1b narrative principles plus a tested animation cheatsheet.
• 🚫 Works without LaTeX — most machines don't have a multi-GB TeX install, so it renders math with Pango Text + Unicode and documents every hidden-LaTeX trap (DecimalNumber, axis labels) that crashes naïve renders.
• 🔍 Verifies before delivering — extracts frames so the agent looks at the layout and catches off-screen / overlapping bugs before the final render.
• 📚 Ships ready-made recipes — derivatives, Riemann sums & integrals, the Fundamental Theorem, limits/ε–δ, the chain rule, Taylor series, vectors.

It is MIT-licensed and free for any AI agent or human to use.

Demo

The bundled example, examples/derivative_x2.py, answers "why is d/dx[x²] = 2x?" two ways — as the slope of the parabola (a sliding tangent line with a live slope read-out) and as a geometric proof (a growing square). Render it:

bash scripts/setup_env.sh demo          # one-time: venv + engine + system libs
cd demo && source .venv/bin/activate
cp ../examples/derivative_x2.py .
bash ../scripts/render.sh derivative_x2.py DerivativeOfXSquared h   # 1080p60
# -> media/videos/derivative_x2/1080p60/DerivativeOfXSquared.mp4

Use it with an AI agent

Claude Code

Install the skill where Claude Code discovers skills, then just ask:

# personal (all projects):
git clone https://github.com/ahkamboh/chalktalk.git \
  ~/.claude/skills/chalktalk

# or per-project:
git clone https://github.com/ahkamboh/chalktalk.git \
  .claude/skills/chalktalk

Then in a session:

"Make a video explaining Riemann sums." "Animate why the chain rule works." "Turn the Fundamental Theorem of Calculus into a 60-second explainer."

The skill triggers automatically — the agent reads SKILL.md, sets up the environment if needed, writes the scene from the closest recipe, draft-renders, checks frames, and delivers the final mp4.

Any other agent

The skill is just Markdown + shell + Python. Point your agent at SKILL.md as the entry point; it links out to the references and scripts as needed (progressive disclosure).

What's inside

chalktalk/
├── SKILL.md                      # entry point: the end-to-end workflow
├── scripts/
│   ├── setup_env.sh              # bootstrap venv + engine + cairo/pango/ffmpeg
│   ├── render.sh                 # render with the right library paths set
│   └── verify_frames.sh          # extract frames for visual QA
├── references/
│   ├── animation-cheatsheet.md   # core animation API patterns
│   ├── no-latex-mode.md          # render math without LaTeX (and the gotchas)
│   └── topic-recipes.md          # blueprints: derivatives, integrals, FTC, …
├── examples/
│   └── derivative_x2.py          # full working no-LaTeX scene
└── assets/
    └── preview.gif

Requirements

• Python 3.11 or 3.12 (some native dependencies may lack wheels on the very newest Python).
• ffmpeg, plus cairo, pango, pkg-config — installed automatically by scripts/setup_env.sh (Homebrew on macOS, apt on Debian/Ubuntu).
• LaTeX is NOT required. Install it only if you want publication-grade typesetting — see references/no-latex-mode.md.

Why no-LaTeX by default?

A full TeX distribution is several gigabytes and a frequent source of latex not found render failures. For the vast majority of explainers, Pango Text with Unicode math (x², ∫, Δ, →, …) looks great and just works everywhere. The skill documents the exact spots where the renderer secretly calls LaTeX (DecimalNumber, Integer, axis labels) and how to route around them — and how to opt back into real LaTeX if you want it.

Credits

• Inspired by 3Blue1Brown (Grant Sanderson), whose videos defined this visual style of mathematical storytelling.
• This repo is an independent, community skill — not affiliated with or endorsed by 3Blue1Brown.

License

MIT © 2026 Ali Hamza Kamboh. Use it freely, for agents and humans alike. PRs adding new topic recipes are very welcome.