Manim's API, a Rust GPU engine, and a 90× speedup claim
ManimGX reimplements the Manim CE Python API on a Rust + wgpu renderer so agents can produce math and 3D videos in seconds instead of minutes.

What it does It’s a drop-in-ish animation engine: you write scenes in Python using Manim CE’s API, and a Rust core built on wgpu renders the MP4. It handles 2D and 3D together — surfaces, meshes, moving cameras, LaTeX/Typst text, plots — and ships fonts, typesetting, and video encoding so nothing else needs installing. It also runs in the browser via Pyodide and WebGPU, with an npm package and no render server.
The interesting bit
The pitch is aimed at coding agents, not just humans: there’s an llms.txt for agent setup, and a manimgx inspect command that gives storyboards and layout checks so an agent can verify its animation without watching it. Fully statically typed keyword arguments and .animate chains mean agents get real autocomplete and error feedback instead of guessing at a loose API.
Key highlights
- Benchmarked at 90.6× faster than ManimCE, 11.5× faster than ManimGL, and 18.7× faster than Blender Workbench (36.2 s of 1080p60 video rendered in 3.08 s on an Apple M4 Pro; methodology published in the repo).
- 3D lighting adapted from Google’s Filament: physically based materials, IBL, ambient occlusion, bloom, AgX tone mapping.
- LaTeX input works without a LaTeX install —
mitextranslates it to Typst, which does the typesetting. self.say()syncs animations to spoken words for narrated videos.- Linux wheels bundle Mesa’s lavapipe for CPU Vulkan rendering when no GPU is available.
Caveats
- Requires Python 3.13+, which is newer than many environments run.
- “Compatible with Manim CE’s API” is a target, not a guarantee — the docs include a dedicated “where behavior differs” page for migrating scenes, so expect some divergence.
- The headline benchmark is one machine (Apple M4 Pro) and five scenes; your mileage will vary.
Verdict If you (or your agent) generate math explainers or 3D visualizations and Manim’s render times hurt, this is worth a serious look. If you need pixel-identical Manim CE output or run older Python, wait for the compatibility story to mature.
Frequently asked
- What is academa-labs/manimgx?
- ManimGX reimplements the Manim CE Python API on a Rust + wgpu renderer so agents can produce math and 3D videos in seconds instead of minutes.
- Is manimgx open source?
- Yes — academa-labs/manimgx is open source, released under the MIT license.
- What language is manimgx written in?
- academa-labs/manimgx is primarily written in Python.
- How popular is manimgx?
- academa-labs/manimgx has 1.1k stars on GitHub.
- Where can I find manimgx?
- academa-labs/manimgx is on GitHub at https://github.com/academa-labs/manimgx.