The Year in AI Papers
The year/Google DeepMind

Paper 2601.23265

PaperBanana: Automating Academic Illustration for AI Scientists

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

Published
Jan 2026
Research lab
Google DeepMind
Citations
19
GitHub
6.9K stars

01 In brief

Summary

PaperBanana is an agentic framework that automates the generation of publication-ready academic illustrations, such as methodology diagrams and statistical plots.

It uses five specialized agents—Retriever, Planner, Stylist, Visualizer, and Critic—powered by VLMs and image generation models.

The Retriever selects relevant reference examples, the Planner creates a detailed description, the Stylist applies an auto-summarized aesthetic guideline, and the Visualizer-Critic loop iteratively refines the output over three rounds.

For statistical plots, the Visualizer generates Python Matplotlib code instead of images.

The authors introduce PaperBananaBench, a benchmark of 292 methodology diagrams from NeurIPS 2025, and evaluate using a VLM-as-a-Judge with reference-based scoring on faithfulness, conciseness, readability, and aesthetics.

PaperBanana outperforms baselines (e.g., vanilla Nano-Banana-Pro) by +2.8% faithfulness, +37.2% conciseness, +12.9% readability, +6.6% aesthetics, and +17.0% overall.

It also extends to statistical plots, slightly surpassing human performance in some dimensions.

Limitations include raster output, style standardization, and fine-grained faithfulness errors, with future work on editable vector graphics and improved evaluation metrics.

02 From the paper

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.