Meta labels real photos as AI-generated – and vice versa

On Facebook, Instagram and threads, photos have been marked as AI-generated – but they are real. Many AI images lack a reference.

Save to Pocket listen Print view
Futuristic, distorted depiction of the face of a beautiful woman with blue eyes, biting her lower lip.

Definitely not AI-generated. Or is it?

(Image: sakkmesterke/Shutterstock.com)

3 min. read
This article was originally published in German and has been automatically translated.

At the beginning of the year, Meta announced that it would label photos with an AI label if they were AI-generated or edited using AI. Apparently, however, there are numerous false positives, i.e. false matches or assessments. Of course, an AI also decides, among other things, which AI hints are given. However, it does not recognize AI-generated images particularly well.

Photographers and people who upload photos to Facebook, Instagram and Threads are complaining that their pictures have been marked as "Made with AI", even though they are supposed to be normal shots. Among those affected is Pete Souza, former photographer of the White House in Washington, on Instagram. He told TechCrunch magazine that Instagram practically forced him to use the AI label for his images, even though they were not even AI-edited. Souza suspects a change at Adobe when saving images as JPEG as the reason for the incorrect assumption. After all, Meta has apparently removed the incorrectly stored label - if you search for Souza's images, it no longer appears.

Meta's intentions are, of course, only good. Photorealistic images in particular are to be given the AI label so that people cannot be fooled. At the launch, Nick Clegg, Meta Manager, wrote: "Our users have told us that they appreciate transparency around this new technology." The "Made with AI" label should initially be added to all images that already contain this information in the metadata. Meta has joined the Coalition for Content Provenance and Authenticity (C2PA) and uses its standard as well as the International Press Telecommunications Council (IPTC) standard.

But there are not only false references to AI in real photos. Obviously, AI-generated and photorealistic-looking images still appear on the platforms without any indication of how they were created. There are entire pages and groups that consist almost exclusively of AI-generated images. Fraudsters, for example, take advantage of the attention generated by images that are as eye-catching as possible: Shrimp Jesus and the like are classic scam attempts.

However, there are also sites that are simply looking for interactions and advertising revenue. For example, there are numerous groups and photos of tiny houses. The only thing is that, on closer inspection, you can play a game with friends: Who can find more bugs. Strange staircases, tiles of different sizes in the backsplash, dishwashers under or in the sink, windows that you don't know whether they want to be windows or lamps. Photorealistic, but without an AI hint.

Supposedly real photo of a tiny house.

(Image: mst)

(emw)