PaperPlot

PaperPlot is an agentic AI tool for researchers that generates methodology diagrams and statistical plots from text or rough sketches, with a self-critique refinement loop aimed at publication-ready output.

PaperPlot is an agentic framework built for AI researchers who need publication-ready figures without spending hours in drawing software. It works from two kinds of input: a written description of a methodology or a rough hand-drawn sketch. From a text prompt it plans a layout and renders model architecture diagrams, flowcharts and system overviews. From a sketch it interprets the visual intent and returns a polished, vector-style diagram while keeping the original layout and style consistent. The workflow runs through specialized agents rather than a single image call. A Retriever gathers context, a Planner designs the layout, a Renderer produces the first image and a Critic evaluates the result and drives iterative refinement against faithfulness, conciseness and aesthetics. Beyond diagrams, the tool also generates statistical plots intended for academic presentation, offering publication-style charts in 2K or 4K resolution with aspect ratios from 1:1 up to 21:9. This is aimed squarely at academic authors, graduate students and research engineers preparing papers for venues like NeurIPS, where figure quality is judged alongside the science. The team backs its claims with PaperPlotBench, an open benchmark of 292 curated test cases from NeurIPS 2025, and positions the output as outperforming general-purpose image generators on scientific accuracy and layout discipline. The practical trade-off is specialization. PaperPlot is not a general illustration or design tool, and its value depends on how precisely you can describe your method in words. Researchers who already maintain careful figure templates in Illustrator or TikZ may find the agentic output needs manual cleanup, but for teams that treat figure creation as a bottleneck, it targets the most repetitive part of the job: the first draft.