Repolode
Submit
marcelo-earth

marcelo-earth/generative-manim

Generative Manim turns text prompts into Manim animation code using LLMs such as GPT-4o or Claude, then renders it as video.

View on GitHubWebsite
Stars
939
Forks
135
Open issues
0
Last commit
2 months ago

What is it?

Generative Manim is a set of tools for making videos with Manim, the Python animation engine used for math and educational content. Users describe an animation in plain language, a large language model writes the corresponding Manim code, and the project renders that code into a video. The goal is to let people without programming or video editing experience create animations from text.

The project supports a wide range of model backends, including OpenAI GPT models, Anthropic Claude models, Google Gemini, Moonshot Kimi, open-weight models hosted through Featherless, and any model accessible via LiteLLM. It ships a web app, a FastAPI-based API, and a legacy Streamlit interface. The repository also includes a training pipeline for fine-tuning open-source code models on Manim with SFT, DPO and GRPO, using the renderer as a pass/fail reward signal, plus an executable benchmark for evaluating Manim code generation.

The project is also the basis of Animo, a desktop application that packages the same workflow with local rendering.

Key features

  • 01Generates Manim code from natural language prompts and renders it to video.
  • 02Supports OpenAI, Anthropic, Google Gemini, Kimi, Featherless and LiteLLM model engines.
  • 03Provides a web demo, an API, and a legacy Streamlit interface.
  • 04Includes an executable benchmark with render-based scoring and pass@k metrics.
  • 05Contains a training pipeline for fine-tuning open-source models on Manim code.

Use cases

  • Creating explainer animations for math or science lessons without learning Manim.
  • Prototyping Manim scenes quickly from text descriptions.
  • Comparing how different LLMs perform on Manim code generation.
  • Researching fine-tuning of code models with a renderer as a verifier.

Things to know

  • Most models call hosted LLM APIs such as OpenAI, Anthropic, or Featherless, which require an API key and usually a paid account.
  • Several open-source models listed in the README are still marked as in progress (🚧) and are not yet usable.

In collections

AI video generation

Topics

#manim#llm#text-to-video#animation#python#openai#code-generation#2d#gpt#gpt-4

FAQ

What is generative-manim?
Generative Manim turns text prompts into Manim animation code using LLMs such as GPT-4o or Claude, then renders it as video. Generative Manim is a set of tools for making videos with Manim, the Python animation engine used for math and educational content. Users describe an animation in plain language, a large language model writes the corresponding Manim code, and the project renders that code into a video. The goal is to let people without programming or video editing experience create animations from text.
Is generative-manim free to use? What license does it have?
generative-manim is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is generative-manim still maintained?
Updated within 6 months. The last commit to generative-manim was on Sep 17, 2026.
What can I use generative-manim for?
Creating explainer animations for math or science lessons without learning Manim. Prototyping Manim scenes quickly from text descriptions. Comparing how different LLMs perform on Manim code generation. Researching fine-tuning of code models with a renderer as a verifier.
Where can I get generative-manim?
The source code and installation instructions are on GitHub: https://github.com/marcelo-earth/generative-manim

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects