SGA: Plug&Play Geometric Verification for Educational Video Synthesis

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Lopez Jhon, Hinojosa Carlos, Ghanem Bernard
• Venue: arXiv: cs.AI
• Published: 2026-07-20
• Collected: 2026-07-21
• URL: http://arxiv.org/abs/2607.18116v1
• PDF: https://arxiv.org/pdf/2607.18116v1
• DOI: UNKNOWN

Plain-Language Brief

The authors propose the Symbolic Geometric Agent (SGA), a tool for improving the spatial correctness and visual legibility of educational animations generated by Large Language Models (LLMs) using libraries like Manim. SGA is designed to intercept LLM-generated code, extract symbolic scene graphs, and apply refinements when spatial conflicts are detected. The authors also introduce the Manim Visual Quality Score (MVQS), a proxy for spatial integrity. Experiments show that SGA improves MVQS in most cases compared to the raw baseline.

What the Document Says

• The paper discusses SGA, a module for animation pipelines that uses LLMs to generate educational animations with Manim. It focuses on addressing spatial correctness and visual legibility issues. The solution involves intercepting the code, performing partial execution, extracting symbolic scene graphs, and applying refinements when spatial conflicts occur.
• The authors introduce MVQS, a deterministic rendering-free proxy for assessing spatial integrity.
• Experiments were conducted on the MMMC-Code benchmark with four LLM backbones and two agentic pipelines. Results show that SGA achieved a peak MVQS of 73.11 (with Code2Video + GPT-5.1), representing a 16.1% relative improvement over the baseline, and improved MVQS in 7 out of 8 backbone x pipeline configurations.

Key Concepts Mentioned

• Large Language Models (LLMs)
• Manim
• Symbolic Geometric Agent (SGA)
• Manim Visual Quality Score (MVQS)
• spatial correctness
• visual legibility

Claims Made by the Document

• SGA improves spatial correctness and visual legibility of educational animations.
• SGA achieves a peak MVQS of 73.11 (with Code2Video + GPT-5.1), representing a 16.1% relative improvement over the raw baseline.

Evidence Used

• Experiments on the MMMC-Code benchmark showing improvements in MVQS and comparison to the raw baseline.

Methods / Approach

• Intercepting LLM-generated code, performing partial execution, extracting symbolic scene graphs, applying refinements when spatial conflicts occur.
• Introducing the Manim Visual Quality Score (MVQS)

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Recent work leverages Large Language Models (LLMs) to generate executable code for pedagogical animations using libraries such as Manim. However, ensuring spatial correctness and visual legibility remains challenging, as existing frameworks emphasize pedagogical content while overlooking geometric occlusions. We propose the Symbolic Geometric Agent (SGA), a plug-and-play module for code-centric animation pipelines that intercepts LLM-generated code, performs partial execution to extract symbolic scene graphs, and applies targeted refinement when spatial conflicts are detected. We further introduce the Manim Visual Quality Score (MVQS), a deterministic rendering-free proxy for spatial integrity. Experiments on the MMMC-Code benchmark across four LLM backbones and two agentic pipelines show that SGA achieves a peak MVQS of 73.11 (Code2Video + GPT-5.1), corresponding to a 16.1% relative improvement over the raw baseline, and improves MVQS in 7 of 8 backbone x pipeline configurations.