ManimGX: a Manim-compatible 3D animation engine on Rust and wgpu

Anyone who has made math animations with Manim has waited: change one line, then watch the CPU grind through frames. ManimGX (a Python API on a Rust engine, MIT, about 1.5k stars as of 2026-10-16), shared over WhatsApp on 16 October, replaces the renderer with Rust and wgpu while keeping the Manim CE API — the same scene.py, and manimgx render outputs a 1080p60 MP4.

Core features

• Manim CE API compatibility: you still write Python ThreeDScene, self.play() and .animate chains; what changes is the renderer underneath. The project positions itself as “a 3D animation engine for agents”.
• An official benchmark with stated conditions: five scenes totalling 36.2 seconds of video, on an Apple M4 Pro at 1080p60, finish in 3.08 seconds from launch to MP4 — against ManimCE’s 278.83 s (90.6×), ManimGL’s 35.28 s (11.5×), Blender Workbench’s 57.61 s (18.7×) and Blender EEVEE’s 601.40 s. Every engine used its fastest encoding preset; the README includes same-scene quality comparisons, and the scenes, settings and runs live in benchmarks/.
• 2D and 3D on one canvas: surfaces, meshes and moving cameras alongside text, equations and plots; LaTeX and Typst typesetting without a local LaTeX install.
• It runs in the browser: npm install manimgx, with Pyodide running Python in a Web Worker and WebGPU rendering the scene — a scene can be embedded in a page with no render server.
• A feedback channel for agents: manimgx inspect produces storyboards and 2D layout checks at scene ends or timestamps you choose.
• Speech in sync: self.say() aligns animation with spoken words.

Typical use cases

• Creators making math and science explainers: iteration speed improves by an order of magnitude, and changing one frame no longer costs a coffee.
• Developers having agents produce teaching animations: the README ships the prompt — paste “Follow https://manimgx.academa.ai/llms.txt. Install ManimGX, make a 3D animation of the solar system, and open it” and the agent handles setup.
• Teams embedding animations in the web: the browser rendering path removes a render server.

Quick start

pip install manimgx      # Python 3.13+ on macOS, Linux or Windows

Save the README’s scene.py, then:

manimgx render scene.py

You get Hello3D.mp4. Fonts, typesetting and video encoding all ship with the package, so there are no extra dependencies.

Summary

For people making math and science animations with Manim who are tired of waiting, and for teams that want agent-produced animations embeddable in the web. It is not a drop-in guarantee for legacy projects relying on the full Manim community scene ecosystem — the README claims API compatibility, but edge behaviors of community Scene subclasses are yours to verify. Python API with a Rust/wgpu core, MIT, 1,536 stars as of 2026-10-16, last pushed 2026-10-15 — development is active.

Read the benchmark with its conditions: the 90.6× figure is five specific scenes on an M4 Pro with each engine’s fastest encoding preset, not a universal speedup; on different hardware or more complex scenes the ratio will move. If what you want is a full video pipeline rather than a programmable animation engine, MoneyPrinterTurbo and ComfyUI are the other two routes.