Manim Video Lab

Production pipeline for mathematical and technical animations using Manim Community Edition.

This skill is Docker-first so users do not need to install Manim, LaTeX, or ffmpeg directly on their machine. Setup builds a small local image, manim-video-runtime:v0.20.1, from the pinned official manimcommunity/manim:v0.20.1 base plus ffmpeg, then writes outputs back into the project directory.

Attribution

This project is built with gratitude for Manim, the open-source mathematical animation engine originally written by Grant Sanderson and now maintained by the Manim Community. It is an independent helper skill for agent workflows and is not affiliated with or endorsed by Grant Sanderson, 3Blue1Brown, or the Manim Community.

Maintainer

Created and maintained by Rakesh Sahni.

For bugs and feature requests, please open a GitHub issue. For quick questions, suggestions, or feedback, you can also reach me on X.

License

MIT. See LICENSE.

What It Does

Creates 3Blue1Brown-style animated videos from text prompts. The agent handles creative planning, Python scene generation, rendering, scene stitching, and iterative refinement.

For ambitious output, the skill uses a director-packet workflow before coding: extract the core insight, prompt objects, object/symbol ledgers, beat graph, layout zones, focus handoffs, camera purpose, pacing, payoff frame, source anchors, and preflight gates. This is the main quality layer that keeps outputs from becoming generic slide-like Manim scenes.

Key quality references:

references/director-packet.md - portable Text2Animation-style directing contract.
references/preflight-gates.md - contract, prompt-preservation, LMIR, operator, source-anchor, direction-execution, assertion, visual-audit, and review gates.
references/attention-and-retention.md - quantified retention grammar mined from the 3Blue1Brown source corpus: change cadence, attention thread, camera fusion, overlap dial, hook architecture, and a deterministic dead-air audit.
references/source-backed-patterns.md - distilled Manim pattern atlas: stable scaffolds, live drivers, inspectors, memory rails, dual-space stages, state-space zooms, and template selection.
references/algorithm-choreography.md - logical motion for algorithm and data-structure videos: sync law, state-to-style contracts, swap/compare/traversal idioms, synced pseudocode panels, and step-trace-driven choreography.
references/codegen-craft.md - current-Manim implementation guardrails for token-safe morphs, rate functions, live dependencies, stable inspectors, protected lanes, and source anchors.
references/visual-polish.md - the premium finish layer: glow, sheen, spotlight dimmers, micro-motion, organic entrances, and a strict polish budget.
references/worked-example.md - a complete render-verified exemplar scene with its mini director packet, showing the full doctrine stack operating in ~100 lines.
references/review-loop.md - direct video review or Docker/ffmpeg frame extraction fallback with bounded repair passes, plus an ffmpeg freezedetect dead-air scan.

Use cases:

Concept explainers: "Explain how neural networks learn"
Equation derivations: "Animate the proof of the Pythagorean theorem"
Algorithm visualizations: "Show how quicksort works step by step"
Data stories: "Animate our before/after performance metrics"
Architecture diagrams: "Show our microservice architecture building up"

Install The Skill

This is an Agent Skills package, not a Codex-only package. The portable core is SKILL.md plus bundled scripts/ and references/. The agents/openai.yaml file is optional OpenAI/Codex UI metadata; other tools can ignore it.

Keep the folder intact:

manim-video/
  Dockerfile
  SKILL.md
  scripts/      # Bash and PowerShell Docker wrappers
  references/
  agents/

Install the folder as manim-video so it matches name: manim-video in SKILL.md. If you clone a repository with another folder name, rename it when copying into a skills directory.

Recommended cross-client install path:

mkdir -p .agents/skills
cp -R /path/to/manim-video .agents/skills/manim-video

User-level cross-client install path:

mkdir -p ~/.agents/skills
cp -R /path/to/manim-video ~/.agents/skills/manim-video

Tool-specific paths:

# Claude Code, project-level
mkdir -p .claude/skills
cp -R /path/to/manim-video .claude/skills/manim-video

# Claude Code, personal
mkdir -p ~/.claude/skills
cp -R /path/to/manim-video ~/.claude/skills/manim-video

# Codex, personal
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
cp -R /path/to/manim-video "${CODEX_HOME:-$HOME/.codex}/skills/manim-video"

For Antigravity, OpenCode, VS Code/Copilot, Goose, Amp, and other Agent Skills-compatible tools, prefer .agents/skills/manim-video/ when the tool scans it. If a tool uses its own skills directory, copy the same folder there. No content changes are needed.

Quick layout after project-level shared install:

your-project/
  .agents/
    skills/
      manim-video/
        SKILL.md
        scripts/
        references/

Setup

Install Docker Desktop or Docker Engine, then run:

macOS, Linux, WSL, or Git Bash:

cd /path/to/manim-video
bash scripts/setup.sh

Windows PowerShell:

cd C:\path\to\manim-video
.\scripts\setup.ps1

The setup script checks Docker, builds manim-video-runtime:v0.20.1 if needed, and verifies Manim, ffmpeg, and LaTeX inside the container.

Render

From a Manim project directory:

macOS, Linux, WSL, or Git Bash:

/path/to/manim-video/scripts/manim.sh -ql script.py SceneName

Windows PowerShell:

& "C:\path\to\manim-video\scripts\manim.ps1" -ql script.py SceneName

Production render:

/path/to/manim-video/scripts/manim.sh -qh script.py SceneName

Stitch scenes:

mkdir -p outputs
/path/to/manim-video/scripts/ffmpeg.sh -y -f concat -safe 0 -i concat.txt -c copy outputs/final.mp4

Windows PowerShell:

New-Item -ItemType Directory -Force outputs
& "C:\path\to\manim-video\scripts\ffmpeg.ps1" -y -f concat -safe 0 -i concat.txt -c copy outputs/final.mp4

Useful overrides:

MANIM_DOCKER_IMAGE=your-custom-image /path/to/manim-video/scripts/manim.sh -ql script.py SceneName
MANIM_BASE_IMAGE=manimcommunity/manim:stable bash /path/to/manim-video/scripts/setup.sh --rebuild
MANIM_PROJECT_DIR=/path/to/project /path/to/manim-video/scripts/manim.sh -ql script.py SceneName

Windows PowerShell overrides:

$env:MANIM_DOCKER_IMAGE = "your-custom-image"
& "C:\path\to\manim-video\scripts\manim.ps1" -ql script.py SceneName

$env:MANIM_BASE_IMAGE = "manimcommunity/manim:stable"
& "C:\path\to\manim-video\scripts\setup.ps1" -Rebuild

$env:MANIM_PROJECT_DIR = "C:\path\to\project"
& "C:\path\to\manim-video\scripts\manim.ps1" -ql script.py SceneName

Smoke Test

macOS, Linux, WSL, or Git Bash:

mkdir -p /tmp/manim-smoke
cd /tmp/manim-smoke
cat > script.py <<'PY'
from manim import *

class HelloManim(Scene):
    def construct(self):
        self.camera.background_color = "#1C1C1C"
        title = Text("Manim works", font_size=48, color=BLUE)
        self.play(Write(title))
        self.wait(1)
PY

/path/to/manim-video/scripts/manim.sh -ql script.py HelloManim

Windows PowerShell:

$SmokeDir = Join-Path $env:TEMP "manim-smoke"
mkdir $SmokeDir -Force
cd $SmokeDir
@'
from manim import *

class HelloManim(Scene):
    def construct(self):
        self.camera.background_color = "#1C1C1C"
        title = Text("Manim works", font_size=48, color=BLUE)
        self.play(Write(title))
        self.wait(1)
'@ | Set-Content script.py

& "C:\path\to\manim-video\scripts\manim.ps1" -ql script.py HelloManim

The output appears under media/videos/script/480p15/HelloManim.mp4.

Output Hygiene

The Docker wrappers remove the container after each run, but files written inside /manim are written to the local project directory and persist. Manim writes draft scene renders and caches under media/; ffmpeg writes wherever the output path points.

Recommended convention:

Keep final/shareable files under outputs/, such as outputs/final.mp4.
For single-scene renders, copy the accepted Manim render from media/videos/... into outputs/ before handoff.
Treat media/ as generated render output/cache, not the canonical final location; it can be deleted and regenerated.
Treat review-frames/, concat.txt, and audio_concat.txt as temporary review/stitching artifacts. Clear review-frames/ before new frame extraction and delete it after final review.
Avoid copying final videos into hidden tool-specific cache folders unless the host UI requires an embedded artifact. If a tool creates that secondary copy, keep outputs/ as the primary project-local location.
Keep script.py, plan.md, source assets, narration audio, and approved files under outputs/.

Notes

The default runtime image is manim-video-runtime:v0.20.1, built from manimcommunity/manim:v0.20.1 plus ffmpeg for reproducible open-source support.
Set MANIM_BASE_IMAGE=manimcommunity/manim:stable and run scripts/setup.sh --rebuild or scripts\setup.ps1 -Rebuild to rebuild the local runtime from the latest Manim release.
Set MANIM_DOCKER_IMAGE=your-custom-image to skip the local image and use a fully custom runtime.
Supported host path is intentionally small: Docker Desktop/Engine plus Bash or PowerShell. Python, Manim, LaTeX, and ffmpeg do not need host installs.
Docker preview/file-browser flags such as -p and -f are not useful inside the container.
The official Manim base includes TeX Live but not a standalone ffmpeg binary, so this skill's Dockerfile installs ffmpeg. For extra LaTeX packages or Python plugins, extend the Dockerfile or set MANIM_DOCKER_IMAGE.

Official Docker docs: https://docs.manim.community/en/stable/installation/docker.html

Manim Video Lab

README

Manim Video Lab

Attribution

Maintainer

License

What It Does

Install The Skill

Setup

Render

Smoke Test

Output Hygiene

Notes

Comments (0)