How to Design a Diagram with Labels That Communicates

You've finished the anatomy, checked the arrows, and matched the terminology to the manuscript. Then a reviewer asks you to relabel a structure, replace a unit, or make the panel hierarchy clearer. Because the labels were placed directly on the artwork, the “small” correction becomes a rebuild.

That failure is common in medical and scientific communication. A diagram with labels isn't just an illustration with text added at the end. It's a coordinated system of typography, spacing, hierarchy, contrast, leader lines, captions, and accessible descriptions. Treat labels as their own design layer, and revisions become controlled edits instead of emergency surgery.

Table of Contents

• The Moment Every Labeled Figure Has to Be Rebuilt
  • Labels carry meaning
  • The workflow change
• Planning the Figure Before You Touch a Canvas
  • Start with the audience and venue
  • Build the label inventory
  • Choose the right figure structure
• Labeling Conventions That Survive Peer Review
  • Set typography once
  • Place panels and units systematically
  • Control leader lines
  • Reduce rather than squeeze
• Color, Contrast, and Designing for Every Reader
  • Test the label in context
  • Write for screen readers too
• File Formats, Resolution, and Handoff
• Producing Labeled and Unlabeled Variants From One Source
  • Organize the source file
  • Decide where information belongs
• A Practical Checklist and Common Questions
  • What should I do when there are too many labels?
  • How can one figure work in print and on screen?
  • What belongs in alt text?
  • Should labels stay on the image or move to the caption?

The Moment Every Labeled Figure Has to Be Rebuilt

The request usually sounds harmless: move this callout, clarify that structure, make the panel letters consistent. In practice, the label may be embedded in the background, the leader line may pass through a critical feature, and the available margin may already be occupied by another annotation. Changing one label shifts several others, so the artist starts nudging objects until the figure loses its original balance.

I've seen this happen with anatomy plates, surgical schematics, pathology figures, and workflow diagrams. The artwork itself was accurate. The problem was that nobody had defined how the labels should behave when the figure changed.

Labels carry meaning

A label identifies more than a part. Its position can imply ownership, sequence, hierarchy, or connection. A line that terminates on a membrane rather than the lumen can change the interpretation. A panel letter that sits too close to the artwork can be mistaken for a data annotation. A unit omitted from a measurement can make an otherwise correct figure incomplete.

The AI2D corpus helped formalize this problem by modeling diagrams through four element types, text, graphics, arrows, and arrowheads, along with ten semantic relations between them. Its successor, AI2D-RST, extended that work to a corpus of 1,000 science diagrams with expert annotations covering grouping, macro-grouping, connectivity, and discourse structure. The research shows why labels shouldn't be treated as informal decoration. Text and image elements participate in a structured visual language. The AI2D-RST research provides a useful foundation for thinking about those relationships systematically.

Practical rule: Design the label system before you finalize the illustration. If the label layer can't be edited independently, the figure isn't ready for review.

The workflow change

Start with a label inventory, not a brush or shape tool. Decide which terms must appear on the figure, which belong in the caption, and which require an accessible description. Establish the typeface, hierarchy, line style, and contrast before placing the first callout.

This approach also respects the educational role of labeled diagrams. Annotated drawing has long been used to help learners identify structures, distinguish diagram features, and reason about the function of labeled parts. Historical evidence places scientific illustration within a much older visual tradition, from illustrations used to document nature to early medical illustration in Hellenic Alexandria. Science education research on annotated drawing connects drawing, labeling, and explanation with the construction of conceptual understanding.

The label is not a finishing touch. It's one of the figure's primary instruments for teaching, identifying, and verifying meaning.

Planning the Figure Before You Touch a Canvas

A short planning pass prevents most labeling rework. Before drawing, write down what the viewer must identify, compare, follow, or remember. That purpose determines whether you need a single panel, multiple panels, or an exploded view.

Start with the audience and venue

A surgeon reviewing an operative approach needs different labeling from a student studying anatomy. A journal reader may need precise structures and units, while a presentation audience needs a faster visual scan. A malpractice attorney may need a clear evidentiary sequence, not every anatomical substructure.

Define the delivery context first:

1. Publication: Keep on-figure text concise, precise, and legible at final page size. Put methodological detail in the caption or legend.
2. Presentation: Increase visual hierarchy and reduce competing labels. A slide viewed from a distance needs fewer, stronger callouts.
3. Teaching: Prepare both an answer version and a clean version. The unlabeled version supports recall rather than recognition.
4. Compliance or review: Preserve terminology, revision history, and a stable mapping between each label and its target.

For presentation planning, a reusable scientific presentation template can help establish panel proportions and text hierarchy before the figure is populated.

Build the label inventory

Make a list of every structure, process, measurement, and panel that could require identification. Then assign each item a destination:

• On-figure label: Use for objects the reader must identify immediately.
• Caption or legend: Use for conditions, methods, abbreviations, and explanatory context.
• Accessible description: Use for relationships, sequence, and information that visual positioning alone communicates.

Sketch the panel layout using empty rectangles before you draw the anatomy. Mark the intended reading order and reserve margins for leader lines. If a label has nowhere to go in the sketch, it won't magically find a good location later.

Choose the right figure structure

A single panel works when one view tells a coherent story. A multi-panel figure is better when the reader must compare stages, treatments, or perspectives. An exploded view earns its complexity when spatial separation clarifies relationships that overlap in the intact object.

Avoid the temptation to create a “master” figure containing every available detail. More content can produce less understanding when the viewer can't distinguish the primary message from supporting information. Each panel should answer a clear question, and the label budget should reflect that question.

Pick the typeface and approximate size during planning, not after the artwork is crowded. Typography determines how much margin you need, how long leader lines must be, and whether a panel can support its intended number of callouts.

Labeling Conventions That Survive Peer Review

Reviewers usually don't object to labels because they dislike annotation. They object when labels make the figure difficult to interpret or impossible to cross-check. Consistency gives the reader a stable visual grammar.

Set typography once

Use one typeface throughout the figure unless a journal style sheet requires otherwise. Keep label size, capitalization, weight, and punctuation consistent. For publication figures, 8 to 10 point text is a practical legibility range specified in AHA/ASA guidance, and AVMA figure instructions emphasize readable, consistent figure labeling.

Use typographic hierarchy sparingly. Panel letters can be bold, labels can use regular weight, and a legend can use a slightly different treatment if needed. Don't use several fonts to distinguish categories that could be separated through position, line style, or grouping.

A label should remain readable at the figure's final displayed size. Don't judge it only on a large working canvas. Shrink the figure to the approximate manuscript, slide, or web size and inspect whether the text still reads without effort.

Place panels and units systematically

Panel letters usually belong in a fixed position, commonly the upper-left corner, with the same font and size in every panel. Label panels first. Add units in parentheses next. Then check that each panel is referenced consistently in the text, caption, and file name.

This sequence catches errors before submission. A figure can contain scientifically correct data and still require revision if the manuscript calls one panel “B,” the file names it “C,” and the caption uses a different label.

Control leader lines

Leader lines should terminate precisely on the intended structure. Keep them thin, direct, and visually subordinate to the artwork. Straight lines are preferable when they don't cross important features. Use an elbow or curved path only when it prevents collisions or clarifies the target.

Never let leader lines cross one another casually. Crossings create ambiguous intersections, especially in dense vascular, neural, or process diagrams. If several labels point to adjacent structures, align the text blocks and create a deliberate routing pattern rather than allowing each line to find its own path.

A leader line is an address, not decoration. If the endpoint is ambiguous, the label has failed even when the text is correct.

For teams creating animated explainers or motion graphics, the same discipline applies to changing on-screen text. A practical overview of AI video text workflows with RemotionAI is useful when labels must remain synchronized with visual movement rather than sit permanently on a static figure.

Reduce rather than squeeze

Overlabeling is the most frequent design failure. When every repeated structure receives a callout, the figure becomes a thicket of lines and text. Label the first instance, representative structure, or clinically relevant feature, then explain repetition in the legend.

If the image is crowded, move supporting detail into the caption. A strong caption should provide enough information to interpret the figure without forcing the reader to search the main text, while the image itself should preserve quick visual recognition. Nature Methods guidance emphasizes consistent type, stable label relationships, and legibility at final page size. Those principles matter more than adding another decorative annotation.

Color, Contrast, and Designing for Every Reader

Color can organize a diagram, but it shouldn't carry meaning alone. A reader may view the figure under poor lighting, print it in grayscale, encounter it on a low-quality display, or use assistive technology that can't infer color relationships. A trustworthy diagram with labels remains interpretable when color is removed.

Use dark text on a light background for most labels. White text can work on a dark, solid fill, but only when the fill remains stable behind every character. Avoid placing text directly over a textured image unless you add a controlled background panel or halo that preserves the letter shapes.

Red and green are a risky primary pairing because some readers won't distinguish them reliably. Separate categories with shape, line pattern, symbol, position, or line weight as well as hue. A dashed boundary, distinct arrowhead, or explicit legend can preserve meaning when the palette changes.

Test the label in context

Contrast belongs to the label and its immediate background, not to the overall figure. A black label may be highly readable over white space and nearly disappear over a dark anatomical surface. Check every callout at its actual endpoint, including areas where the background changes.

Use a contrast ratio checker tool during the design pass rather than waiting for an accessibility review. The verified accessibility guidance for scientific figures recommends concise alt text, sufficient contrast, non-color cues, and longer stepwise descriptions when a complex process can't be summarized adequately in a short phrase. ACS guidance on accessible scientific figures frames accessibility as part of figure construction, not a cosmetic audit.

Test the figure in three views:

• Full size: Check endpoints, line crossings, and local contrast.
• Thumbnail size: Check hierarchy and whether the primary message survives reduction.
• Grayscale or color-vision simulation: Confirm that categories remain distinct without hue alone.

The accessibility context also matters for public-facing work. The supplied guidance identifies WCAG 2.1 AA as becoming a legal standard for public agencies in April 2026. That projection makes it sensible to create accessible source files now, particularly for hospitals, universities, and public institutions.

Write for screen readers too

Alt text shouldn't repeat every visible label. It should summarize the figure's purpose and explain relationships that a sighted reader obtains from layout, arrows, grouping, or sequence. For a complex process, provide a concise summary plus a longer structured description elsewhere.

A useful pattern is: identify the figure type, state the central relationship, name the major groups, and describe the direction or sequence. Don't write “a diagram with labels” and stop there. That tells the reader what the object is, not what it communicates.

File Formats, Resolution, and Handoff

Export decisions can undo careful label design. Vector formats such as SVG, PDF, and EPS preserve text and line sharpness when the receiving workflow supports them. Raster formats such as PNG and TIFF are appropriate when the destination requires pixels, but the label must be evaluated at the final dimensions rather than only at the enlarged working view.

Keep the source file layered. Store base artwork, labels, leader lines, panel letters, legends, and accessibility notes separately. Embed fonts where the format and workflow allow it, or convert a controlled delivery copy to outlines after preserving an editable master. Font substitution can alter line breaks, label width, and alignment during handoff.

Delivery Target	Recommended Format	Resolution	Notes
Publication	PDF, EPS, or journal-approved vector format	Use vector output when accepted	Confirm final page-size legibility and font handling
Presentation	PDF, SVG, or PNG	Match the slide's displayed dimensions	Test from the intended viewing distance
Web article	SVG or optimized PNG	Match the rendered web size	Preserve readable text and meaningful alt text
Print poster	PDF or high-quality TIFF	Use the printer's requested setting	Inspect labels at the actual printed scale

Raster export needs particular care. The supplied publication guidance requires figures to remain intelligible at final page size, with labels easily legible and captions sufficiently informative for interpretation. Don't solve a tiny label by exporting a larger file if the label remains tiny in the final layout.

For web and commerce workflows, image optimization advice such as resize images for Shopify listings can inform file-size reduction without careless compression. The principle transfers to scientific web figures, but readability remains the priority. Before delivery, open the exported file on a different device and verify that labels, units, panel letters, and symbols survived.

When a figure includes quantitative plots, preserve the source data and editable chart structure where possible. Guidance on creating a scatter plot in Excel is relevant when a chart may need relabeling or alternate exports later. A flattened image is convenient for placement, but it's a poor master for future revisions.

Producing Labeled and Unlabeled Variants From One Source

A publication figure, teaching handout, presentation slide, and compliance record often need different amounts of visible text. Rebuilding each version separately invites terminology drift. The safer approach is to create one source illustration with controlled visibility.

Organize the source file

Use named layers or groups for:

• Base anatomy or structural artwork
• Arrows and flow indicators
• Primary labels
• Leader lines
• Panel letters
• Legends and units
• Accessibility notes and descriptive text

The labeled publication version combines the base artwork and required annotation. The clean presentation version may hide secondary labels while retaining a few orientation cues. The unlabeled teaching version hides the label group entirely, leaving the structures available for recall or testing.

A compliance or legal review version may need visible labels plus a separate document that defines each structure and its relationship to the evidence. It shouldn't rely on the image alone to carry every explanation.

Decide where information belongs

Ask what the reader needs to do with each piece of information:

• If the reader must identify a structure immediately, put the term on the image.
• If the reader needs experimental conditions or procedural context, put that information in the caption.
• If the information describes spatial relationships, sequence, or visual grouping, include it in alt text or a longer accessible description.
• If the label would obscure data or background, move the explanation into the legend and retain only a concise identifier on the image.

This decision prevents the common mistake of forcing one artifact to serve incompatible purposes. Accessibility guidance for complex diagrams supports using concise labels together with separate explanatory text when the visual alone can't communicate the full process.

For medical teams starting from clinical imagery, Natomy's medical illustration workflow describes a source-to-illustration approach that can be adapted to a layered labeling system. The important control is not the generation method. It's whether the resulting source remains editable, reviewable, and capable of producing each required variant without redrawing the anatomy.

Name exports clearly. Distinguish labeled, clean, teaching, and accessible-description packages, and keep the terminology identical across them. A variant system only works when the source of truth is shared.

A Practical Checklist and Common Questions

Before delivery, run the figure through this compact review:

• Purpose: Can you state what the viewer must identify or understand?
• Inventory: Is every required label accounted for?
• Typography: Is one typeface used consistently, with publication text in the 8 to 10 point range when appropriate? AHA/ASA figure guidance
• Placement: Are panel letters fixed, leader lines direct, and endpoints unambiguous?
• Units: Are units present in parentheses where required?
• Accessibility: Does meaning survive without color, and is useful alt text available?
• Variants: Can you export labeled and unlabeled versions by toggling layers?
• Output: Does the chosen format match the delivery target?
• Thumbnail: Does the hierarchy remain clear when the figure is reduced?

What should I do when there are too many labels?

Separate essential identification from explanation. Label the first or most important instance, group repeated structures, and move supporting detail into the legend or caption. If the figure still feels crowded, split it into panels rather than shrinking the type.

How can one figure work in print and on screen?

Design and test at the smallest important final size. Use stable contrast, avoid delicate lines, and export a vector file when accepted. If the print and screen contexts demand different density, create variants from the same layered source instead of forcing one version to serve both.

What belongs in alt text?

State what the figure shows, its central relationship, and the major sequence or grouping. Don't copy every visible label. Add a longer description when arrows, spatial arrangement, or process steps carry information that a short summary can't convey.

Should labels stay on the image or move to the caption?

Keep essential identifiers on the image. Move methods, conditions, definitions, and extended explanations into the caption or legend. If a label would cover data or damage scanning, shorten it on the figure and provide the full explanation elsewhere.

---

Natomy helps doctors, scientists, and researchers create medical and scientific illustrations from source material, including publication-ready anatomical visuals and short animations. Use a layered labeling plan with the resulting artwork, then visit Natomy to explore a workflow that can support publication, presentation, teaching, and review-ready figure variants.