So... what can AI bring to children’s literature?

There is good reason for all the excitement.

Generative AI dramatically lowers some of the traditional barriers associated with visual creation. A person who cannot draw can suddenly explore character designs, settings, moods, compositions and visual styles. A creator can move from an idea to a visible concept rapidly and can experiment with multiple possibilities before settling on a particular direction.

For educators, student teachers and emerging authors, this can be especially empowering.

Ratner et al. (2025), for example, found that AI-generated story experiences could encourage playful engagement, multilingual participation and shared interaction between children and caregivers. Their work suggests that generative tools can have value as supplementary creative resources, particularly when they are carefully designed and mediated by adults.

Similarly, Bai et al. (2025) found that children learning from digital picture-book videos responded comparably to AI-generated and human pedagogical agents. AI-generated appearance and voice did not significantly increase cognitive load or negatively affect reading performance. Importantly, this suggests that children do not automatically reject something simply because part of its visual presentation is AI-generated.

The technology itself, then, is not necessarily the problem.

The important question is how it is used.

The tricky bit: a beautiful image is not necessarily a good storybook illustration

Generative AI is remarkably good at producing something that looks impressive at first glance. Picture books require us to look again.

Character consistency matters

Imagine opening a story about Mia.

On page 1, Mia has curly black hair, round glasses and a purple cardigan.

On page 4, her cardigan is blue.

On page 6, her glasses disappear.

On page 9, her hairstyle changes entirely.

By page 12, Mia appears to be Mia’s distant cousin.

This remains one of the practical challenges of creating sequential illustrations with generative AI. Ho (2025) notes that AI-generated picture books can contain illustrations that are visually inconsistent from page to page, sometimes displaying markedly different styles or inadequately edited visual errors.

In traditional picture-book illustration, visual continuity helps construct a stable storyworld. Children need to recognise that this is still the same character, in the same world, continuing the same story.

That means AI-supported illustration requires deliberate planning: defining characters before generating scenes, establishing visual reference points, maintaining clothing, colours, proportions and distinctive features, and repeatedly checking new images against what has already been created.

In other words:

Consistency does not happen by accident. It has to be designed.

The picture and the words have to talk to each other

There is another form of consistency that is perhaps even more important: image-text alignment.

Ratner et al. (2025) found recurring instances in which generated illustrations failed to depict important elements of the accompanying narrative. This matters because picture-book meaning emerges from the relationship between image and language (Evans & Saint-Aubin, 2005, as cited in Ratner et al., 2025).

If the story tells us:

Benny discovers a tiny red door beneath the tree.

and the illustration contains:

a perfectly lovely tree... but no door...

we have a problem.

The image does not need to illustrate every single word literally...

Remember: good picture books often allow illustrations to extend the story rather than merely duplicate it. BUT important narrative information should not contradict what the child is reading or hearing.

This is why creators need to ask:

Does this illustration actually tell the same story as my text?

Representation: “diverse” does not automatically mean inclusive

This is where the conversation becomes particularly important. Generative AI systems learn patterns from enormous datasets. Those patterns are not culturally neutral.

Ar (2026) examines AI-generated children’s narratives featuring African protagonists and demonstrates an important tension. On the surface, the stories appear inclusive: they celebrate African hair, family, heritage and cultural pride. Yet closer examination shows recurring narrative patterns, limited engagement with historical experiences and what the author describes as forms of coded whiteness and limited cultural empathy.

The lesson extends beyond written narratives.

AI can generate an image containing a culturally diverse character without necessarily generating an image that offers culturally meaningful representation.

There is a difference between:

visibility: Is someone represented?

and

representation: How are they represented?

Creators therefore need to look beyond whether an image appears diverse. We should ask what assumptions the system has made about skin tone, hair, clothing, family, disability, gender, community, socioeconomic context, architecture, culture and place.

Ar (2026) draws on research showing that AI-generated imagery may reproduce racial biases, including the privileging the representation of certain populations groups (Yang, 2025, as cited in Ar, 2026). Ho (2025) similarly warns that generative systems may default to what their training data suggest is the “most likely” character unless creators deliberately interrogate and guide those assumptions.

A useful question when reviewing any AI-generated illustration is therefore:

What did I ask the AI to include, and what did the AI assume all by itself?

Sometimes, the most revealing part of an AI-generated picture is not what appears. It is what does not.

Then there is the creativity question...

Can AI be creative?

That question can keep us occupied for a VERY LONG time.

A more useful question for children’s literature might be:

Where should human creativity sit within an AI-supported process?

Ho (2025) raises concerns about large numbers of AI-assisted children's books being produced rapidly, sometimes with limited editing, inconsistent illustration and questionable attention to quality. Generative tools make publishing faster...but faster does not necessarily mean more thoughtful.

Ratner et al. (2025) reached a similar conclusion from a different direction. Although the AI-generated stories they examined could be produced efficiently and at scale, parents frequently described them as lacking humour, dramatic arc and narrative charm compared with familiar human-authored books.

Efficiency is therefore not the same as imagination.

And generation is not the same as authorship.

There is also a broader ethical debate around the intellectual property used to develop generative AI systems and the impact of automated illustration on professional artists. Ho (2025) highlights concerns raised by illustrators regarding AI systems trained using existing creative work and the possibility of AI being used primarily as a cheaper alternative to commissioning human artists.

These debates are still developing.

What remains important is transparency, critical awareness and respect for creative work.

The human STILL matters

Across this emerging research base, one idea appears repeatedly:

Human judgement remains essential.

AI can suggest. AI can generate. AI can surprise us. AI can produce twenty rabbits before we finish our coffee.

But the creator still needs to decide:

• Is the image accurate?

• Does it fit the story?

• Is the character still recognisable?

• Is the illustration developmentally appropriate?

• Does it contain stereotypes?

• Is the cultural representation respectful?

• Does the image add meaning?

• Is the composition suitable for the words that still need to appear on the page?

• Would I actually want a child to encounter this image?

Bai et al. (2025) ultimately position AI-generated agents as potentially complementary rather than replacements for human educators. Ratner et al. (2025) similarly argue that AI-generated storybooks currently have greater value as supplementary resources than as replacements for human-authored children's literature.

That distinction is useful for creative work too.

Perhaps the most productive way to think about AI is not as the illustrator, but as part of an iterative human-AI creative process.

You imagine. You direct.

The system generates.

You inspect. You reject. You revise. You regenerate. You edit.

You make the final decision.

The human does not disappear from the creative process. The human becomes responsible for it.

An interesting contradiction: children may accept AI images even when adults notice their limitations

Bai et al. (2025) found no significant differences in children's overall preferences between human and AI-generated pedagogical agents. In fact, 55% of the children in their study awarded the picture-book videos a five-star rating across the experimental conditions.

Yet Ratner et al. (2025) found that children were also remarkably observant. They noticed when illustrations did not match stories, while one child asked: “Did a robot write that?”

Children, then, should not be imagined as passive recipients of AI-generated media.

• They notice.

• They interpret.

• They question.

And that gives educators another opportunity: AI-generated stories and images can themselves become material for critical AI literacy.

Instead of only asking children, “What is happening in this picture?”, we might eventually also ask:

“Does this picture make sense?”

“What would you change?”

“Do the words and image agree?”

“Who might be missing from this picture?”

That moves AI from being merely a production tool to becoming something we can think about, question and critique.

So... should we use AI to illustrate children’s stories?

There is no useful answer that is simply yes or no.

Generative AI offers genuine possibilities for experimentation, accessibility, rapid visualisation, personalisation and creative exploration.

It also introduces genuine risks involving: quality, consistency, bias, representation, copyright, authorship, transparency and over-reliance on automated output.

The research therefore points towards neither technological enthusiasm nor technological panic.

It points towards thoughtful use.

The goal should not be to generate more pictures simply because we can. The goal should be to create visual stories that are meaningful, coherent, inclusive, engaging and appropriate for children.

And achieving that requires something increasingly important in an AI-rich world:

PROMPT LITERACY.

From idea to image: Why prompt literacy matters

AI cannot see the picture in your head.

It only has the information you give it.

A prompt such as:

“A rabbit in a forest.”

may produce a perfectly acceptable rabbit and a perfectly acceptable forest.

• But your rabbit?

• In your visual style?

• Showing the correct emotion?

• Standing in the correct part of the composition?

• Leaving enough space for your story text?

• Looking exactly like the rabbit from the previous page?

That requires considerably more intentional communication.

Prompt literacy involves learning how to translate a visual idea into clear instructions that a generative system can interpret. It means thinking carefully about who is in the scene, where the scene occurs, what is happening, how the moment should feel, what visual style should be used, and how the final image should be composed.

It also means recognising that good prompting is not a magic formula.

It is an iterative process:

Plan it. Prompt it. Generate it. Check it. Refine it.

And sometimes...

prompt it again.