AI-Generated Photos Used to Deceive Partners, Experts Warn

The increasing accessibility and realism of AI image generation tools present new challenges in verifying digital content, with experts warning that these technologies could be used to fabricate evidence in personal disputes. The concern is that fabricated images can be used to create false alibis, mirroring the way they might be used in public scandals.

A recent instance highlighted how AI-generated elements within a photograph could be used to deceive. A woman became suspicious of her boyfriend after he sent her a photo of a park, claiming he was out for a walk. The image appeared to depict a couple enjoying a picnic, but the woman suspected it had been digitally altered with an AI-created couple overlaid onto a real scene.

AI experts identified several subtle indicators suggesting the picnic scene was not authentic. One of the primary signs was a noticeable difference in the quality and rendering of the couple compared to the rest of the photograph. The couple appeared softer and less detailed, a common artifact when AI technology merges visual elements of varying quality. "The quality of the couple is different to the rest of the picture," AI expert Andrew Hulbert told the Daily Mail. "It's smeared and softer." This effect occurs because AI often lifts images from other sources to integrate into a current picture, and these sources can have inconsistent resolutions.

Another clue was the lack of sharp, clear details within the picnic scene. For example, the woman's hair strands appeared to blend unnaturally into each other, lacking the crisp definition expected from a modern smartphone camera. Similarly, hands and objects on the picnic blanket lacked clear definition, and some picnic items had unusual shapes. Parts of the woman's dress, blanket, and grass seemed to merge, according to Emily Hatton, founder of AI in Agencies. "Around the picnic couple, the hands and the object they're holding lack clear definition, some of the picnic items have unusual shapes, and parts of the woman's dress, blanket and grass seem to merge," Hatton said.

The composition of the scene also raised suspicions, appearing more like a staged advertising shoot than a candid moment. "It all looks like a magazine photo shoot but low detail. It's also unlikely to catch that perfect pose. It's too staged," Hulbert noted.

The most significant indicator was an inconsistency in focus. The couple, though supposedly within the park scene, did not appear to match the photographic characteristics and focus of their surroundings. This discrepancy is often a result of the AI's stitching process when creating composite images.

However, experts caution that individual visual irregularities alone are not definitive proof of AI manipulation, as image compression, resizing, and repeated transmissions can also introduce visual artifacts. "These can be signs that people have been added using AI, although the low image quality means I couldn't confirm it definitively," Hatton stated.

Sunil Manjunath, co-founder of Techoor, emphasized the broader implications: "The same tools that let someone create a fake vacation photo can just as easily fake an alibi, like a photo suggesting someone was somewhere they weren't." He added, "As these tools get more realistic and accessible, I believe we'll see this show up more in everyday relationships, not just public scandals." The challenge, Manjunath suggested, is shifting from asking "can I see a photo?" to "can I confirm where this photo actually came from?" a verification process many people are not yet equipped to handle.

The woman in the case initially shared the image on Reddit's 'Is This AI?' forum, seeking opinions on its authenticity. While she could not pinpoint the exact issue, the unnatural appearance of the people in the photo made her question its genuineness. Her suspicions were eventually confirmed through other means beyond the photograph itself.