unknown

PaperBanana is an agentic framework that automates the generation of publication-ready academic illustrations for AI research papers. The system uses five specialized agents (Retriever, Planner, Stylist, Visualizer, and Critic) powered by vision-language models to create methodology diagrams and statistical plots. The framework is evaluated using PaperBananaBench, a benchmark with 292 test cases from NeurIPS 2025 papers, and demonstrates superior performance in faithfulness, conciseness, readability, and aesthetics compared to baselines. The system can also enhance existing human-drawn diagrams and generate statistical visualizations, though it faces challenges with connection errors in complex diagrams.

PaperBanana: Automating Academic Illustration for AI Scientists

Squad for devs building apps with GenAI and LLMs! Connect, share, exchange ideas, and get feedback. 

Build With GenAI

PaperBanana is an agentic framework by Google that generates publication-ready academic illustrations from methodology descriptions.

<p>PaperBanana is an agentic framework by Google that generates publication-ready academic illustrations from methodology descriptions.</p>

<p>This is a clever use of agents—turning paper figures from a time sink into something assistive. Curious how well it handles complex diagrams without introducing subtle visual errors.</p>
