Fully Automated Mathematical Video Generation System
MathVizAI is a complete end-to-end system that takes mathematical problems and automatically generates polished educational videos with synchronized visualizations and audio narration. The entire pipeline is automated from problem input to final video output.
- Intelligent Problem Solving: Uses LLM to solve complex mathematical problems with detailed proofs
- Automated Validation: Self-correcting system with evaluator that ensures solution accuracy
- Natural Audio Scripts: Generates conversational explanations optimized for text-to-speech
- Voice Cloning TTS: neuTTS-air integration for natural voice narration
- Automated Video Rendering: Renders all Manim scenes with quality settings
- Audio-Video Synchronization: Aligns audio with video and adjusts durations automatically
- Text Slide Generation: Creates text slides when video is missing
- Beautiful Visualizations: Creates Manim animations that sync with narration
- Segmented Output: Breaks content into 15-20 second segments for optimal processing
- Multiple Scene Support: Handles complex problems with unlimited animation scenes
- Final Video Assembly: Concatenates all segments into one polished video
Query Input: "Solve x² + 5x + 6 = 0"
↓
┌─────────────────────┐
│ Math Solver │ ← Solves problem with proof
└─────────────────────┘
↓
┌─────────────────────┐
│ Solution │ ← Validates correctness
│ Evaluator │ (Auto-retry until perfect)
└─────────────────────┘
↓
┌─────────────────────┐
│ Script Writer │ ← Creates audio narration
└─────────────────────┘ (15-20s segments)
↓
┌─────────────────────┐
│ Video Generator │ ← Generates Manim code
└─────────────────────┘ (multiple scenes)
↓
┌─────────────────────┐
│ TTS Generator │ ← Creates audio files
└─────────────────────┘ (voice cloning)
↓
┌─────────────────────┐
│ Video Renderer │ ← Renders all scenes
└─────────────────────┘ (automated)
↓
┌─────────────────────┐
│ Video │ ← Syncs audio + video
│ Synchronizer │ (duration matching)
└─────────────────────┘ (text slides for missing video)
↓
Final Polished Video
(ready to publish)
- Python 3.8 or higher
- GitHub Token with API access
- FFmpeg (required for Manim video rendering)
- Manim (optional, for automated video rendering)
- neuTTS-air (optional, for audio generation)
- Clone the repository
git clone https://github.com/anirudhsengar/MathVizAI.git
cd MathVizAI- Install dependencies
pip install -r requirements.txt- Set up environment variables
Create a .env file or set environment variables:
# Windows PowerShell
$env:GITHUB_TOKEN = "your_github_token_here"
# Linux/Mac
export GITHUB_TOKEN="your_github_token_here"Alternatively, create a .env file:
GITHUB_TOKEN=your_github_token_here
- Optional: Install Manim for automated video rendering
pip install manimFor detailed Manim installation instructions, visit: https://docs.manim.community/en/stable/installation.html
FFmpeg Installation (required for Manim):
- Windows:
choco install ffmpegor download from https://ffmpeg.org/ - Linux:
sudo apt install ffmpeg - Mac:
brew install ffmpeg
- Optional: Set up neuTTS-air for audio generation
See TTS_SETUP.md for detailed instructions.
Quick setup:
# Install soundfile
pip install soundfile
# Clone and install neuTTS-air
git clone https://github.com/anirudhsengar/neutts-air.git
cd neutts-air
pip install -e .Note: TTS is optional. Without it, the system will generate all other components and skip audio generation.
python main.pyThen enter mathematical problems when prompted:
Enter math problem: Prove that the square root of 2 is irrational
Each query generates a timestamped folder in output/ with:
output/
└── 20250118_143022_Prove_that_the_square_root_of_2/
├── metadata.json # Session metadata
├── original_query.txt # Your original question
├── solver/
│ ├── solution_attempt_1.txt
│ └── solution_final.txt # Approved solution
├── evaluator/
│ ├── evaluation_attempt_1.txt
│ └── evaluation_final.txt # Final validation report
├── script/
│ ├── audio_script.txt # Complete script
│ ├── segments.json # Parsed segments
│ └── segment_XX_visual.txt # Visual cues
├── audio/
│ ├── segment_01.wav # Generated audio (if TTS enabled)
│ ├── segment_02.wav
│ ├── segment_XX_audio.txt # Audio script text
│ └── audio_metadata.json # Audio generation info
├── video/
│ ├── manim_visualization.py # Executable Manim code
│ └── rendering_instructions.txt
└── final/
└── (final video will go here)
After generation, render the Manim visualization:
cd output/[your_session_folder]/video
manim -qh manim_visualization.py MathVisualizationQuality options:
-ql: Low quality (480p15) - fast preview-qm: Medium quality (720p30)-qh: High quality (1080p60) - recommended-qk: Production quality (1440p60)
Edit config.py to customize:
# Temperature settings for each component
TEMPERATURE_SOLVER = 0.4 # Solution generation
TEMPERATURE_EVALUATOR = 0.0 # Strict evaluation
TEMPERATURE_SCRIPT_WRITER = 0.6 # Creative narration
TEMPERATURE_VIDEO_GENERATOR = 0.2 # Consistent code
# Retry settings
MAX_SOLVER_RETRIES = 5 # Max attempts before accepting solution
# Video settings
VIDEO_RESOLUTION = "1080p"
VIDEO_FPS = 60- Mathematical problem solving with proofs
- Automated solution validation with retry logic
- Audio script generation with segments
- Manim visualization code generation
- TTS audio generation (neuTTS-air integration)
- Audio-video synchronization
- Final video assembly
- Background music and sound effects
- Subtitle generation
- Batch processing mode
- Web interface
MathVizAI/
├── main.py # Entry point
├── config.py # Configuration
├── requirements.txt # Dependencies
├── system_prompts/ # LLM prompts
│ ├── solver.txt
│ ├── evaluator.txt
│ ├── script_writer.txt
│ └── video_generator.txt
├── pipeline/ # Core pipeline modules
│ ├── orchestrator.py # Main pipeline controller
│ ├── solver.py # Math solver
│ ├── evaluator.py # Solution validator
│ ├── script_writer.py # Audio script generator
│ └── video_generator.py # Manim code generator
├── utils/ # Utility modules
│ ├── llm_client.py # LLM API wrapper
│ ├── prompt_loader.py # Prompt management
│ └── file_manager.py # File operations
└── output/ # Generated content
- Custom Prompts: Edit files in
system_prompts/to customize AI behavior - Pipeline Stages: Add new stages in
pipeline/directory - Configuration: Add settings to
config.py
Example problems you can try:
- "Prove that the square root of 2 is irrational"
- "Solve the integral of e^x * sin(x) dx"
- "Prove the Pythagorean theorem using similar triangles"
- "Find the derivative of x^x"
- "Explain the Fundamental Theorem of Calculus"
Contributions are welcome! Areas where help is needed:
- TTS integration (neuTTS-air)
- Audio-video synchronization
- Manim animation templates
- Additional visualization types
- Testing and bug fixes
MIT License
- Manim Community - Mathematical animation engine
- Microsoft Phi-4 - Reasoning model via GitHub Models
- Azure AI Inference - LLM API infrastructure