It starts like this, someone texts their family group chat, slightly panicked:
“You won’t believe this. An old classmate just walked into our house. I think he’s homeless now. He’s sitting on the sofa and drinking coffee!”
They attach a photo of a scruffy man sitting comfortably in the living room, half-smiling, a mug in his hand. The curtains, the sofa, the lighting; everything looks exactly like their house. It seems real. Parents panic, call the police and within minutes, patrol cars are on their way.
But the man does not exist. He was created using a text-to-image generator as part of a new social media trend where people edit a stranger into photos of their own homes. With recently developed new AI tools, it is super easy to make a person appear realistic in almost any scene. You can adjust the pose, the light, the reflections and even the texture of their clothes.
For people who regularly work with these tools, there are small signs that give it away: hands that look slightly strange, soft textures that blur in the wrong places or lighting that doesn’t quite match the rest of the room. But parents who receive such a picture on WhatsApp do not see those details most of the time. They only see someone who shouldn’t be there and they react immediately.
The Dutch police yesterday warned about this exact prank after several emergency calls were made by parents who thought a homeless person had broken into their home. According to NOS, some even sent officers and helicopters before realising it was an AI-generated joke.
Although the trend may sound funny at first, it shows how easily generative AI blurs the line between imagination and reality. What begins as a joke can quickly create confusion and lead to unnecessary emergency responses. This also raises the question about responsibility of AI developers. How can they prevent such misuse? One possible solution would be to include automatic watermarks or origin tags in generated images. These would not be visible to the eye but could be detected by social media platforms or authorities. Another solution could be just clearer labelling on AI-edited content so that all users instantly know when an image has been altered.
Still, it is easy to understand why this trend became popular. The technology makes it effortless to create something that looks convincingly real. Tools such as DALL·E or Photoshop’s “Generative Fill” make that possible within seconds. I actually tried it myself, within five minutes I had a sheep standing in my hallway, and it looked suprisingly real.
People are out here pranking their parents with fake squatters. The real crime here is how good these generators are getting!
Wow, I don’t use social media that much so I had not come across this trend yet, but this seems concerning. I thought that the gen AI would mostly be used inappropriately for fraud. I had not considered that the kids would use AI for their pranks. This way it become a dangerous tool, I can image what kind of stuff I would do if i had AI tools as a kid 😂. So I think you are right, there should definitely be some measures to avoid people misusing it, especially for the youth.
Great post Aukje, I loved reading it! I actually got these videos a lot recently too on my fyp. And I have to admit I found it really funny too at first, especially how realistic it looked and the reactions. But when people say “I am going to call the police” that does goes a bit too far for the joke. I didn’t know that even helicopters were being sent, that hits how serious it can get, a harmless joke that wasted emergency resources. I do think you are right that there should be some measures to avoid this, especially for the people who don’t use social media that often. I think watermarks or labelling is a great idea.