Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AutoFigure: Generating and Refining Publication-Ready Scientific Illustrations

Minjun Zhu, Zhen Lin, Yixuan Weng, Panzhong Lu, Qiujie Xie, Yifan Wei, Sifan Liu, Qiyao Sun, Yue Zhang

TL;DR

This work tackles the bottleneck of producing publication-ready scientific illustrations from long-context documents by introducing FigureBench, a large-scale benchmark of 3,300 long-text–illustration pairs, and AutoFigure, a two-stage, agentic framework based on the Reasoned Rendering paradigm. Stage I performs conceptual grounding and layout planning to produce a symbolic, structurally coherent blueprint, while Stage II renders this blueprint into high-fidelity visuals with an erase-and-correct post-processing pipeline to ensure textual accuracy. Through extensive automated and human evaluations, AutoFigure consistently surpasses baselines in visual design, communication, and content fidelity, achieving publication-ready quality across diverse document types. The work demonstrates a practical path toward automated, high-quality scientific visualization, with implications for broader AI-assisted scientific communication and future extensions to dynamic and interactive figures.

Abstract

High-quality scientific illustrations are crucial for effectively communicating complex scientific and technical concepts, yet their manual creation remains a well-recognized bottleneck in both academia and industry. We present FigureBench, the first large-scale benchmark for generating scientific illustrations from long-form scientific texts. It contains 3,300 high-quality scientific text-figure pairs, covering diverse text-to-illustration tasks from scientific papers, surveys, blogs, and textbooks. Moreover, we propose AutoFigure, the first agentic framework that automatically generates high-quality scientific illustrations based on long-form scientific text. Specifically, before rendering the final result, AutoFigure engages in extensive thinking, recombination, and validation to produce a layout that is both structurally sound and aesthetically refined, outputting a scientific illustration that achieves both structural completeness and aesthetic appeal. Leveraging the high-quality data from FigureBench, we conduct extensive experiments to test the performance of AutoFigure against various baseline methods. The results demonstrate that AutoFigure consistently surpasses all baseline methods, producing publication-ready scientific illustrations. The code, dataset and huggingface space are released in https://github.com/ResearAI/AutoFigure.

AutoFigure: Generating and Refining Publication-Ready Scientific Illustrations

TL;DR

This work tackles the bottleneck of producing publication-ready scientific illustrations from long-context documents by introducing FigureBench, a large-scale benchmark of 3,300 long-text–illustration pairs, and AutoFigure, a two-stage, agentic framework based on the Reasoned Rendering paradigm. Stage I performs conceptual grounding and layout planning to produce a symbolic, structurally coherent blueprint, while Stage II renders this blueprint into high-fidelity visuals with an erase-and-correct post-processing pipeline to ensure textual accuracy. Through extensive automated and human evaluations, AutoFigure consistently surpasses baselines in visual design, communication, and content fidelity, achieving publication-ready quality across diverse document types. The work demonstrates a practical path toward automated, high-quality scientific visualization, with implications for broader AI-assisted scientific communication and future extensions to dynamic and interactive figures.

Abstract

High-quality scientific illustrations are crucial for effectively communicating complex scientific and technical concepts, yet their manual creation remains a well-recognized bottleneck in both academia and industry. We present FigureBench, the first large-scale benchmark for generating scientific illustrations from long-form scientific texts. It contains 3,300 high-quality scientific text-figure pairs, covering diverse text-to-illustration tasks from scientific papers, surveys, blogs, and textbooks. Moreover, we propose AutoFigure, the first agentic framework that automatically generates high-quality scientific illustrations based on long-form scientific text. Specifically, before rendering the final result, AutoFigure engages in extensive thinking, recombination, and validation to produce a layout that is both structurally sound and aesthetically refined, outputting a scientific illustration that achieves both structural completeness and aesthetic appeal. Leveraging the high-quality data from FigureBench, we conduct extensive experiments to test the performance of AutoFigure against various baseline methods. The results demonstrate that AutoFigure consistently surpasses all baseline methods, producing publication-ready scientific illustrations. The code, dataset and huggingface space are released in https://github.com/ResearAI/AutoFigure.
Paper Structure (33 sections, 1 equation, 21 figures, 12 tables)

This paper contains 33 sections, 1 equation, 21 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. FigureBench: A Benchmark for Automated Scientific Illustration Generation
  4. AutoFigure
  5. Stage I: Conceptual Grounding and Layout Generation
  6. Concept Extraction and Symbolic Construction.
  7. Critique-and-Refine.
  8. Stage II: Aesthetic Synthesis and Text Post-processing
  9. Experiments
  10. Automated Evaluations
  11. Human Evaluation with Domain Experts
  12. Ablation Studies
  13. Conclusion
  14. Data Curation Details for FigureBench
  15. Human Evaluation Protocol
  16. ...and 18 more sections

Figures (21)

  • Figure 1: The composition of the FigureBench dataset. It features a rich collection of text-figure pairs from four distinct sources (Paper, Survey, Blog, and TextBook), demonstrating the benchmark's capability to evaluate automatic illustration generation across various domains and complexities.
  • Figure 2: An Overview of the AutoFigure, which decouples structural layout generation from aesthetic rendering. Stage 1 ensures structural fidelity by having a multi-agent system generate and iteratively self-correct a symbolic layout (SVG). Stage 2 renders the validated layout and employs an erase-and-correct module—using OCR and cross-verification—to guarantee perfect textual accuracy with high-fidelity vector overlays. This figure is also produced by AutoFigure and serves as a qualitative showcase of its generation quality.
  • Figure 3: Examples showcasing the versatility of AutoFigure in generating complex scientific illustrations from a diverse range of academic texts. Note that we employ a unified default style (Delicate and cute cartoon comic style, using Morandi color palette) solely to ensure visual consistency and readability for comparative analysis. This is a choice of presentation rather than a limitation of the method; users can freely specify or mix arbitrary styles as needed (see in Appendix \ref{['app:style']}). We present a diverse range of results in Appendix \ref{['appendix:cases']} to further illustrate our approach.
  • Figure 4: Human evaluation results from 10 first-author experts assessing AI-generated figures for 21 of their own publications. The comprehensive study required experts to perform three tasks: (a) a forced-choice holistic ranking of six AI models against the original reference to determine a win rate, (b) a publication intent selection, and (c-e) multi-dimensional scoring on a 1-5 Likert scale for accuracy, clarity, and aesthetics.
  • Figure 5: Ablation studies of the AutoFigure framework. Subplots compare different backbone models on (a) pre-rendering symbolic layouts versus (b) final rendered outputs. Also shown are (c) performance scaling with increased test-time refinement iterations and (d) the impact of different intermediate sketch formats.
  • ...and 16 more figures