showlab/Code2Video

Leverages multi-agent LLM orchestration (Planner, Coder, Critic) to synthesize executable Manim animation code from educational topics, ensuring reproducible and debuggable video generation rather than relying on pixel-based diffusion models. Produces high-fidelity educational videos with precise visual control by executing Python code that directly specifies animations, spatial layouts, and temporal sequencing—enabling greater consistency and correctness compared to text-to-video baselines.

1,585 stars.