PaperBanana: Automating Research Diagrams With an Agentic AI Framework

PaperBanana is Google AI's agentic framework that converts natural language descriptions into publication-ready research figures using a multi-agent architecture. It employs a critic-generator loop with four agents (Planner, Code Generator, Executor, Critic) that iteratively refine outputs until they meet defined publication standards. Unlike single-pass chart generation tools, this approach addresses multi-criteria quality requirements such as typography, color accessibility, and journal-specific formatting.