PaperBanana is an open-source, AI-powered platform that converts text descriptions into publication-ready academic illustrations, including methodology diagrams, statistical charts, and infographics, using a five-agent collaborative architecture.
What is PaperBanana?
PaperBanana is an AI-powered tool that automates the creation of publication-ready academic illustrations from natural language text descriptions. It takes text input and produces images such as methodology diagrams, statistical plots, system architectures, flow charts, and poster assets. The platform is web-based, with a credit-based freemium model, and is developed by researchers at Peking University (as indicated by the GitHub repository owner dwzhu-pku). It is open source and available on GitHub, with a related research paper.
Key Features
- Multi-agent architecture: Employs a closed-loop five-agent system (Planner, Visualizer, etc.) to ensure precise, logically accurate diagrams.
- Code-based statistical plots: Generates executable Python Matplotlib code for charts, eliminating numerical hallucination and ensuring mathematical precision.
- Nano-Banana-Pro model: A built-in model optimized for rendering scientific icons, shapes, and connectors with high aesthetic quality.
- Sketch refinement: Accepts rough hand-drawn sketches or whiteboard notes and transforms them into publication-quality figures.
- Multiple diagram types: Supports methodology diagrams, statistical charts, system architectures, flow charts, educational infographics, and poster assets.
- Style and aspect ratio control: Offers dropdown selection for visual style (e.g., Methodology Diagrams, Statistical Plots) and aspect ratio (Auto, square, etc.).
- Credit-based generation: Each generation costs 5 credits; users must sign in to generate images (freemium model with limited free credits likely).
Who is it for?
- Academic researchers: Prepare methodology diagrams and statistical charts for papers submitted to venues like NeurIPS, ICML, and ICLR.
- Graduate students: Create thesis figures and presentation visuals without manual drawing or complex software.
- Educators: Develop clear, scientifically accurate infographics for lectures and teaching materials.
- Scientific communicators: Produce polished illustrations for blog posts, posters, and supplementary materials.
What can you do with PaperBanana?
- Generate methodology diagrams: Describe a neural network architecture or algorithm pipeline, and PaperBanana produces a structured diagram with accurate shapes and connectors.
- Create publication-quality statistical charts: Input raw data and get precise bar charts or line plots as Python code, not pixel-based approximations.
- Polish existing sketches: Upload a rough hand-drawn diagram and have PaperBanana refine its layout, color scheme, and typography while preserving scientific content.
How does PaperBanana work?
PaperBanana uses a closed-loop five-agent architecture. A Planner agent interprets the text description and creates a structured visual layout. A Visualizer agent then renders the diagram using the Nano-Banana-Pro model, producing shapes, connectors, and icons. Additional agents handle aesthetic refinement and validation, ensuring the output meets top-venue standards. The entire process is automated, with no manual editing required.
Pricing
PaperBanana operates on a freemium model. Users can sign in to generate images, with each generation costing 5 credits. Free credits are likely available upon registration, and additional credits can be purchased (specific pricing not listed on the page).
FAQ
Is PaperBanana open source?
Yes, PaperBanana is open source. The code is available on GitHub.
How does PaperBanana ensure data accuracy in statistical plots?
Unlike general image generators that render charts as pixels, PaperBanana generates executable Python Matplotlib code from raw data. This code-based approach ensures that bars, axes, and scales are mathematically precise, eliminating numerical hallucination.
What inputs does PaperBanana need to generate a diagram?
PaperBanana requires a text description of the desired figure, up to 15,000 characters. Users can also select a visual style (e.g., Methodology Diagrams, Statistical Plots) and configure options like aspect ratio, quality, and format.
Can PaperBanana improve my existing hand-drawn sketches?
Yes, PaperBanana can transform rough hand-drawn sketches or whiteboard notes into professional-grade illustrations. It applies auto-summarized aesthetic guidelines to refine layout, color palette, typography, and iconography while preserving the original structure.
