Artificial intelligence models that convert text into an image can really mess up. You don’t have to look far for examples – just an alleged photo of Pope Francis in a Balenciaga jacket is enough to see that even humorous creations can move the media around the world.

To counteract these and much more serious abuses, Google has decided to introduce into the Google Images search engine special markings – labels from which users will learn about the origin of the image. The feature is part of Google’s broader initiative to combat disinformation and was announced as part of this year’s Google I/O conference.

Google Images as a deep fake detector? These markings will tell you if the image was generated by artificial intelligence

Now, each image displayed in search results will have a special “About this image” section (“About this image”, similar to the currently available “About this result” in the main search engine). We will learn from it, m.in. when the image and similar results were first indexed on Google, on which site it probably appeared for the first time, and on which pages the image was published.

According to Google, this information will allow users to understand whether “the photo is credible or needs to be looked at again.”

For example, with “About this picture”, you would be able to see that news articles indicated that this image depicting a staged moon landing was generated by artificial intelligence.

-writes Cory Dunton, product manager at Google, describing a sample image generated by AI

At the same time, Google announced that the images generated by artificial intelligence models belonging to the company will contain a special marking in the file that will allow Internet users to quickly identify whether the photo is real or is just an illustration generated by AI – even if it is published outside of Google services.

In addition, content creators will be able to add similar tags to images indexed on Google but from generative AIs that are not owned by the company.

Thanks to this, Google Images users will not only be informed about the artificial origin of the image, but also about the model in which it was generated. The company has already announced cooperation in this area with m.in. Midjourney and Shutterstock.