Write the subject, action, and key elements in plain language.
2
Pick a publication style
Scientific, line, 3D, editorial, and 2 more — or let Auto skip the style overlay.
3
Generate, edit, export
SciFig renders your text to figure in seconds. Refine text or regions, then ship.
Text-to-Figure FAQ
Common questions about generating figures from text descriptions.
1.
Type your figure. SciFig draws it.
Plain Language Prompts: Paste a description like "signaling pathway of insulin response" and SciFig text to figure generates the full diagram — no drawing skills required.
Every Element Editable: Labels, arrows, legends, molecules — each independently selectable on the canvas. Fix anything in seconds.
Domain-Aware: SciFig understands scientific context across biology, chemistry, materials science, medicine, and physics — not generic AI image generation.
2.
Yes. We show you what top journals already did.
Topic-to-Layout Proposals: Type "mitochondrial apoptosis pathway" and SciFig proposes 3–5 text to figure layout options based on how peer-reviewed journals have visualized similar concepts.
Iterate on Any Panel: Pick the layout that clicks, then refine specific parts with text prompts — no need to restart from scratch.
3.
Those are the three most-requested. And a lot more.
Top 3 Text to Figure Use Cases: Mechanism diagrams, signaling pathways, experimental workflows — the figures researchers actually need for papers.
Full Domain Coverage: Molecular structures, organ cross-sections, cell-cycle diagrams, gel/blot schematics, field-sampling workflows — across biology, chemistry, materials science, pharmacology, and agriculture.
No Template Library Limit: Unlike BioRender's fixed icon set, SciFig text to figure can generate anything you can describe — even niche lab equipment or custom protocols.
4.
No. SciFig runs on the latest top-tier models — outputs hit journal-quality from the first generation.
Top Models, Your Choice: Access and switch freely between GPT Image 2, Nano Banana Pro, and Nano Banana 2 — pick the model that fits your figure type and goal.
Journal-Style Output: Line weights, typography, and palette match peer-reviewed journal conventions — no generic-art look, no obvious AI tells.
You Polish the Last 1%: AI gets the figure 99% there; you fix any remaining label, structure, or color before export. Reviewers will assume you hired an illustrator.
5.
Click and edit. Every element of your text to figure output is independently editable.
Fix in Seconds: Typos, wrong enzymes, misplaced arrows — click the object, type the correction, done. No starting over.
Regenerate Just One Part: Don't like one section? Lock what's finished and regenerate just the part that doesn't fit.
6.
Yes. Lock what works, redo what doesn't.
Section-Level Regeneration: Lock the sections you've finalized; regenerate just the one that needs work. Multi-round refinement, one section at a time.
Element-Level Re-prompt: Swap a specific molecule, rewrite a legend, change one arrow's direction — without touching anything else.
7.
Yes. Full multilingual support, plus the STEM typography reviewers expect.
Any Working Language: Type labels in English, your working language, or mixed. Useful for bilingual presentations, language-localized textbooks, or non-English journals.
Greek Letters, Sub/Superscripts, Math: Full scientific typography — αβγ, H₂O, CO₂, subscripts, superscripts, math symbols — all rendered correctly.
One-Click Language Swap: Generate in your working language for the lab meeting, swap to English for journal submission — same figure, no redraw.
8.
Every format your text-to-figure output needs for paper, deck, or poster.
Direct Exports: PNG (recommended for journal submission), JPG (smaller files for sharing or email), and PPTX (a 16:9 PowerPoint slide with the figure embedded — drop straight into a deck).
Vectorize for Layered Editing: Open the result in Vector Canvas and run Vectorize. One pass produces clean SVG (opens in Illustrator, Inkscape, Figma, or Affinity) plus an editable-shape PPTX where every element is a separately selectable PowerPoint object.
Resolution: PNG/JPG up to 8K — clears 300 DPI at standard journal column widths.
9.
Yes. Your prompts, uploads, and outputs never train our models.
Prompts Aren't Training Data: The text you type — including draft hypotheses and unpublished method descriptions — is used only to generate your figure. Never recycled into model training.
Draft-Safe: Typing out an unpublished experimental design is as protected as uploading a manuscript.
Yours to Delete: Remove any prompt or figure from your history and it's gone from our servers — zero retention.
AI Text-to-Figure Generator — Prompt to Publication | SciFig
