PaperBanana is an advanced academic illustration generator specifically designed for researchers, transforming raw paper text, references, and even rough sketches into publication-ready methodology diagrams, statistical plots, and clean AI research figures. It aims to eliminate the tedious manual process of redrawing figures, offering a streamlined workflow for creating high-quality visuals for papers, posters, and presentations.
Key Features and Tools:
- Image to Image Generation: Users can upload sketches, draft diagrams, screenshots, or reference images and describe the desired research figure for generation. This tool is ideal for quickly converting initial concepts into structured visuals.
- AI Image Editor: This feature allows for precise editing and cleaning of existing visuals using text instructions. Researchers can revise figures, clean compositions, adjust details, or transform drafts into stronger, more polished versions.
- Image Upscaler: For low-quality drafts or screenshots, the Image Upscaler enhances clarity and output resolution, ensuring figures are crisp and ready for publication.
- Context-Driven Generation: Unlike generic AI art tools, PaperBanana prioritizes scientific faithfulness by starting from source context. Users paste method sections, system overviews, or notes, which guides the AI in planning a cleaner, more accurate academic illustration.
- Agentic Layout Planning: The generator plans the figure structure—mapping stages, blocks, and data flow—before rendering visuals. This ensures logical consistency and adherence to scientific principles, crucial for methodology diagrams.
- Iterative Refinement: The platform supports an iterative review process. If an initial draft is crowded or vague, the system can critique and revise the figure based on feedback, improving wording, spacing, and composition without requiring a complete restart.
- Configurable Generation: Users have control over various parameters, including figure caption (to define the type of figure needed), maximum iterations (for speed or deeper review), and aspect ratio (to match paper, poster, or slide layouts).
- Cost Transparency: PaperBanana provides credit estimates upfront, and unused credits are refunded if a figure converges early, offering transparent cost management.
- Benchmark-Backed Quality: The platform is benchmarked for research use through "PaperBananaBench," a test set of 292 curated cases linked to NeurIPS 2025-style research figures, ensuring product quality is aligned with real academic needs.
Use Cases:
- Methodology Diagram Generation: Generate complex model architecture diagrams, algorithm pipelines, and system overviews directly from paper text or notes, saving significant time and effort.
- Academic Illustration Cleanup: Transform rough hand-drawn sketches, whiteboard notes, or early slide screenshots into polished, conference-ready academic illustrations with consistent labels and improved visual style.
- AI Research Figure Generation: Create various statistical plots, including bar charts, line charts, comparison charts, ablation studies, and benchmark figures. It also helps in preparing multi-panel layouts for posters and presentations, ensuring clear and structured data visualization.
PaperBanana is built for PhD students, lab researchers, startup research teams, and paper authors who need to accelerate figure production, achieve cleaner layouts, and ensure higher scientific faithfulness in their visuals, all without requiring specialized design skills. It offers a unified workflow for diverse academic illustration needs, from initial drafts to final publication-ready outputs.
