Manim Cursor - AI Animation Generator

A FastAPI-based backend and React frontend to convert natural language prompts into 3Blue1Brown-style animations using the Manim library. Built with OpenAI's GPT model via OpenRouter API.

Live Demo: https://renderconcepts.com

---

Features

• Generate math visualizations from plain English.
• Automatic tool_call generation using LLM.
• Supports 20+ animation tools (text, LaTeX, graphs, shapes, transforms, etc.)
• Exposes Manim videos and generated code via static routes.
• React + Tailwind frontend for easy UX.

---

Tech Stack

Frontend

• Next.js (App Router)
• TailwindCSS
• Clerk (for auth + usage limits)
• Deployed on Vercel

Backend

• FastAPI
• Manim Community Edition (subprocess-based rendering)
• Supabase (optional for storing usage)
• Deployed on AWS EC2 (Ubuntu 22.04)

---

How It Works

1. Prompt Input: A user provides a natural language prompt.
2. Classification: Prompt is checked for complexity.
3. LLM Interaction:
   • If simple, a direct Manim Python script is generated.
   • If complex, LLM generates a list of tool_calls in JSON.
4. Code Generation: Each tool_call is converted into valid Manim Python code.
5. Rendering: Code is saved and rendered using manim.
6. Result: A video (.mp4) and manim code file (.py) are returned via API.

---

Rate Limiting

• Guests: 3 video credits.
• Logged-in users: 10 total credits.
• Data stored in Supabase.

---

Project Structure

backend/
├── app/
│   ├── main.py              # FastAPI app, LLM calls, route handling
│   ├── executor.py          # Validation, scene builder, Manim subprocess call
│   ├── dispatcher.py        # tool -> code generator mapping
│   ├── code_generators.py   # Python code templates per tool
│   ├── registry.py          # Tool schema + validation
│   ├── schemas.py           # Pydantic models for all tool args
│   ├── prompt.py            # LLM system prompts
│   └── render/
│       ├── code/            # .py files generated
│       └── output/videos/   # .mp4 output files
frontend/
├── app/
│   ├── page.tsx             # Main UI with prompt input
│   └── components/ # Input form logic

---

Local Dev Setup

Backend Setup (FastAPI + OpenRouter + Manim)

cd backend
python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows
pip install -r requirements.txt

Create a .env file:

OPENAI_API_KEY=your-openrouter-api-key

Start the server:

uvicorn app.main:app

Make sure you have Manim CE installed and accessible in your PATH.

Frontend Setup (Next.js + Tailwind)

cd frontend
npm install
npm run dev

---

Supported Tools

Includes but not limited to:

• add_text, add_latex
• plot_graph, set_axes, label_axes, add_axis_markings
• add_shape, move_object, fade_in, fade_out
• wait, transform_object, rotate_object, scale_object
• group_objects, draw_arrow, set_color
• create_number_line, animate_text_change, surround_with_shape
• set_background_color

Full argument validation handled via Pydantic models in schemas.py.

---

Example Prompts

"Show 'Hello World' in red"

"Plot cosine graph with yellow axes and custom ticks"

"Create a rectangle and transform it into a circle after 2 seconds"

---

Credits

Idea inspired by Harkirat Singh's YouTube video.

---

Future Ideas

• Prompt history with previews
• Advanced tool combinations (3D, camera zoom, etc)
• Shareable links
• Export code directly

---

Troubleshoot

You might have to use these commands, if you face error while running pip install -r requirements.txt during installation of pangocairo.

sudo apt install -y python3-dev pkg-config

sudo apt install -y build-essential libcairo2-dev

sudo apt install -y libpango1.0-dev libgirepository1.0-dev

---

License

MIT License. See LICENSE file.