Vector-Based Editable Diagram Generation via GUI Agents
Drive professional vector software via GUI agents to generate editable SVG and PDF diagrams directly from textual style plans.
Source paper
Dawei Zhu, Rui Meng, Yale Song, Xiyu Wei, Sujian Li, Tomas Pfister, Jinsung Yoon
Generated by AI. The paper’s authors did not write these.
Drive professional vector software via GUI agents to generate editable SVG and PDF diagrams directly from textual style plans.
Extract scene graphs using object detection models to symbolically verify structural connectivity before aesthetic review.
Combine executable Python plotting code for precise geometry with controlled image-to-image diffusion models for publication-ready aesthetics.
Apply test-time scaling to generate a diverse spectrum of aesthetic variants before final rendering to prevent style standardization.
This paper introduces PaperBanana, an agentic framework designed to automate the generation of publication-ready academic illustrations, specifically methodology diagrams and statistical plots. The system orchestrates five specialized agents (Retriever, Planner, Stylist, Visualizer, and Critic) powered by Large Language Models and image generation models to convert textual methodology descriptions into high-fidelity, aesthetically compliant figures.
The framework utilizes a retrieval-augmented generation pipeline where a 'Retriever' selects relevant visual examples, a 'Planner' creates a structural description, a 'Stylist' enforces academic aesthetic guidelines, and a 'Visualizer' renders the image. A 'Critic' agent then performs iterative self-correction loops to refine the output based on faithfulness and visual quality, evaluated against a newly introduced benchmark, PaperBananaBench.