paperbanana

PaperBanana is a community implementation of the paper "PaperBanana: Automating Academic Illustration for AI Scientists". The README states clearly that it is unofficial, not affiliated with or endorsed by the paper's authors or Google Research, and may differ from the original system. It generates publication-style methodology diagrams and statistical plots from text descriptions, using Google Gemini for both planning and image generation, and offers a CLI, a Python API, an MCP server for IDEs and Claude Code skills for `/generate-diagram`, `/generate-plot` and `/evaluate-diagram`.

The pipeline has two phases and five agents. A Retriever picks relevant examples from a curated set of 13 methodology diagrams, a Planner writes a detailed description of the target diagram, and a Stylist refines it with NeurIPS-style guidelines. Then a Visualizer renders the image (Gemini 3 Pro image preview for diagrams, Matplotlib code for plots) and a Critic reviews it against the source text, repeating for up to three rounds. It needs Python 3.10 or newer and a Google Gemini API key, which the README says is available free from Google AI Studio; `pip install paperbanana` and `paperbanana setup` walk through the key. `paperbanana generate --input method.txt --caption "..."` produces a diagram. It is MIT licensed.