To make math animations with AI, you describe what you want in plain English and let the model generate and render it. Here's how QuantumSketch does it — and why it works.

Why Manim?

Manim (Mathematical Animation Engine) renders animations by executing Python that describes geometry, transforms, and timing. Every frame is deterministic and pixel-perfect — unlike generative video, which hallucinates physics.

The problem: Manim has a steep API surface. Writing a clean derivative animation from scratch takes 2–4 hours for someone new.

The QuantumSketch pipeline

Prompt → LLM storyboard → Manim code → render → TTS narration → merge

1. Prompt: "Explain the chain rule visually using f(g(x)) = sin(x²)"
2. LLM phase: breaks the concept into 4–6 narrative beats, writes Manim Scene classes per beat
3. Render phase: ManimGL renders each scene as a video chunk
4. TTS phase: narration is synthesized and aligned to animation timing
5. Merge: ffmpeg concatenates + mixes audio

What you can animate

| Topic | Example prompt | |---|---| | Calculus | "Show the limit definition of a derivative" | | Linear algebra | "Visualize matrix multiplication as a transformation" | | Statistics | "Animate the central limit theorem with n=5, 30, 100" | | Data structures | "Show merge sort step by step" | | Physics | "Animate a pendulum's phase space portrait" |

Try it

→ quantumsketch.app — paste a concept, get a rendered video.

The manim-coding-skill that powers this is also open-source on GitHub.

---

Written by Shihab Shahriar Antor · Shahriar Labs · Also see: LetX