Developer argues structured specs, not prompts, are key to accurate scientific diagrams

A developer working on text-to-figure generation found that vague prompts like 'Draw the MAPK pathway' leave critical scientific decisions to the AI model, often producing visually polished but scientifically misleading results. To address this, they began drafting small figure specifications before generation, explicitly defining purpose, entities, relationships, and what to exclude. They also advocate for typed edge definitions that distinguish between activation, inhibition, transport, and sequence, rather than relying on ambiguous arrows or color alone. The developer further recommends tagging every figure element by its evidence category — measured data, cited literature, or interpretation — to prevent generated illustrations from implying false precision. This structured approach, they argue, makes scientific figures easier to validate, review, and correct before visual rendering obscures the underlying assumptions.