InDesign has recently introduced a new feature: generative AI can now be used within the software to automatically create alternative descriptions of images (also called alt-text).

Alt-texts are short textual descriptions that explain the visual content of an image to people with visual disabilities and are among the minimum requirements of the accessibility guidelines (WCAG – SC 1.1.1 Non-text Content).

The new tool is part of the Generative AI ecosystem—Adobe’s service already available in other Creative Cloud software, which is accessed by users through the generative credits included in their subscription plan.

We tested this feature to understand its limitations and potential.

Limitations identified during pre-testing

At present, the feature has some significant limitations:

• it is only available in the English-language version of the software, so all descriptions are generated in English, regardless of the language of the text;
• it is active by default. This means an alternative description is generated for every image, even when not necessary, and thus it automatically reduces the generative credits included in the subscription plan—without them being intentionally “spent” by the user.

Further limitations

Transparency of the alt-text generation mechanism

In its Content Analysis FAQ, Adobe specifies that content entered into the software is not used to train AI models. However, the underlying mechanism for generating alt-texts remains opaque. Are images processed locally or sent to external servers? Under what terms and conditions? All these questions must be asked before adopting this feature for those working with confidential or rights-protected images.

Issues for the software user

When an image without an alt-text is imported into InDesign, the AI automatically generates an alternative description. This is signalled by the appearance of an icon  in the lower margin of the frame.

The phrase “AI generated content” is added to every auto-generated description. This correctly indicates how the content was created, but in documents with many images it can become redundant for people who read with assistive technologies. It should be noted that this addition can be disabled from the Edit > Preferences > Generative AI panel.

If an image already has a manually written alt-text, the icon does not initially appear when the image is imported. However, opening the Text Object Exportation panel to review and/or edit the alt-text causes the icon to appear even for that image. This happens because both auto-generated and manually written alt-texts fall under the Custom option in the Alt-text Source menu.

The lack of a distinction can be a problem for those working in InDesign. A quick visual indicator to tell apart images with AI-generated alt-texts from those that have human-written ones would make the workflow considerably more practical.

Testing and evaluating the quality of alt-texts generated by InDesign’s AI

Over time, image recognition algorithms have been particularly optimised for photographs. This type of image is very common in certain kinds of books, but it is not representative of the complexity and diversity of images present in other publishing genres and sectors—think of school textbooks, university manuals, comics, and so on. A book can contain decorative images, geographical maps, explanatory models, charts, comic strips, and much more.

To test this feature, we selected three different types of images: a photograph of an animal, a histogram, and a diagram.

The alternative text contains two errors:

• the animal is incorrectly identified—it is an iguana, not a lizard;
• the animal’s position is incorrect—it is not looking at the camera, but straight ahead.

Here, the alternative text reproduces the chart title but does not specify the chart type. More importantly, it does not describe the data at all.

In this case too, although this is a fairly simple diagram, the alt-text covers only some components while overlooking entire phases of the workflow.

As mentioned above, the automatic generation of alternative descriptions is active by default for users whose plan includes AI implementation. To disable it, click on Edit > Preferences > Generative AI and deselect the first option in the panel.

LIA’s research on images

Fondazione LIA has extensive, long-standing expertise in this area. It has been conducting research on the subject since 2019—before many generative AI tools emerged—starting with the pilot project “Automating the alternative descriptions of images>”.

This work continued with “ALT-GPT”, a project carried out as part of the PhD Program of National Interest in Cultural Heritage, launched in 2023 with Università di Roma Tor Vergata. The project analyses the potential of multimodal generative AI in developing scalable methodologies for alt-text production.

Within this R&D path, Caterina Morelli presented a contribution at the 18^th international conference of the Association for the Advancement of Assistive Technology in Europe (AAATE) 2025. The paper was published in the conference proceedings by Springer under the title: Enhancing Accessibility in Publishing: Leveraging GAI for Effective Alt-Text Solutions. It explores how GenAI can support alt-text production in publishing, enabling more efficient processes and promising lower production costs—while maintaining a firm human-in-the-loop approach to ensure quality and compliance with accessibility requirements.

Conclusions

Creating accurate alternative descriptions requires specific accessibility expertise and the ability to contextualise images within a given publication. In its current state, AI cannot assess whether visual information is already present in the surrounding text, nor determine what additional information a visually impaired or blind reader may need.

The new automatic alt-text generation feature in InDesign is an interesting and promising innovation. At present, however, we do not consider it ready for use in accessible digital publishing, where precision and contextualisation remain non-negotiable.

The examples above show that this tool does not yet produce alternative descriptions of sufficient quality for the complexity of images found in editorial publications. For now, it is essential that the workflow includes human review of auto-generated descriptions for simple images, and review or rewriting of alt-texts by competent editorial professionals and/or subject-matter experts for complex ones.

As part of its quality assurance and e-book certification activities, Fondazione LIA also provides support in reviewing alt-texts—so that images can speak even to those who cannot read them through the eyes—and in integrating accessibility into workflows, towards “Born Accessible” publications.

If you are interested in finding out which service best suits your organisation’s needs, send an email to segreteria@fondazionelia.org.