Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PaperBanana: Automating Academic Illustration for AI Scientists

Dawei Zhu, Rui Meng, Yale Song, Xiyu Wei, Sujian Li, Tomas Pfister, Jinsung Yoon

TL;DR

PaperBanana tackles the bottleneck of producing publication-ready academic illustrations by introducing an agentic framework that coordinates Retriever, Planner, Stylist, Visualizer, and Critic to generate methodology diagrams and extend to statistical plots. It leverages reference-driven planning and self-critique, validated by the PaperBananaBench benchmark of NeurIPS 2025 diagrams evaluated with a VLM-based judge across faithfulness, conciseness, readability, and aesthetics. Experimental results show PaperBanana outperforming Vanilla, Few-shot, and Paper2Any baselines across all four dimensions and the overall score, with ablations confirming the importance of each agent and iterative refinement. The work also demonstrates the framework's applicability to statistical plots via code-based rendering, discusses human-alignment and evaluation reliability, and outlines limitations and future directions for editable vector outputs, style diversity, and broader domain extension.

Abstract

Despite rapid advances in autonomous AI scientists powered by language models, generating publication-ready illustrations remains a labor-intensive bottleneck in the research workflow. To lift this burden, we introduce PaperBanana, an agentic framework for automated generation of publication-ready academic illustrations. Powered by state-of-the-art VLMs and image generation models, PaperBanana orchestrates specialized agents to retrieve references, plan content and style, render images, and iteratively refine via self-critique. To rigorously evaluate our framework, we introduce PaperBananaBench, comprising 292 test cases for methodology diagrams curated from NeurIPS 2025 publications, covering diverse research domains and illustration styles. Comprehensive experiments demonstrate that PaperBanana consistently outperforms leading baselines in faithfulness, conciseness, readability, and aesthetics. We further show that our method effectively extends to the generation of high-quality statistical plots. Collectively, PaperBanana paves the way for the automated generation of publication-ready illustrations.

PaperBanana: Automating Academic Illustration for AI Scientists

TL;DR

PaperBanana tackles the bottleneck of producing publication-ready academic illustrations by introducing an agentic framework that coordinates Retriever, Planner, Stylist, Visualizer, and Critic to generate methodology diagrams and extend to statistical plots. It leverages reference-driven planning and self-critique, validated by the PaperBananaBench benchmark of NeurIPS 2025 diagrams evaluated with a VLM-based judge across faithfulness, conciseness, readability, and aesthetics. Experimental results show PaperBanana outperforming Vanilla, Few-shot, and Paper2Any baselines across all four dimensions and the overall score, with ablations confirming the importance of each agent and iterative refinement. The work also demonstrates the framework's applicability to statistical plots via code-based rendering, discusses human-alignment and evaluation reliability, and outlines limitations and future directions for editable vector outputs, style diversity, and broader domain extension.

Abstract

Despite rapid advances in autonomous AI scientists powered by language models, generating publication-ready illustrations remains a labor-intensive bottleneck in the research workflow. To lift this burden, we introduce PaperBanana, an agentic framework for automated generation of publication-ready academic illustrations. Powered by state-of-the-art VLMs and image generation models, PaperBanana orchestrates specialized agents to retrieve references, plan content and style, render images, and iteratively refine via self-critique. To rigorously evaluate our framework, we introduce PaperBananaBench, comprising 292 test cases for methodology diagrams curated from NeurIPS 2025 publications, covering diverse research domains and illustration styles. Comprehensive experiments demonstrate that PaperBanana consistently outperforms leading baselines in faithfulness, conciseness, readability, and aesthetics. We further show that our method effectively extends to the generation of high-quality statistical plots. Collectively, PaperBanana paves the way for the automated generation of publication-ready illustrations.
Paper Structure (41 sections, 7 equations, 12 figures, 3 tables)

This paper contains 41 sections, 7 equations, 12 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Task Formulation
  3. Methodology
  4. Benchmark Construction
  5. Data Curation
  6. Evaluation Protocol
  7. Experiments
  8. Baseline Methods and Models
  9. Evaluation Settings.
  10. Main Results
  11. Ablation Study
  12. PaperBanana for Statistical Plots Generation.
  13. Discussion
  14. Enhancing Aesthetics of Human-Drawn Diagrams
  15. Coding vs Image Generation for Visualizing Statistical Plots
  16. ...and 26 more sections

Figures (12)

  • Figure 1: Examples of methodology diagrams and statistical plots generated by PaperBanana, which show the potential of automating the generation of academic illustrations.
  • Figure 2: [Generated by , textual description to reproduce this diagram is presented in Appendix \ref{['app_sec:reproduce_diagram']}.] Overview of our PaperBanana framework. Given the source context and communicative intent, we first apply a Linear Planning Phase to retrieve relevant reference examples and synthesize a stylistically optimized description. We then use an Iterative Refinement Loop (consisting of Visualizer and Critic Agents) to transform the description into visual output and conduct multi-round refinements to produce the final academic illustration.
  • Figure 3: [Generated by ] Statistics of the test set of PaperBananaBench (totaling 292 samples). The average length of source context / figure caption is 3,020.1 / 70.4 words.
  • Figure 4: [Generated by ] Vanilla Gemini-3-Pro vs. PaperBanana for statistical plots generation.
  • Figure 5: [Generated by ] Coding vs. Image Generation for visualizing statistical plots.
  • ...and 7 more figures