Code2Video - Show Lab open source AI teaching video generation framework

What is Code2Video?

Code2Video is an innovative open source project by the Show Lab team at the National University of Singapore that automatically converts code snippets into high-quality video content (mp4 format). The project through a unique code-centered paradigm , using carbon-now-cli tools to generate code into beautiful pictures , the use of ffmpeg to stitch these picture sequences into a complete teaching video . The core functionality includes four main components: code segmentation, image generation, resizing and video compositing, which can simulate the effect of entering code line by line, and is especially suitable for programming teaching and technology demonstration content.

Features of Code2Video

• Code-driven generation paradigmThe Manim code is a unified medium that uses executable Manim code to precisely control the temporal sequence and spatial layout of the video through structured commands, ensuring the logical rigor and reproducibility of the generated content.
• Modular Multi-Intelligent Body Collaboration Framework: Using Planner (planning storyboards), Coder (generating debuggable code) and Critic (optimizing layouts through anchors) three agents work together to achieve an end-to-end automated process from requirements resolution to code generation.
• High quality vector animation outputThe Manim engine generates resolution-independent vector animations that support complex mathematical formulas, dynamic transformations and smooth transitions, ensuring clear, professional-grade video content.
• Multi-dimensional assessment and optimization mechanisms: A comprehensive evaluation system covering knowledge accuracy, visual aesthetics (e.g., layout reasonableness, animation smoothness) and generation efficiency (Token consumption, rendering time) to support iterative optimization.
• Standardized benchmarks and scalabilityMMMC benchmarks (117 educational topics) are provided to support rapid single-concept generation and batch task processing, adapting to the needs of scenarios of different complexity.
• Cross-scene compatibility: Suitable for multiple fields such as math visualization, science demonstration, and programming instruction, it balances educational rigor with creative flexibility.

Code2Video's core strengths

• Code-driven precision controlManim Code is a unified medium that uses executable Manim code to precisely control the temporal sequence and spatial layout of the video through structured commands, ensuring that the generated content is logically rigorous and highly reproducible.
• Modular Multi-Intelligence for Efficient Collaboration: Through Planner (planning storyboards), Coder (generating debuggable code) and Critic (optimizing layouts through anchors) three agents work together to achieve end-to-end automation from requirements resolution to code generation, dramatically improving development efficiency.
• Industry-leading generation qualityBased on the Manim engine to generate resolution-independent vector animation, the output video is close to the official 3Blue1Brown course in terms of clarity, aesthetics, and teaching effect, providing a professional-grade visual experience.
• Comprehensive evaluation and optimization systemThe system has a multi-dimensional evaluation system covering knowledge accuracy (TeachQuiz), visual aesthetics (AES) and generation efficiency (Token consumption, rendering time), which supports iterative optimization and ensures the quality of the final output.
• Strong standardization and scalabilityMMMC: Provides the first code-driven video benchmark MMMC (including 117 educational topics), supporting single-point rapid generation and batch task processing, adapting to the needs of different complexity scenarios, and facilitating secondary development and customization by the community.
• Supported by rich ecological resourcesThe video is based on IconFinder, Icons8 and other high-quality icon libraries, based on the Manim Community and the mainstream open source ecosystem of large models, which significantly improves the visual richness and development flexibility of the video.

What is Code2Video's official website?

• Project website:: https://showlab.github.io/Code2Video/
• Github repository:: https://github.com/showlab/Code2Video
• arXiv Technical Paper:: https://arxiv.org/pdf/2510.01174

Who Code2Video is for

• Educators and content creators: Suitable for teachers, online education practitioners and science content creators who need to quickly generate high-quality teaching videos, which can transform abstract knowledge into intuitive animated presentations, enhancing teaching efficiency and attractiveness.
• Technology developers and researchers: Provides researchers working in the fields of educational technology, multimodal generation, or code-driven content synthesis with reproducible benchmarks (e.g., MMMC) and modular frameworks to support algorithmic iteration and customized development.
• Manim Community Users and Animation Enthusiasts: Aimed at users who are familiar with or want to learn Manim programming, it reduces the threshold for manually creating complex instructional animations by precisely controlling the details of the animation through code.
• Automated video production on demandIt is suitable for organizations that need to generate standardized educational videos in batch (e.g., online course platforms, training institutions), and automate the end-to-end production process through multi-intelligence collaboration.