Wispr Flow - Don’t type, just speak
Mini toolsPaperBanana
By dwzhu-pku
PaperBanana automates academic illustration for AI researchers. It uses a team of specialized agents, powered by modern VLMs and image generators, to retrieve reference figures, plan content, design a consistent style, render diagrams or code-based plots, then self-critique and refine them. The authors introduce PaperBananaBench, a NeurIPS-derived benchmark with 292 test and 292 reference diagrams, and show that PaperBanana improves faithfulness, conciseness, readability and aesthetics versus baselines while also generating high-quality statistical plots and enhancing human-drawn figures.
Multimodal
Gen 3
Overview
PaperBanana is an agentic framework that turns raw scientific content into publication-ready methodology diagrams and plots, orchestrating multiple AI agents plus a dedicated benchmark, PaperBananaBench.
Last updated: April 6, 2026
