AutoFigure: Generate Publication-Ready Scientific Diagrams from Text

By Prahlad Menon 2 min read

Creating figures for research papers is tedious. You know what you want to show, but translating that into a polished diagram means hours in draw.io, PowerPoint, or—if you’re ambitious—TikZ.

AutoFigure generates publication-ready scientific illustrations from text descriptions or directly from research papers.

How It Works

AutoFigure uses a dual-agent system with iterative refinement:

  1. Generate — An LLM creates initial SVG/XML from your description
  2. Evaluate — A critic agent scores quality (0-10) and provides feedback
  3. Refine — Loop continues until the figure meets publication standards

This Review-Refine loop is key. First drafts are rarely perfect, but AutoFigure keeps improving until the diagram is actually usable.

Two Modes

Text-to-Figure — Describe what you want:

"A flowchart showing the transformer architecture with 
attention mechanism, feed-forward layers, and residual connections"

Paper-to-Figure — Upload a PDF, and AutoFigure extracts the methodology and creates visual diagrams automatically. No manual description needed.

Output Formats

  • SVG — Vector graphics, scales perfectly
  • mxGraph XML — Fully compatible with draw.io for manual editing

The draw.io compatibility is clever—you get AI-generated starting points that you can refine manually if needed.

Getting Started

git clone https://github.com/ResearAI/AutoFigure.git
cd AutoFigure
pip install -e .
playwright install chromium

Basic usage:

from autofigure import generate_figure

figure = generate_figure(
    description="Neural network architecture with encoder-decoder structure",
    output_format="svg"
)

There’s also a web interface for interactive generation and editing.

Why This Matters

Scientific communication relies heavily on figures. A good diagram can explain in seconds what takes paragraphs of text. But figure creation is a bottleneck:

  • Researchers spend hours on diagrams instead of research
  • Quality varies wildly based on design skills
  • Iteration is slow when changes require manual redrawing

AutoFigure treats figure generation as a first-class AI task, not an afterthought.

My Take

I’ve seen many “AI diagram generators” that produce unusable output. AutoFigure’s iterative refinement approach is what makes it different—it actually critiques and improves its own work.

The paper-to-figure mode is particularly interesting for literature reviews. Feed it papers and get visual summaries without manually reading methodology sections.

Links: