Google Earth experimented – only to withdraw it after 24 hours – a function which, at least on paper, could open up very useful scenarios for engineers, researchers and other professionals who deal with landscapes on a daily (or almost daily) basis. With this new function it was possible to create customized geographical images starting from satellite maps, aerial photographs and 3D models of the planet. Too bad that geographic deepfakes have started to spread like wildfire, forcing the Mountain View giant to withdraw, at least for the moment, the function-based Nano Banana 2, its generative artificial intelligence model.
A function created for a noble purpose and examples of deepfakes
The initial idea was simple: allow users to virtually modify real places through AI. With just a few commands it would have been possible, for example, to visualize a dream house in a park, imagine new infrastructure or transform a modern place to show how it might have looked in the past. The function, available only from a computer, allowed you to generate images directly from the Google Earth environment and save them within your projects.
The line between creativity and the creation of geographical fakes, however, proved to be very thin and was soon overcome by users who used the function in question. This is why some researchers quickly highlighted that a similar tool could be exploited to create true geographic deepfakes, i.e. artificial images that show real places modified by artificial intelligence to appear authentic, at least at first.
Demonstrations of the risks arrived almost immediately. Some experts showed how it would be possible to add non-existent elements on top of real satellite images: a nuclear power plant in Iran, destroyed monuments (such as the Eiffel Tower and the Great Pyramid of Giza, in the cover image of this article), accidents or structures that were never built. The fear was that images of this type could be used for disinformation campaigns, that is, to spread false content with the aim of influencing opinions and behaviour.
This concern also arises from the particular role that Google Earth has assumed over the years. The service is often used by journalists and researchers to verify photographs and videos from crisis areas, checking whether a certain event actually happened in a certain place or not. If a platform considered reliable allows you to create artificial images based on its maps, it becomes much more difficult to distinguish real evidence from an invented reconstruction.
Independent investigator Henk van Ess, in his newsletter, showed how quickly it was to create deceptive scenarios using the new feature. According to his analysis, tasks that previously required several steps with separate tools could be accomplished in seconds directly within Google Earth, thanks to a typing bar and a button. Among the most striking examples presented by the investigator is the reproduction of a scene in which thousands of refugees are seen near the border between Mexico and the USA which, as you can see from the following video, was created in just a few seconds.
Examples of deepfakes created with Google Earth
If this intrigues you, we leave you with a small overview of some examples of deepfakes generated with Google Earth, so that you can understand the extent of the phenomenon even better. All the images you will see below are obviously fake.
Will AI return to Google Earth in the future?
Google initially defended the feature by explaining that the images generated included SynthID, an invisible digital watermark that allows it to recognize content created with the company’s artificial intelligence tools. According to Google, this system was supposed to help users check whether an image was generated by AI. But some researchers have questioned the effectiveness of relying on this method alone. Fake images, in fact, often do not circulate as original files: they can be shared as screenshots, screen recordings or photographs taken of a monitor, losing part of the information that allows automatic recognition.
Here Google has decided to withdraw the function and, in a post entrusted to X, explained:
We’ve seen geospatial professionals use this feature for a variety of useful purposes, but we’ve also seen people share screenshots of generated images that appear to violate our policies. (…) We are therefore restoring this functionality to Google Earth while we work on implementing stronger protective barriers.
The fact that Google has declared that it is working on «more resistant protective barriers», according to some it is an allusion to the fact that Earth’s AI functions have only been stopped temporarily and that they could become usable again in the future, when there are greater guarantees in terms of safety.
