Paper Banana is an innovative AI academic figure generator designed specifically for researchers, PhD students, and scientists. It addresses the significant challenge of creating publication-ready diagrams and statistical plots, which traditionally consumes valuable time using complex software like Illustrator or TikZ. With Paper Banana, users can transform their methodology text, experiment descriptions, or raw data into high-quality academic figures in minutes.
The platform offers a suite of powerful features to streamline the figure generation process:
-
Instant Diagram Generation from Text: Simply paste your methodology section, abstract, or system description. Paper Banana's AI intelligently extracts key entities, relationships, and data flows to render structured, publication-quality diagrams, eliminating manual drawing and alignment. This includes complex methodology and pipeline diagrams for various domains like agent/reasoning, vision, generative models, and scientific workflows.
-
Mathematically Accurate Statistical Plots: Unlike general image generators that might hallucinate data, Paper Banana generates executable Python Matplotlib code directly from your raw data (JSON or CSV). This ensures mathematical accuracy for all statistical plots, including bar charts, line graphs, scatter plots, heatmaps, and confusion matrices, with correct bars, data points, error bars, and scales.
-
Professional Polish for Existing Diagrams: Users can upload rough sketches or draft figures, and Paper Banana's AI will refine the layout, colors, and typography while preserving the original scientific intent. The tool is rigorously benchmarked on PaperBananaBench, a dataset of 292 real diagrams from NeurIPS 2025, ensuring outputs meet the stringent aesthetic and quality standards of top-tier conferences like NeurIPS, ICML, CVPR, and ACL.
-
Conference Style Compliance: The system automatically applies aesthetic guidelines from leading AI conference publications, ensuring correct palettes, typography, line weights, and layouts that seamlessly integrate with proceedings.
-
Multi-Format Input Support: Paper Banana is highly flexible, accepting various input formats including plain text descriptions, full methodology sections, JSON data objects, and CSV tables. This allows researchers to paste their content as-is without needing to pre-structure it.
-
High-Resolution Output: All generated figures are exported as high-resolution PNGs, suitable for journal submissions, conference papers, posters, and theses, with configurable aspect ratios to match specific venue requirements.
What sets Paper Banana apart is its sophisticated 5-Agent AI Pipeline, developed in collaboration with Peking University and Google Cloud AI Research. This pipeline comprises specialized agents—Retriever, Planner, Stylist, Visualizer, and Critic—working in sequence to produce high-quality figures. The Critic agent can perform up to 10 refinement cycles, ensuring meticulous attention to detail. The platform is designed for accessibility, requiring no coding or design skills and presenting no learning curve. Furthermore, it supports iterative refinement, allowing users to describe desired changes in plain English for continuous improvement until satisfaction.
The process is straightforward:
- Paste Your Paper Content: Input your text, data, or even a rough sketch.
- AI Generates and Refines: The 5 AI agents process your input, plan the layout, apply academic styling, and render the figure, typically in under three minutes, with real-time results.
- Download and Submit: Obtain a high-resolution PNG ready for publication. Further refinements can be requested with simple text descriptions.
