Feature Proposal: Fully Automated Scientific Illustration Generation from Text

Dear OpenAI Team,

I’m writing to propose the development of a fully automated, scientifically accurate illustration generation tool that creates diagrams and visual content directly from structured or natural language input, with applications in:

  • Anatomy & Medical Education
  • Biology & Biochemistry
  • Pharmacology and Systems Biology
  • Environmental & Earth Sciences
  • Engineering Schematics and Physics Diagrams

Why It’s Needed:
Current general-purpose models like DALL·E 3 and Midjourney can generate visually appealing images, but they:

  • Lack biological or clinical accuracy
  • Often hallucinate or distort anatomical structure
  • Do not produce vector-editable or layered outputs needed for academic use

This creates a workflow bottleneck where AI-generated drafts must be manually cross-checked, corrected, and redrawn in tools like BioRender or Illustrator.

Proposed Features:

  • Text-to-Diagram Capability: Accept structured prompts like “cross-section of kidney nephron with labeled glomerulus, proximal tubule, and loop of Henle.”
  • Domain Accuracy: Trained or fine-tuned on expert-verified datasets (e.g., Netter, Gray’s Anatomy, KEGG pathways).
  • Vector Output: Generate SVG or layered editable formats suitable for scientific publication.
  • Optional Data Input: Accept raw biological data (e.g., gene expression profiles, molecular interactions) to automatically create pathway maps or expression heatmaps.

Impact:

  • Drastically reduce time for educators and researchers to produce visuals
  • Improve accessibility for students and clinicians without design backgrounds
  • Expand OpenAI’s value in science, education, and knowledge communication

AM Fouda (MD, PhD)
Professor of Clinical Pharmacology and Applied Therapeutics
Mansoura University Faculty of Medicine, Mansoura 35516, Egypt