AI - Manim Video Generator

Automatic video generation using an agentic LLM flow in combination with the manim python library.

Overview

The project experiments with automated Manim video creation. An agent workflow delegates code drafting to a Code Writer and validation to a Code Reviewer, using LiteLLM for model routing so different models from different providers can be compared on the same task. The flow focuses on reducing render failures and improving visual consistency through iterative feedback and optional vision inputs.

Current Flow (subject to change)

Installation

1. Clone the repository:

git clone https://github.com/makefinks/manim-generator.git
cd manim-generator

2. Install the requirements

With uv (recommended):

uv sync

Or using pip directly:

pip install -e .

3. Additional dependencies

Install ffmpeg and, if you plan to render LaTeX, a LaTeX distribution.

Windows (using Chocolatey):

choco install ffmpeg
choco install miktex

macOS (using Homebrew):

brew install ffmpeg
brew install --cask mactex

Linux (Debian/Ubuntu):

sudo apt-get update
sudo apt-get install texlive texlive-latex-extra texlive-fonts-extra texlive-science

4. Configure environment variables

Create a .env file from the provided template:

cp .env.example .env

Then edit .env and add your API keys. Providers available via openrouter are supported through LiteLLM.

Usage

1. Execute the script

With uv (recommended):

uv run manim-generate

Or if you've activated the virtual environment:

source .venv/bin/activate
manim-generate

Or using Python directly:

python -m manim_generator.main

2. CLI Arguments

The script supports the following command-line arguments:

Video Data Input

Argument	Description	Default
--video-data	Description of the video to generate (text string)	-
--video-data-file	Path to file containing video description	"video_data.txt"

Model Configuration

Argument	Description	Default
--manim-model	Model to use for generating Manim code	"openrouter/anthropic/claude-sonnet-4"
--review-model	Model to use for reviewing code	"openrouter/anthropic/claude-sonnet-4"
--streaming	Enable streaming responses from the model	False
--temperature	Temperature for the LLM Model	0.4
--force-vision	Adds images to the review process, regardless if LiteLLM reports vision is not supported	-
--provider	Specific provider to use for OpenRouter requests (e.g., 'anthropic', 'openai')	-

Process Configuration

Argument	Description	Default
--review-cycles	Number of review cycles to perform	5
--manim-logs	Show Manim execution logs	False
--output-dir	Directory for generated artifacts (overrides auto-naming)	Auto (e.g., manim_animation_20250101_120000)
--success-threshold	Percentage of scenes that must render successfully to trigger enhanced visual review mode	100
--frame-extraction-mode	Frame extraction mode: highest_density (single best frame) or fixed_count (multiple frames)	"fixed_count"
--frame-count	Number of frames to extract when using fixed_count mode	3
--headless	Suppress most output and show only a single progress bar	False

Reasoning Tokens Configuration

Argument	Description	Default
--reasoning-effort	Reasoning effort level for OpenAI-style models (choices: "minimal", "low", "medium", "high")	-
--reasoning-max-tokens	Maximum tokens for reasoning (Anthropic-style)	-
--reasoning-exclude	Exclude reasoning tokens from response (model still uses reasoning internally)	-

Note: You cannot use both --reasoning-effort and --reasoning-max-tokens at the same time.

Providing --output-dir skips the automatic descriptor-based folder name and uses the supplied path instead.

Example

uv run manim-generate --video-data "Explain the concept of neural networks with visual examples" --manim-model "openrouter/anthropic/claude-sonnet-4" --review-model "openrouter/anthropic/claude-sonnet-4" --review-cycles 3

Or with the command directly (if virtual environment is activated):

manim-generate --video-data "Explain the concept of neural networks with visual examples" --manim-model "openrouter/anthropic/claude-sonnet-4" --review-model "openrouter/anthropic/claude-sonnet-4" --review-cycles 3

3. Configure image support

Images are available only when the reviewer model supports multimodal input.

• https://openrouter.ai/models?modality=text+image-%3Etext
• https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json

Contributing

Focus areas include prompt improvements, review loop refinements, code quality, and new features or optimizations.

Known issues

• Streaming: current streaming implementation does not provide syntax highlighting
• Prompting / environment setup: the selected LLM version may not match the local installation.