An agentic framework for AI researchers. Generate high-quality methodology diagrams and plots from text or references with PaperBanana.
🍌 Generator Configuration
Unused credits will be refunded if the task completes early
Empowering Every Creative Workflow
From initial concepts to final production assets, PaperBanana fits seamlessly into your process.
Diagram Generation
from Scratch
PaperBanana accepts textual descriptions and orchestrates agents to plan layout and render components, bridging the gap between abstract ideas and visual diagrams.
- Model architecture diagrams
- Complex flowcharts
- PaperBanana System overviews
Polishing
Hand-Drawn Sketches
Upload a rough sketch. With PaperBanana, PaperBanana's multimodal capabilities interpret your visual intent and refine it into a polished, professional diagram.
- Sketch-to-Digital conversion
- Style consistency
- Layout preservation
Statistical Plots
Visualization
Beyond diagrams, PaperBanana effectively generates high-quality statistical plots, ensuring data is presented with clarity and academic rigor.
- Publication-style charts
- Accurate data representation
- Vector-quality output
Designed for Scientists
PaperBanana addresses the pain point of creating figures. Focus on your research, not on drawing tools.
PaperBanana Benchmark
Examples from PaperBananaBench (292 test cases from NeurIPS 2025)
Why PaperBanana?
Solve the Time Bottleneck
Generating publication-ready illustrations is a labor-intensive bottleneck. PaperBanana lifts this burden, allowing you to focus on core research rather than manual drafting.
Standardized Aesthetics
Consistently outperforms leading baselines in faithfulness, conciseness, and aesthetics. Our PaperBanana framework ensures your figures meet the rigorous standards of top-tier AI conferences.
Open Source Collaboration
We introduce PaperBananaBench, a comprehensive benchmark of 292 curated test cases from NeurIPS 2025, provided to the community to foster innovation in automated scientific illustration with PaperBanana.
Multimodal Capability
Powered by state-of-the-art VLMs and image generation models. PaperBanana seamlessly handles both text-to-image generation and iterative refinement via self-critique.
Frequently Asked Questions
Everything you need to know about PaperBanana
PaperBanana is an automated agentic framework designed for AI researchers. It automates the process of creating high-quality, publication-ready academic illustrations, including methodology diagrams and statistical plots, directly from textual descriptions or rough references.
Our framework employs a multi-step process: 'Retrieve' to gather context, 'Plan' to design the layout, 'Render' to generate the initial image using state-of-the-art models, and 'Refine' to iteratively improve the output through self-critique for maximum accuracy and aesthetics.
PaperBanana is versatile. It excels at creating complex methodology diagrams (like model architectures and flowcharts) as well as precise statistical plots. It can virtually handle any visual requirement for an academic paper.
Yes! Beyond generating from scratch, PaperBanana features a powerful polishing capability. You can input your rough hand-drawn sketches or draft diagrams, and the system will refine them into professional, vector-style illustrations.
Absolutely. PaperBanana is benchmarked against standards from top AI conferences like NeurIPS. Our evaluation metrics focus on faithfulness, conciseness, readability, and aesthetics to ensure the output meets publication standards.
Yes, PaperBanana is an open-source project. We aim to support the research community. Please check our GitHub repository for the latest code and updates.
Not at all. The goal of PaperBanana is to bridge the gap between your research ideas and visual communication. You provide the scientific context, and our agents handle the design principles.
Ready to automate your figures?
Let PaperBanana handle the drawing, so you can handle the research.
UNOFFICIAL SERVICE DISCLAIMERThis website is an independent platform built to provide online accessibility for the PaperBanana open-source framework. We are strictly a third-party service provider and are NOT affiliated, associated, authorized, endorsed by, or in any way officially connected with Google, Peking University, or any of their subsidiaries. The official PaperBanana project belongs to its original authors.