Content creators are making AI-generated images of historical events and figures. Do they belie the past?

The line separating historical fact from fiction is blurring fast as creators are making images of the pre-modern past in mere seconds, by using text-to-image AI models. Generative AI chatbots have made it possible to turn historical paintings of Indian rulers and warriors into lifelike portraits. Other AI images illustrate atrocities against enslaved subjects, women, and common warriors who may have been left out of historical records.

But unlike the lavishly illustrated miniature paintings or carefully carved temple sculptures, AI-generated images are flawed due to poorly rendered settings and historical errors. While some mistakes may be obvious to viewers, others are only noticeable to professional historians.

Screengrab from an Instagram Reel showing AI-generated images of Indian history | Photo Credit: @history_unfolded_5 on Instagram

Some scholars argue that the gaps in historical archives should be left as they are, in order to keep the space open for further research or to show how valuable sources of evidence have been destroyed, lost, or kept out of public reach. Despite this, content creators who build their repository of historical events in India, are choosing to make AI-generated images.

This crop of media tends to confirm the creator’s personal biases and emotional interpretations of their historical subjects. AI models allow a person to create a flattering photo of a strong and victorious ruler. It also allows a person who detests the same ruler to generate a pitiful image of a man who has lost a battle.

Whose image carries more weight? Social media algorithms and followers will decide. Some content filters in AI models stop users from creating particularly gory or explicit historical images, but users with experience can easily bypass such restrictions.

Meanwhile, some educators are generating AI images for more academic reasons.

Ruchika Sharma, a historian of medieval India, makes YouTube videos to educate her followers about centuries-old topics that cause controversy even today, like religious relations in the Mughal empire and the destruction of Hindu temples.

Sharma explained that she enters prompts into Adobe Firefly to generate images for her narration-driven videos. Her reasons for using AI-generated images are diverse: her dislike of editing videos; a wish to save time; the need to engage viewers; and the difficulty in getting permission to use licensed media.

Ruchika Sharma uses Adobe Firefly to create AI images of Indian history | Photo Credit: Ruchika Sharma on X

“The idea is just to sort of try and recreate a scene from the past and I think AI can do it very well,” Sharma said.

“I think if you are teaching history to a group of people who don’t have a background in history, who find history boring, illustrations really go a long way in catching your attention, really, because the visual medium has an impact,” she explained, adding that pictures of other humans carry a lot of emotions and stay with learners longer.

Sharma said she uses such pictures as representational images, similar to non-AI illustrations of historical figures that are already available online. For example, Sharma pointed out that even non-AI images of Chanakya online have the ability to shape how users imagine his appearance and could perpetuate stereotypes, which she wanted to avoid.

By contrast, Sharma felt that generating historical AI images herself helped her depict rulers in a more “neutral” and “matter-of-fact” way when compared to some illustrations of historical figures already published online.

“For example, talking about Alauddin Khalji, the kind of pictures that you have online are just so menacing. So I don’t want to do that [and use those images], I don’t want to do that for any kind of king. I think of temple destruction as just a particular aspect in history. And I don’t think any ruler who has sort of broken temples is in any way menacing. So I think AI does that: it does not create this menacing ruler, you know, with this grin on his face and doing something very evil,” she explained.

When The Hindu looked up images of Alauddin Khalji on Google Chrome, we spotted a mixed set of neutral sketches and paintings depicting the ruler who died in the 14th century, as well as photos of actor Ranveer Singh who portrayed Khalji in the 2018 film Padmaavat. Upon scrolling further down, a few non-historical illustrations showed Khalji engaging in intimate acts with a man.

Sharma said that YouTube did not ask her to flag any AI content in her videos, and that she trusted her viewers to know that an AI-generated image was not a primary source or a real photo of a medieval event.

However, Sharma clarified she would not use AI-generated images when teaching her students, because she could use licensed images such as historical paintings in her classroom presentations. She also observed that AI-generated images would not likely be welcomed at conferences with other academics.

Giving AI history a go

To a prompt: “Generate a historically accurate picture of the Mughal emperor Akbar”, Google’s Gemini chatbot churned out a painting-style image of a court scene with a thickly-bearded man seated on a throne, surrounded by other men and women.

While the picture looks passable at first glance, a second look reveals warped fingers, disproportionately sized people, and distorted objects. These mistakes are obvious to most viewers.

Google’s Gemini generated an image of the Mughal emperor Akbar that was full of historical errors | Photo Credit: Gemini

However, those who have read about Akbar or viewed Mughal paintings of him will notice contextual errors in the AI-generated image. These include inaccurate architectural features, incorrect gender dynamics for the time period, and even features in Akbar’s personal appearance and style that are completely at odds with the historical records describing him.

For example, Akbar’s son Jahangir noted in his memoirs that his father had a mole on the left side of his nose. That feature is missing in the AI-generated image.

Despite errors such as these and the risk of confusing viewers, some creators of historical content believe an AI-generated image can be better than no image.

Medieval Indian History is an educational account on Instagram that discusses periods of Indian history such as the Delhi Sultanate and the Mughal empire, sharing informative posts with its more than 62,000 followers.

The Instagram account sometimes used AI-generated photos, but its team told The Hindu that such images were only used when subject matter was “scanty or absent.”

Some content creators only use AI images of history as a last resort | Photo Credit: @medievalindianhistory on Instagram

The team recently posted an AI-generated photo of Sultan Balban’s son Muhammad, who lived during the 13th century.

“In the case of the content on Sultan Balban’s son Muhammad, we have not found contemporary paintings, sketches or appropriate material to attach with the concerned content. While labelling, we try to link the image to one or more historical event(s), personalities et al that are a part of the content along which it is attached,” explained Team Medieval India in a statement to The Hindu.

The team stressed that it only used AI-generated content if no other contemporary paintings, modern day sketches, images, clips from documentaries, or films on the subject were available for use.

When past and future collide

Historian Ira Mukhoty, whose past works have examined Mughal royalty and their legacies today, said that new technologies can reflect racist and gender-normative biases even in the hands of well-intentioned users.

“My concern in India would therefore be with the intentions of the creators of these images - is this being done for purely educational reasons, or are there hidden motives behind the project[?]” wondered the historian, who authored Akbar: The Great Mughal—a comprehensive biography of the Mughal emperor who lived between 1542 and 1605.

Mukhoty noted that without photography during the time period, images of the Mughals would be based on contemporary written records, which did not focus greatly on the emperors’ looks. Another source, Mughal miniature painting, was not always realistic either, she pointed out.

“Instead, AI is likely to use negative and caricatural modern stereotypes, which can become very problematic if this is done with the intention of manipulating peoples’ sentiments,” Mukhoty warned.

“Images and pictures can generate an affective intimacy with the past - an experience that is founded on emotions, rather than purely rational knowledge. Humans are visual creatures and images can provoke an immediate feeling of grief, love, hate or anger very quickly, and thereby project a distorted idea of history,” she said.