Paper Banana:
Automating Academic Illustration

Paper Banana generates two main types of academic figures: methodology diagrams (system architectures, model pipelines, experimental workflows, algorithm flowcharts) and statistical plots (bar charts, line graphs, scatter plots, heatmaps, confusion matrices). All outputs are exported as high-resolution PNG images suitable for direct insertion into journal papers, conference submissions, theses, and presentation slides.

Your input passes through five specialized AI agents in sequence. The Retriever searches a curated database of reference diagrams from top venues like NeurIPS and ICML. The Planner translates your text into a detailed structural description. The Stylist applies academic visual standards — colors, fonts, spacing. The Visualizer renders the final image. Finally, the Critic evaluates the output against your source text and triggers refinements if needed. This iterative loop produces figures that consistently meet publication standards.

Each iteration costs 5 credits. With the default 3 iterations, a typical generation uses 15 credits. You can choose fewer iterations (faster, cheaper) or more iterations (higher quality, more expensive). The Critic agent may approve the result early, in which case unused iteration credits are not charged.

Paste your data as JSON with labeled arrays. For example: {"labels": ["Model A", "Model B"], "accuracy": [94.2, 91.5]}. You can also include multiple datasets for grouped charts. Describe the visualization you want in the caption field — bar chart, line plot, heatmap, etc. — and Paper Banana will select the optimal chart type and styling.

Yes. Paper Banana applies formatting conventions from top academic venues including NeurIPS, ICML, CVPR, Nature, IEEE, and ACM. Output resolution goes up to 4K, and figures are exported as PNG — the standard format accepted by all major publishers and LaTeX document systems.

No. Just paste your paper text or data and describe what you want in plain English. Paper Banana handles all visual design decisions — layout, color palette, typography, arrow routing, spacing — automatically, based on the conventions of your target venue.

The output is a high-resolution PNG image. You can import it into any image editor (Photoshop, Figma, Canva) for minor adjustments. For structural changes, we recommend modifying your input text and regenerating — Paper Banana's iterative pipeline often produces better results than manual post-editing.

A typical generation with 3 iterations takes about 2 minutes. The main time cost is the image rendering step (~60 seconds per iteration). You can monitor real-time progress in the Pipeline Log. If you need faster results, set iterations to 1 — you'll get a result in about 90 seconds, though with fewer refinement passes.