Images from ancient Pompeii. Animated scenes from World War II. Surveillance footage of the Chernobyl nuclear disaster. Artificial intelligence is increasingly being used to create these and countless other fabricated images and videos depicting the past.  

"It seems that AI can be used to reconstruct images from the past, and to create synthetic images that never existed but might have," media researcher Roland Meyer told DW. "In this way, it can appear to fill supposed or actual ‘gaps’ in the archives and historical record. That strikes me as a deceptive promise."  

DW Fact check explores why these videos are taking off — and why using AI to imagine the past can be dangerous. 

Telltale signs of AI-generated content 

AI-generated content is becoming easier to identify, thanks to watermarks, labels on social media platforms and Community Notes on X. 

At the same time, AI tools are constantly evolving. AI-generated videos are more and more convincing, and watermarks indicating the use of AI are often removed. This makes critical image analysis as important as ever. 

This TikTok video depicting Pompeii before, during and after the eruption of Mount Vesuvius is labelled as AI-generated on TikTok. When Mount Vesuvius erupted in AD 79, the Roman city was buried under volcanic ash and debris. 

The video displays several telltale signs of AI-generated content: The people move unnaturally, and the video’s female protagonist is depicted in more modern dress, unlike the look typically associated with women in ancient Rome. 

Big moments, striking images 

A quick look online reveals just how widespread the trend has become. 

This video on X, which supposedly shows surveillance footage of the 1986 Chernobyl nuclear disaster, was also generated using AI. Although it does not contain a watermark, it was identified as AI-generated content in the notes attached to the post, known as Community Notes. 

Another example: This YouTube video was produced using Veo, a generative AI tool. The camera moves in toward a group of soldiers holding the US flag and shows their faces as they look into the distance. 

The video is based on the iconic photograph of the flag-raising on the Japanese island of Iwo Jima. Taken by Joe Rosenthal on February 23, 1945, the image became a powerful symbol of the US military's victory over Japan and one of the most famous photographs of World War II. 

The authentic photograph of Iwo Jima can be found in the US National Archives. However, no authentic video footage of that moment exists. 

The YouTube account that published the AI-generated video promises to present "patriotic stories" as well as "educational and entertaining content that brings America's story to life." 

Clichés and stereotypes 

AI models draw on both authentic sources and staged representations when generating such images. The datasets used to train these models include historical photographs as well as digitized paintings, film scenes and images from video games. 

Meyer notes: "The images of the past generated by AI mainly reproduce the familiar patterns, clichés and stereotypes through which that past has been portrayed so far. Above all, they are images of our expectations: Not a window into the past, but a mirror of our present." 

Such AI creations also raise questions about responsibility. Unlike the use of reenacted scenes in documentaries, where directors or producers consciously decide how the past should be portrayed, AI draws on vast datasets. 

And it does so without any understanding of historical context. "It produces something that appears plausible at first glance, but not because it is historically accurate. Rather, it looks convincing because it matches our expectations," Meyer explains.

AI depictions created in collaboration with experts 

The use of AI to create "historical" images is therefore controversial. However, there are also examples of AI-generated depictions being developed in collaboration with experts. 

In Pompeii, Italy, for example, archaeologists have used artificial intelligence for the first time to reconstruct the appearance of a victim of the AD 79 eruption of Mount Vesuvius. 

The depiction is based on archaeological findings and is intended to make research results more accessible and easier to understand. It is presented as a scientific reconstruction, however, rather than an authentic depiction. 

When AI-generated history competes with real sources 

The video is based on a real historical event but adds details and perspectives that are not documented in the historical record, presenting them within a patriotic narrative. This is why it is important to take a close look at both the images themselves and their sources. Where does the image come from? Who is presenting it, for whom, and why? 

Many viral AI-generated "historical" videos that claim to depict the past occupy a space somewhere between fact and entertainment. 

While such videos can spark interest in historical events, experts warn that they can also give the past a sense of certainty that often did not exist. Historian Jürgen Zimmerer puts it this way: "Every time a particular scene is visualized, it excludes other possible interpretations that could equally exist." 

It is therefore important for viewers to recognize when they are watching an AI-generated "historical" video and to approach its depiction with the necessary critical distance. 

Media researcher Meyer also warns: "When it comes to images of ancient Rome, we know they are fictional. But with images from the Nazi era, there is a risk that large volumes of AI-generated fakes could displace authentic documents, not least because AI-generated images are tailored to our expectations and viewing habits." 

Originally published on July 19, this article was updated on August 19 to include additional insights from experts.