LibreAI / Design / paperbanana
// design · llmsresearch/paperbanana

paperbanana

PaperBanana — agentic generation of publication-quality academic figures, diagrams and plots from text

★ 2.4k stars⑂ 341 forksMITupdated 2026-09
▦ visual — this project's README ships no embeddable screenshot,
so here's a placeholder. The guide below is complete regardless.
// project facts

Project facts

paperbanana is an open-source design project by llmsresearch (llmsresearch/paperbanana on GitHub). It has 2,388 GitHub stars and 341 forks, is released under the MIT license, and was last updated on GitHub on 17 September 2026.

GitHub stars
2,388
Forks
341
License
MIT
Last updated on GitHub
17 September 2026
Category
Design
Listing
Full install guide
// how it works

How it works

  1. Clarify your goal

    Prepare your methodology text (or a paper PDF) and a precise caption describing the figure you want; for plots, prepare the CSV/JSON data and the chart's intent.

  2. Get PaperBanana

    pip install paperbanana, or clone and install from source with the provider extras you need.

  3. Configure a provider

    Copy .env.example to .env and add an OpenAI, Azure or Gemini key (Gemini's free tier works), or run paperbanana setup for the Gemini wizard.

  4. Generate and refine

    Run paperbanana generate with your input and caption; add --optimize and --auto for input optimization and critic-driven auto-refinement, then review the iterations in outputs/.

// features

Why people love it

Multi-agent figure pipeline

Up to 7 specialized agents: input optimizer (context enricher + caption sharpener), retriever, planner, stylist, visualizer and critic, in a two-phase plan-then-refine flow.

Diagrams and statistical plots

generate makes methodology diagrams; plot renders statistical plots via VLM-generated matplotlib code — no image-generation credentials needed for plots.

Iterative refinement with feedback

Refinement rounds (default 3), --auto to loop until the critic is satisfied, and --continue to resume a run with user feedback.

Batch and PDF inputs

Batch diagrams from a YAML/JSON manifest, batch plots per-item from CSV/JSON, and PDF inputs with per-page selection (optional PyMuPDF extra).

Studio, MCP and skills

A local Gradio Studio (paperbanana studio), an MCP server for IDE integration, and Claude Code skills for /generate-diagram, /generate-plot and /evaluate-diagram.

Venue style packs

Built-in NeurIPS, ICML, ACL and IEEE style packs, with scaffolding to create your own from example figures.

Advertisement
// run it

Run it yourself

  1. Clone the repo

    Clone PaperBanana if installing from source; PyPI installation is step 2's alternative.

    git clone https://github.com/llmsresearch/paperbanana.git
    cd paperbanana
  2. Install

    Install from PyPI, or from source with the provider extras shown in the README.

    pip install paperbanana
    # source: pip install -e ".[dev,openai,google]"
  3. Generate a diagram

    Set your key in .env first (OPENAI_API_KEY or GOOGLE_API_KEY), then generate from the sample input.

    paperbanana generate --input examples/sample_inputs/transformer_method.txt --caption "Overview of our encoder-decoder architecture with sparse routing"
// use cases

What you can build

Advertisement
// faq

Questions, answered

Who maintains PaperBanana?

llmsresearch maintains this unofficial, community-driven implementation of the PaperBanana paper (Zhu et al.); it is not affiliated with or endorsed by Google Research or the original authors.

Is it free?

The code is MIT-licensed. Generation calls external VLM/image APIs — Gemini has a free tier, while OpenAI/Azure/Atlas usage is billed by those providers.

Does it need a paid API key?

It needs an API key for a VLM/image provider, but a free Google Gemini key from AI Studio is enough to start; statistical plots need no image provider.

How does it compare to drawing figures manually or a single image-model prompt?

A single prompt gives one unreviewed image. PaperBanana plans from curated reference diagrams, styles to venue guidelines, and critiques/revises its own output over several rounds — and can continue from your feedback.

How do I get started?

pip install paperbanana, put a Gemini or OpenAI key in .env, and run paperbanana generate on the sample input — or try the Colab quickstart with no local setup.

// keep exploring

Similar tools

Design

openchatbi

Chat with your data — natural-language BI with SQL generation

★ 658 starsMIT
View guide →
Design

draw.io Desktop

draw.io Desktop — the official offline desktop app for diagrams.net: flowcharts, architecture diagrams and wireframes in an Electron app that never sends your diagram data anywhere, GPL-3.0

★ 63.6k starsGPL-3.0
View guide →
Design

GrapesJS

Free and Open source Web Builder Framework. Next generation tool for building templates without coding

★ 26.3k starsCustom (check repo)
View guide →
Design

vscode-drawio

This unofficial extension integrates Draw.io (also known as diagrams.net) into VS Code.

★ 9.5k starsGPL-3.0
View guide →
Design

beamer-academic

Turn your thesis into defense slides with one command.

★ 306 starsMIT
View guide →
Design

diagrams

:art: Diagram as Code for prototyping cloud system architectures

★ 42.7k starsMIT
View guide →
// alternatives

Alternatives in Design

The most-starred other Design projects in the LibreAI catalogue — the realistic shortlist if paperbanana is not the right fit.

Get paperbanana

Free and open — grab the source, star the repo, and start creating.