PaperBanana

Why we included this project

Researchers who dread the hours spent on method diagrams and statistical plots for a manuscript will find a lot to like here. PaperBanana turns a raw method description and a figure caption into publication-quality visuals by running a small team of specialized agents: one retrieves reference figures, another plans the narrative, a stylist enforces academic aesthetics, a visualizer calls an image-generation model, and a critic loops back for revisions. Because it is reference-driven and iterative, the output tends to match the look of figures in established papers rather than generic AI art, which matters when reviewers expect consistency across a submission. It works with Gemini, OpenRouter, and OpenAI image models, and ships with both Gradio and Streamlit interfaces, plus a refine and upscale workflow for fixing an existing diagram. Teams working on computer-science papers will get the most value today, since the curated reference set is strongest in that area.