How to Make Scientific Figures With AI: A Researcher's Step-by-Step Guide

You spent three months collecting data, running analyses, and writing the paper. Then you hit the part nobody warns you about: making the figures.

The bar charts look flat. The color scheme clashes with itself. The resolution is wrong for the journal's requirements. And every time you think it's ready, the reviewer asks you to remake everything at a different aspect ratio.

This is not a skill problem — it's a workflow problem. Most researchers are asked to produce publication-quality figures without being taught how. The good news is that AI-assisted figure tools have changed what's possible in a single afternoon.

This guide walks you through a complete workflow for how to make scientific figures with AI in 2026: from a rough sketch or spreadsheet to a publication-ready figure that passes journal checks. Drawing on experience preparing figures for peer-reviewed publications across multiple research fields, I'll show you where AI helps, where it still falls short, and exactly what to verify before submitting.

Why Scientific Figure Generation Is Different From General AI Image Making

A scientific figure maker needs to do things that general image generation does not prioritize. Understanding this difference upfront saves hours of rework.

What a Research Figure Must Do

General AI image generators are optimized for visual appeal, not for data integrity. An AI scientific figure generator, when used correctly, treats these requirements as constraints rather than afterthoughts.

Where AI Fits in the Scientific Figure Workflow

Before we begin, here is the honest split:

AI excels at: layout suggestions, color scheme selection, diagram generation from natural-language prompts, background removal, resolution upscaling, batch styling, and generating conceptual illustrations (pathways, mechanisms, schematics).

AI cannot do (and you must verify): match numeric data to visual elements, enforce journal-specific style guides, guarantee axis-scale correctness, or replace peer review of figure accuracy.

Before You Start: Collect Your Materials

A streamlined research figure generation workflow starts before you open any tool. You need three things on your desk:

1. Your raw data or data table — the actual numbers, not a screenshot
2. The journal's figure guidelines — DPI, color mode, file format, margin requirements
3. A rough layout sketch or reference — even a napkin drawing saves rounds of AI iteration

Having these ready makes the prompt-to-figure process substantially faster. You will tell the AI what data to use and what format to produce, rather than iterating on "make it look more scientific." With your materials assembled, the next decision is choosing a tool that fits both your figure type and your workflow.

Step 1: Choose Your AI Scientific Figure Generator

Not all tools are equally suited for creating scientific figures. Here is how the common options compare for figure creator for papers purposes.

For this tutorial, we use Paper Banana as the primary tool because it combines natural-language input with export settings designed for academic publishing. Most of the workflow principles apply to any AI figure generator with minor adjustments.

Step 2: Write Your First Prompt — From Text to Figure Draft

The prompt-to-figure process is the core skill in AI-assisted scientific figure creation. A well-structured prompt produces a usable draft in one attempt. A vague prompt produces something artistic but useless for a paper.

The Anatomy of a Scientific Figure Prompt

A complete prompt has five components:

[Figure type] + [Content / data description] + [Layout / panel structure] + 
[Style / color constraints] + [Output format]

Example: Prompt for a Mechanism Diagram

The strong prompt works because it gives the AI specific constraints. You can always adjust layout and color later, but starting with constraints reduces unnecessary generation rounds.

Example: Prompt for a Data Figure

For data-driven figures, be explicit about the variable names and axis ranges:

This level of specificity is what differentiates a professional figure maker for research paper use from casual image generation.

The prompt-to-figure pipeline: from structured prompt to layout draft. The AI interprets the prompt and produces a figure that can be refined in subsequent steps.

Step 3: Refine Layout and Readability

The first draft from your AI scientific figure generator will rarely be submission-ready. Three rounds of refinement is normal.

Round 1: Composition and Alignment

Check these elements in order:

• Margins: Is there enough whitespace around the figure for journal layout?
• Label visibility: Can all axis labels and legends be read at the printed size?
• Panel balance: If multi-panel, do the panels have roughly equal visual weight?
• Font consistency: Are all labels in the same font family and size?

Most AI figure generation tools allow in-app editing for repositioning labels and adjusting spacing. If your tool does not, take the figure into a vector editor (Illustrator, Inkscape, or PowerPoint) for this round.

Round 2: Color Accessibility

Approximately 8% of male reviewers have some form of color blindness. Your figures must work without color.

Quick accessibility check:

• Convert the figure to grayscale. Can you still distinguish all data series?
• Do you rely only on red-green differences? If yes, add shape markers or patterns.
• Is the background pure white (not off-white)? Journals print on white paper.

Use color-blind friendly palettes — the ColorBrewer set or the Wong palette are safe defaults for publication-ready diagrams. A responsible scientific figure maker never assumes color is available to every reader.

Round 3: Journal Style Compliance

Open your journal's author guidelines alongside your figure. Check against these common requirements:

Exporting at the correct resolution from the start avoids quality loss from downscaling later.

Example of a properly annotated multi-panel figure. Panels A and B share consistent axes, font, and legend placement.

Step 4: Annotate and Style for Publication

The figures that impress reviewers have two qualities: clarity and consistency. The scientific figure maker you use should support both.

Annotation Rules

Font and Size Guidelines

For journal-safe limitations, stick to these safe defaults unless your target journal specifies otherwise:

• Font: Arial or Helvetica (sans-serif) for readability
• Label size: 7–9 pt for axis numbers, 9–11 pt for axis titles
• Panel labels: 11–12 pt bold
• Line weight: 0.5–1 pt for data lines, 1.5–2 pt for axes

These sizes ensure that a figure printed at full-page width (roughly 7 inches) remains readable. Going smaller works in digital-only journals but risks being illegible when printed.

Step 5: Data Accuracy Verification — The Most Important Step

This step is not negotiable. Every AI-assisted figure should go through a data accuracy check before it enters your paper.

What to Verify

The Triangle Check

A reliable method: place your raw data table, the AI-generated figure, and your analysis script or notes side by side. Pick three data points at random — one from the highest value, one from the lowest, one from the middle — and verify they appear correctly in the figure.

If even one check fails, audit the entire figure before using it. Data errors in published figures can lead to corrections, retractions, or loss of credibility.

Step 6: Export With the Correct Settings

Your AI scientific figure generator may offer multiple export options. Choosing the wrong format causes issues at submission time.

Export Cheat Sheet

A note on file size: A full-page TIFF at 300 DPI CMYK can easily be 40–80 MB. That is normal. Some journals have file size limits — check before exporting, and use LZW compression (lossless) to stay under the cap.

What to Name Your Figure File

Journals receive dozens of figure files per submission. Use a consistent naming convention:

Figure_1_mechanism_diagram_v3.tiff

Avoid: final_final_actuallyfinal.png, fig1.png (too vague), or untitled.jpg.

Troubleshooting Common AI Figure Problems

Even with a good workflow, AI scientific figure generation can produce unexpected results. Here are the most common scenarios and how to resolve them.

Problem 1: Labels Are Blurry or Distorted

Symptom: Text in the exported image appears pixelated or broken.

Root cause: The AI rendered labels as raster pixels at too low a resolution, or used a non-standard font that did not render cleanly.

Resolution strategy: Re-generate at the target DPI from the start (do not upscale). Specify exact font names in the prompt. For critical labels, add them manually in a vector editor — this is the most reliable fix for publication-ready diagrams.

Problem 2: Color Shifts Between Draft and Export

Symptom: The figure looks correct in the AI tool but colors shift after export.

Root cause: The display uses an RGB color space, but the export uses CMYK conversion with a different profile.

Resolution strategy: Export in the same color mode your journal requires. If you need CMYK, check the conversion before exporting. Most AI figure generators now include CMYK preview mode — use it before final export.

Problem 3: Data Points Misaligned With Axis Ticks

Symptom: Bars or scatter points do not align with the grid or tick marks.

Root cause: The AI interpreted relative positions rather than exact data coordinates.

Resolution strategy: This is one area where AI still consistently struggles. For data-heavy figures, generate the chart in a dedicated plotting tool (Matplotlib, R ggplot2, or GraphPad Prism) and use the AI tool only for layout and styling around it. Do not rely on an AI figure generator to place data points precisely from a text description.

Left: common export problems — blurry labels, color banding, and misaligned data. Right: corrected versions after applying the resolution strategies above.

Problem 4: Single-Panel Figure Needs to Become Multi-Panel

Symptom: You have several related figures that should be combined, but each was generated separately.

Root cause: The prompt-to-figure process produced individual outputs without considering the combined layout.

Resolution strategy: Re-prompt with the combined dataset: "Combine the following figures into a three-panel layout with shared X-axis." Provide each panel's description separately. If the AI tool supports layer import, bring in separate generated elements and arrange them with consistent spacing.

AI-Assisted Editing: What to Automate and What to Keep Manual

Knowing which steps to delegate to AI and which to do manually doubles the efficiency of your research figure creation workflow.

Delegate to AI

Keep Manual

The Complete Research Figure Creation Checklist

Before you submit, run through this checklist. It covers every stage of the workflow from prompt to PDF export.

Pre-Generation

• Raw data table is complete and verified
• Journal figure guidelines are open for reference
• Layout sketch or reference diagram is ready
• Figure type and panel structure are decided
• Target resolution and color mode are set

Prompt Engineering

• All five prompt components are included (type, content, layout, style, format)
• Color palette is accessible (grayscale and color-blind checked)
• Font family and sizes are specified
• Axis ranges and labels are explicitly stated for data figures

Post-Generation

• Data accuracy triangle check is complete
• Labels are crisp and consistent across all panels
• Color accessibility test passed (grayscale conversion)
• Statistical markers are correct
• Error bars are labeled with measure type
• Figure legend is written in complete sentences
• File is exported at correct DPI and color mode
• File size is within journal limits
• File name follows a consistent convention

Pre-Submission

• Colleagues or co-authors have reviewed the figure for clarity
• Data has been independently verified by a second person
• Source data files are archived alongside the figure
• The AI tool and prompt used are noted in lab records

Responsible Use of AI-Generated Figures

Before you submit, consider the ethical and practical guardrails around AI-assisted figure generation.

Always disclose AI assistance. Many journals now require authors to declare whether AI tools were used in figure preparation. Check your target journal's policy before submitting — some require disclosure in the figure caption, others in the methods section.

Keep original data and prompts accessible. Archive your raw data files, analysis scripts, and the exact prompts used alongside the final figure. If a reviewer asks about methodology, you should be able to reproduce the figure from scratch.

Do not let AI styling mask data errors. A figure that looks polished but contains a data error is worse than a plain figure that is accurate. Verify the underlying data representation before applying AI layout and styling.

Document AI tool version in lab records. Record which tool and version number were used for each figure, along with the prompt history. This is becoming standard practice in labs with formal data management protocols and is increasingly requested during manuscript data audits.

Core Summary

Your figures are the visual backbone of your paper — what reviewers scrutinize first and what readers remember. AI-assisted scientific figure creation saves researchers hours of manual layout work, but only when used within its limits. The workflow that consistently produces publication-ready figures follows the same pattern every time: prepare your data first, write a structured prompt, generate a draft, refine layout and color, verify every data point manually, and export at the correct settings.

The tools will improve, but the verification step will not go away. A scientific figure maker is a productivity tool, not a replacement for scientific judgment. Your data, your analysis, and your review process are what make a figure trustworthy.

If you are just starting with research figure generation, begin with a single figure type you already know well (a bar chart, a pathway diagram) and run the full checklist once end-to-end. That one cycle teaches you more than reading ten guides.

Next step: Open your dataset, write your first figure prompt with all five components, generate a draft, and run the triangle check. That is the fastest way to build a reliable scientific figure workflow.