WCAG 1.4.5 Images of Text: The Requirement Teams Keep Getting Wrong
Keisha · AI Research Engine
Analytical lens: Community Input
Community engagement, healthcare, grassroots
AI-assisted · Source-linked · Editorially reviewed · Methodology
Trust note
This article was drafted with AI assistance, reviewed against accessibility.chat editorial standards, and should be treated as research and education rather than legal advice. We prioritize primary sources and correct material errors.

One in four. That's the CDC's estimate (opens in new window) of American adults living with some form of disability. Among them: people with low vision who rely on browser zoom, people with dyslexia who override fonts and line spacing, people with photosensitivity who need high-contrast color schemes. When a designer saves a banner or callout as a flat image rather than styled HTML text, every one of those customizations stops working. That's what WCAG 2.1 Success Criterion 1.4.5 (opens in new window) is designed to prevent.
The Requirement
SC 1.4.5 Images of Text sits at Level AA — the conformance level required by the ADA's 2024 Title II web rule (opens in new window) and the Section 508 refresh (opens in new window). The rule is straightforward: if technology can render the same visual result using actual text, use actual text. Don't substitute an image.
The W3C's Understanding document (opens in new window) frames the intent clearly — this criterion exists to protect users who need to control how text is presented. Font size, foreground and background color, font family, line spacing, alignment: all of these are adjustable by the user when content is real text. None of them are adjustable when the content is a JPEG or PNG.
Three exceptions exist, and they matter:
- Essential presentation — logotypes, brand wordmarks, and type specimens where the specific rendering is the information
- Customizable images — rare cases where users can adjust the image itself to match their preferences
- Technical infeasibility — where achieving the same visual effect with live text would interfere with another criterion (like 1.4.4, Resize Text) or isn't reliably supported across user agents
One clarification the spec makes explicitly: text embedded in photographs, diagrams, charts, or screenshots — where the image contains significant other visual content — falls outside this criterion's scope. A screenshot of a terminal window isn't the same problem as a marketing banner saved as a PNG because the design team wanted a custom font.
Why This Matters
The gap between what a user sees and what a user can use is where this criterion lives.
Consider a low-vision user running Windows with a high-contrast theme enabled. Their operating system inverts colors, strips backgrounds, and boosts text contrast across the interface. Real text responds to this. An image of text does not — it renders exactly as saved, potentially with contrast ratios that become unreadable or visually painful under the user's custom settings.
Now consider someone with dyslexia using a browser extension to switch all body text to a dyslexia-friendly font like OpenDyslexic, or adjusting line height for easier tracking. Again: real text responds. Images don't.
Screen reader users face a different but related problem. An image of text with no alt attribute is invisible to assistive technology. An image with a generic alt like alt="banner" conveys nothing. Even a perfectly written alt attribute — which satisfies SC 1.1.1 Non-text Content — doesn't restore the user's ability to resize, recolor, or reformat the content. Meeting 1.1.1 and meeting 1.4.5 are separate obligations.
This is a pattern worth naming: teams often assume that adding alt text to an image of text resolves the accessibility concern. It doesn't. Alt text addresses discoverability. SC 1.4.5 addresses adaptability. They're solving different problems.
The populations most affected — users with low vision, cognitive disabilities, and reading differences — are often the same populations already navigating the most barriers online. As our research on testing methodology gaps documents, automated tools catch only a fraction of real-world barriers. Images of text are a prime example: a tool can flag an image lacking alt text, but it cannot reliably determine whether an image contains text that should have been live text instead.
How to Meet It
The W3C identifies four sufficient techniques:
CSS Technique C22 (opens in new window) — Use CSS to control visual text presentation. This is the primary path for most web content. Web fonts, letter-spacing, line-height, color, background-color, font-family, text-transform — CSS handles virtually every visual effect that teams reach for images to achieve. If your design system requires a specific display font, load it as a web font and style the live text.
CSS Technique C30 (opens in new window) — Use CSS to replace images of text with styled text, and provide user controls to switch. This technique covers cases where an image of text is used for visual effect but a mechanism exists for users to request the live-text version.
General Technique G140 (opens in new window) — Separate structure from presentation so that content can be presented in different visual forms. This is the architectural principle underlying the more specific techniques.
PDF Technique PDF7 (opens in new window) — For PDF documents, perform OCR on scanned pages to convert image-based text to real text. This addresses the common pattern of scanned documents published as inaccessible PDFs.
Advisory techniques — C12, C13, C14, C8, and C6 — address font size units, font weight, and character spacing in CSS. These aren't required to satisfy the criterion, but they support the broader goal of user-adjustable text presentation.
Common Failure Patterns
| Pattern | Why It Fails SC 1.4.5 | Fix |
|---|---|---|
| Hero banners saved as images with text overlay | Text is not adjustable; may also fail 1.1.1 if alt is missing | Rebuild with HTML/CSS text layer over CSS background image |
| Infographics where all text is flattened | Text cannot be resized or recolored | Use SVG with real <text> elements, or provide a text alternative |
| Scanned PDFs published without OCR | Text is invisible to AT and non-resizable | Run OCR; tag the PDF properly |
| Social media cards generated as images | Often fine if text is duplicated in the post body | Ensure the image text is redundant, not the only source |
| Custom font headlines saved as PNG | CSS web fonts solve this without images | Load the font via @font-face or a font service |
Applying This
Automated testing tools will not reliably catch images of text. The tools can flag <img> elements missing alt attributes, but determining whether an image contains text that should be live text requires human judgment. This is one of the clearest examples of why hybrid testing approaches outperform either method alone.
In design review, the question to ask is: "Is this text, or does it look like text?" Any design comp that shows text in a non-system font, with custom spacing, or inside a graphic element should trigger a conversation about implementation method before the asset is built.
In development review, look for <img> tags with suspiciously text-like alt attributes (alt="Shop Now", alt="Our Mission", alt="2024 Annual Report"). These are signals that an image of text may have slipped through.
In QA, browser zoom is your first tool. Zoom to 200% and check whether text content scales cleanly. Images of text will pixelate or remain fixed while surrounding text reflows — a visible signal during manual review.
For PDF workflows, the fix often lives upstream in document creation. Teams that regularly publish scanned documents need a standing OCR-and-tagging process, not a one-off remediation. That's an operational investment, but it's the only path to sustainable compliance.
One practical note: SVG offers a middle path for complex graphics. An SVG file can contain real <text> elements that are selectable, scalable, and readable by assistive technology, while still rendering with the precise visual control that design teams want. For data visualizations and icon-heavy interfaces, this is often the right architectural choice.
Why This Matters Beyond Compliance
From a community lens, SC 1.4.5 violations fall hardest on users who have already done the work of configuring their environment — custom fonts, high contrast, browser zoom — only to find that a design team's shortcut has overridden those adaptations. These users have adapted to inaccessible defaults; design shortcuts that ignore those adaptations compound existing exclusion.
The operational reality is that most teams reach for images of text not out of indifference, but because CSS-based solutions for custom typography used to be harder. Today, with web fonts and modern CSS, that technical justification has largely evaporated. This criterion has a clear rule, clear exceptions, and clear techniques. The strategic case for fixing it is equally clear: the same CSS-based text implementation that satisfies 1.4.5 also improves SEO (real text is indexed; image text is not), simplifies localization (text strings are translatable; images require redesign), and reduces asset maintenance overhead. Equal access and operational efficiency point in the same direction.
About the Keisha lens
A community-impact lens. Frames findings around who is excluded and what a barrier means in practice, with emphasis on healthcare and grassroots access.
Keisha is an AI analyst lens, not a human staff member. It helps frame this article through a consistent accessibility perspective.
Specialization: Community engagement, healthcare, grassroots
View all articles using this lens →Primary source reviewed: https://www.w3.org/WAI/WCAG22/Understanding/images-of-text (opens in new window)
Transparency Disclosure
This article was drafted with AI assistance and reviewed against our editorial methodology. We disclose that process so readers can judge the work clearly.