InfoAgent: Traceable Generation and Repair of Evidence-Grounded Infographics

2026-09-30

AI Summary

This study addresses the challenge of maintaining consistency among facts, symbols, and visual relationships during infographic rendering and revision, as well as the absence of evidence-dependent tracing mechanisms for element repair. To this end, it proposes InfoAgent, a training-free framework that introduces evidence-bound visual-symbolic program synthesis. By leveraging typed dependency graphs (IVD descriptors) to record facts and their evidential sources, combined with retrieval priors, the framework enables hierarchical execution and dependency verification for both raster and symbolic objects. This approach facilitates traceable, precise local repairs while avoiding costly global regeneration. Evaluated on IGenBench, InfoAgent achieves a Q-ACC of 93.0%, modifying only 12.4% of the canvas during repair—substantially outperforming the 67.3% alteration required by global regeneration—and improves the complete checklist pass rate to 28.5%.

Abstract

Reliable infographic generation requires facts, symbols, and visual relations to remain consistent through rendering and revision. Correcting one element also requires tracking its supporting evidence and the dependencies affected by the change. We present \textbf{InfoAgent}, a training-free framework for \emph{evidence-bound visual-symbolic program synthesis}. Its Infographic Visual Description (IVD) records factual payloads, evidence provenance, execution routes, and verification obligations in a typed dependency graph. Retrieved design priors guide compilation, and layered execution combines raster synthesis with editable symbolic and binding objects while retaining their traces. Dependency-aware repair localizes corrections, rechecks affected dependencies, and requires protected obligations to remain satisfied under the declared checkers. Unresolved obligations remain explicit. On IGenBench, InfoAgent achieves 93.0 Q-ACC and 59.0 I-ACC. We also introduce InfoGraphicBench-Evidence, where complete-checklist pass rates on 200 test requests increase from 21.5\% for Same-IVD Prompt to 23.5\% for the initial layered output and 28.5\% after repair, using the same evidence and initial IVD. On 120 audited repair cases, localized repair edits 12.4\% of the canvas on average, compared with 67.3\% for global regeneration.

Problem

• infographic generation
• evidence grounding
• dependency tracking
• traceable repair
• visual-symbolic consistency

Innovation

• evidence-bound program synthesis
• typed dependency graph
• layered execution
• dependency-aware repair
• training-free framework