Accessibility

Image description: the invisible work that makes an accessible book

· Fernando Tavares

An illustration of image description in accessible books

Summary: Describing images in accessible books takes a qualified professional. AI can help, but human review is what keeps the editorial quality.

When Booknando started working on accessibility in digital books, almost a decade ago, image description was treated as a technical detail, little more than a checklist item. Today we know it is one of the most important elements in making a work genuinely inclusive. More than that: done well, it improves the experience for every reader, not only those who are blind or have low vision.

Audio description, the technical term for putting visual elements into words, is regulated in Brazil by ABNT NBR 16452. The standard sets criteria for quality in production, but what many people do not realise is that those criteria apply to any context where visual information has to be conveyed non-visually. In books that ranges from charts and diagrams to illustrations and photographs.

According to IBGE, Brazil's national statistics institute, 18.6 million people in Brazil have some form of disability, which is 8.9% of the population. Of those, roughly 6.5 million have a serious visual impairment, whether blindness or low vision. They are readers who depend on proper descriptions to reach the content of a book.

What makes a description adequate?

The PNLD, Brazil's national textbook programme, now includes specific digital accessibility requirements in its most recent calls for tender. The technical specification for early years education, for example, directs that image descriptions use the HTML "alt" attribute or techniques described by the DIAGRAM Center, the international initiative that sets standards for accessible digital books. That means publishers taking part in the programme have to deliver works with proper descriptions, or risk being disqualified.

But what makes an image description "adequate"? The answer depends on several things. The audience comes first. A children's book calls for a completely different approach from a technical or academic one. In work for children the description has to capture not only what is in the image but the tone, the emotion, the atmosphere of the scene. It is not enough to say "a boy sitting in a chair". You need to convey "a boy in a red t-shirt sitting in a rocking chair, looking out of the window with a curious expression while the afternoon light comes in and casts shadows on the wooden floor". Editorial judgement makes all the difference.

In technical or educational books, precision matters most. Charts need their data described clearly. Diagrams need the relationship between elements explained. Scientific photographs require specific, correct vocabulary. The W3C, the consortium that sets web standards, offers an alt text decision tree that helps determine what should be described for each kind of image. The DAISY Consortium, the world reference in publishing accessibility, maintains a detailed guide to descriptions in EPUB.

Description is not a trivial task

This is where a crucial point comes in: describing images is not a trivial task. It requires knowledge of the content, an understanding of the audience, command of language and, above all, the judgement to tell which information matters and which is redundant. A description that is too generic does not do its job. One that is too long can tire the reader and get in the way.

In recent years artificial intelligence tools have begun to appear as possible allies. Platforms such as the Brazilian Audiover propose using AI to generate automatic descriptions of images and video. Academic research, such as the work published at the Brazilian Computing Society symposium, shows the technology has potential for static images, particularly in scientific contexts. The journal Ciência da Informação Express published a study showing that AI can make scientific images more accessible, with one fundamental caveat: human review remains indispensable.

The risk of automation without oversight

This is the point Booknando wants to draw attention to. AI tools can speed the process up and help produce a first draft, but they do not replace a professional who understands the subject. A description generated automatically, without review by someone who understands the book's context, its audience and the accessibility criteria, risks being technically correct but editorially wrong. Worse: it can be misleading, or simply useless.

There are cases where AI describes a children's illustration as "an abstract image with coloured shapes", when in fact that image is a key moment in the story, with characters and emotions that need to come across. In technical books we have seen tools describe a bar chart as "a chart with blue and red bars", without mentioning what the data means, what trend it shows, or what conclusion the reader is meant to draw from it.

So the recommendation is clear: if there is no qualified professional to assess and refine the description, it is better to have no description at all. A bad description can exclude more than none, because it creates a false sense of accessibility while failing to deliver the information a blind reader actually needs.

Training and editorial judgement

The way forward, then, is to invest in training. People working in publishing production need to know the basic criteria for image description, which are not the same criteria we use for writing captions or alt text for the web. They depend a great deal on the kind of book, the content, the purpose of the work and, above all, the audience. A good describer needs to understand language, narrative, editorial design and, of course, accessibility.

Booknando has worked with publishers across Brazil to put in place production flows that include image description from the start of the process, rather than as a final, almost corrective step. When accessibility is considered from the design stage onwards, the result is better for everyone: blind readers get full access to the content, readers with intellectual disabilities benefit from clear, direct descriptions, and even search engine indexing improves, because the descriptions work as metadata that helps identify what the images contain.

Publishers who want to go deeper can consult the DIAGRAM Center guides, the ABNT standards and the DAISY Consortium materials. And of course they can count on Booknando to work out how to turn that requirement into real editorial practice, with quality and respect for the reader.

Accessibility, after all, is not a favour. It is a right. And every well-written description is a step towards more people reaching the knowledge that books carry.

References

  • ABNT NBR 16452:2016. Acessibilidade na comunicação: audiodescrição. Associação Brasileira de Normas Técnicas, 2016. Available at: ABNT Coleção
  • ABNT NBR 17225:2025. Acessibilidade digital para Web. Associação Brasileira de Normas Técnicas, 2025. Available at: ABNT Coleção
  • IBGE. Pessoas com deficiência têm menor acesso à educação, ao trabalho e à renda. Agência IBGE Notícias, 2023. Available at: agenciadenoticias.ibge.gov.br
  • FNDE. Edital PNLD Educação Infantil: especificações técnicas e acessibilidade. Fundo Nacional de Desenvolvimento da Educação, 2024. Available at: gov.br/fnde
  • Booknando. Designing inclusive futures: why the PNLD programme matters. 2024. Available at: booknando.com.br/en/case-studies/pnld-digital
  • DAISY Consortium. Image Descriptions: Accessible Publishing Knowledge Base. Available at: kb.daisy.org
  • W3C. Write helpful Alt Text to describe images. Harvard University Accessibility, 2024. Available at: accessibility.huit.harvard.edu
  • Accessible Publishing. A Guide to Image Descriptions. Available at: accessiblepublishing.ca
  • DIAGRAM Center. Image Guidelines for EPUB 3. Available at: diagramcenter.org
  • Audiover. An intelligent platform for describing images and video. Available at: audiover.com.br
  • SBC. Inteligência artificial para audiodescrição de imagens. Simpósio Brasileiro de Computação, 2024. Available at: sol.sbc.org.br
  • Ciência da Informação Express. Inteligência artificial pode tornar imagens científicas mais acessíveis, mas revisão humana continua essencial. 2024. Available at: cienciadainformacaoexpress.com
  • Bookwire. PNLD e livros acessíveis a todos. Bookwire Brasil blog, 2024. Available at: br.bookwire.net
  • MokLabs. EPUB Acessível: guia completo para criar e-books inclusivos. 2024. Available at: moklabs.com.br
  • Booknando on Medium. O ePub3 que o PNLD quer. 2024. Available at: medium.com/booknando
Fernando Tavares

Fernando Tavares

Founder and CEO of Booknando. AI Coordinator at Faculdade LabPub. PublishNews columnist. More than 20 years in digital publishing.

Related articles