PaperBanana - AI-powered academic illustration generator for methodology diagrams and research figures

What Is PaperBanana — Your Dedicated AI Assistant for Academic Figures

Anyone who's ever prepared a conference submission knows the feeling. It's 2 AM, the deadline is in 12 hours, and you're still tweaking arrow alignments in PowerPoint, adjusting font sizes in Illustrator, or wrestling with color schemes in Figma. What should be a straightforward task—turning your research into clear, professional figures—can eat up hours, sometimes days, of precious time.

That's the problem PaperBanana was built to solve. Unlike general-purpose AI image generators like Midjourney or DALL·E, which prioritize visual novelty and aesthetic appeal, PaperBanana is an academic illustration generator designed specifically for research contexts. It focuses on what matters most for scientific communication: structural clarity, label readability, and faithful representation of your methodology.

The key difference? PaperBanana doesn't start from a blank canvas. Instead, it works from your original paper text, reference materials, or hand-drawn sketches—first planning the chart's layout (what goes where, how elements connect) and then rendering the visual output. This "structure-first" approach ensures that the resulting figure is not just pretty, but scientifically sound and publication-ready.

It's already being used by research teams at top institutions worldwide, including Seoul National University (SNU), Stanford, UC Berkeley, CMU, Tsinghua, SJTU, IIT Madras, and many more. And to back its claims with transparency, PaperBanana also released PaperBananaBench, an open-source benchmark with 292 test cases based on NeurIPS 2025 styling—giving the academic community a standardized way to evaluate figure generation quality.

The Features Your Research Workflow Actually Needs

PaperBanana packs six core capabilities into one platform. Here's what each one does for you in practice.

Methodology Diagram Drafting — You can use it to turn dense method sections into clean, structured architecture diagrams. Just paste your paper text, add a figure title, and the AI plans the layout—model architectures, algorithm pipelines, system workflows—before rendering. Each generation costs 29 credits.

Sketch Cleanup and Redraw — You can use it to polish rough whiteboard sketches, tablet scribbles, or slide screenshots into professional academic illustrations. The system preserves your original layout structure while improving label readability, spacing, arrows, and overall visual style. This one's 15 credits per run.

AI Figure Editor — You can use it to make targeted edits to existing figure drafts. Upload your current version, describe what needs to change (tweaked labels, re-routed arrows, updated color palette), and let the AI produce a cleaner version. Two workflows are available: Sketch Cleanup (15 credits) and Figure Cleanup Before Submission (15 credits).

Statistical Plots & Charts — You can use it to generate publication-ready bar charts, line graphs, scatter plots, ablation study comparisons, ROC/PR curves, confusion matrices, and loss curves—all from your research text or data description. The system automatically determines chart type, axis variables, grouping, and statistical annotations.

Reference-Guided Figure Variants — You can use it to explore alternative visual presentations while keeping your core information intact. Upload a reference image as a style and layout guide, and PaperBanana generates variants that adopt better spacing, cleaner labels, or more polished formatting. Cost: 20 credits per variant.

Self-Critique Loop — You can use it to let the system automatically evaluate its own output for figure fidelity, label clarity, and spatial arrangement. Set the iteration count: fewer iterations for quick drafts, more for deep review. The loop refines wording, spacing, and composition without requiring manual redraws.

Who Uses PaperBanana — Real Scenarios That Put It to Work

Not sure if PaperBanana fits your workflow? Here are the situations where it shines brightest.

Doctoral students racing against a deadline. Suppose you're submitting to a top conference in three days. You need five methodology diagrams, three benchmark comparison charts, and two ablation studies—and you've never been comfortable with graphic design tools. Paste your method text into PaperBanana, let it draft the figures, then use the self-critique loop to polish them. What used to take hours now takes minutes.

Lab researchers turning whiteboard ideas into paper figures. You sketched a clever system architecture on the whiteboard during a lab meeting. It captures your idea perfectly, but it's not going to fly in a NeurIPS submission. Snap a photo, upload it to Sketch Cleanup and Redraw, and PaperBanana preserves the original layout while sharpening labels, arrows, and spacing. No need to redraw from scratch.

Independent authors without design skills. Not every researcher has a design background. If you're a solo author or a student publishing your first paper, learning Illustrator or Figma on top of doing actual research is a heavy lift. PaperBanana lowers the barrier: provide clear source text and a figure title, and the AI handles the rest.

Lab teams standardizing figure styles. When five lab members each create figures in their own preferred tool and style, the resulting paper collection looks inconsistent. Using the same PaperBanana workflows and prompt templates across the team ensures everyone outputs publication-level figures with a unified visual language.

Responding to reviewer feedback quickly. A reviewer asks you to update specific labels and values in Figure 3, but the rest of the figure is fine. PaperBanana's Strict Label Rename feature edits only the specified elements—no need to regenerate the entire chart. You can turn around revisions in minutes, not hours.

Preparing conference posters. Poster sessions require multi-panel layouts that integrate the problem, method, results, and impact into a cohesive visual. PaperBanana's Poster Panels workflow automatically organizes multi-panel layouts with coordinated spacing and typography, giving you a conference-ready poster in one go.

Pricing That Keeps It Simple — Pick the Plan That Fits

PaperBanana's pricing philosophy is straightforward: show you the credit cost before every run, and refund unused credits when tasks complete under budget. First-time purchases within 14 days are eligible for a pro-rata refund.

Monthly and annual subscriptions are available, along with one-time credit packs (Pro and Premium tiers) for short-term bursts. New accounts may receive Starter credits to try the platform, and the Free Draft preview lets you test core ideas before committing.

Our advice: If you're an independent researcher or PhD student working on a single paper, the Pro plan (500 credits/month) is usually enough—it covers around 15–20 methodology diagram generations plus cleanup tasks. If your lab is running multiple paper tracks simultaneously, the Premium plan (2,000 credits/month) offers better value and more iteration room for each figure.

What Makes PaperBanana Different Under the Hood

The technical design of PaperBanana reflects a single priority: producing figures that are scientifically accurate, structurally sound, and ready for publication.

Agentic Layout Planning — Instead of generating pixels from random noise like most image models, PaperBanana first plans the entire chart structure: where the title goes, how subcomponents are arranged, which arrows connect which boxes, and how labels are positioned. Only after this layout is finalized does the rendering engine produce the visual output. The result: figures that make logical sense, not just visual sense.

Self-Critique Loop — After generating a figure, the system runs an automatic evaluation on three criteria: fidelity to the source text, clarity of labels, and spatial arrangement. If it detects issues (crowded labels, unclear arrows, poor spacing), it adjusts the layout and re-renders. You control the iteration depth—quick mode for rough drafts, deep mode for publication-quality output.

Multi-Model Support — Choose the engine that fits your needs: GPT Image 2 for general-purpose quality, Nano Banana for academically optimized output, or Nano Banana Pro for higher precision on complex figures. Each model trades off between speed, cost, and quality.

PaperBananaBench Open-Source Benchmark — With 292 test cases based on NeurIPS 2025 styling, this benchmark provides a standardized evaluation framework for academic figure generation. It's fully open-source, so the community can verify performance claims and track progress over time.

7 Specialized Workflows, 4 Categories — The platform covers the full academic figure creation pipeline: Generation (methodology diagrams, statistical charts), Refinement (sketch cleanup, figure editing), Enhancement (poster upscaling), and Video (beta). Each workflow is purpose-built for its specific task.

SVG Vector Output (Beta) — Need to make manual tweaks after generation? The Image to Editable SVG feature converts bitmap figures into vector files, letting you adjust labels, borders, arrows, and colors in any vector editor. It's still in beta, but it's already useful for fine-tuning.

Frequently Asked Questions