PaperBanana is an agentic framework designed to automate academic illustration. Users provide methodology descriptions, datasets, sketches, or concept text; five specialized agents (Retriever, Planner, Stylist, Visualizer, Critic) collaborate to fetch references, devise layout and content plans, apply academic styling, render high-resolution images or code-based plots, and iteratively refine outputs until they meet publication quality. Outputs include polished diagrams, accurate statistical plots with downloadable Python code, and enhanced sketches ready for insertion into papers or presentations.