Google Should Never Have Added the Nano Banana to Google Earth.
Last week, Google unveiled its AI-powered image generation model , Nano Banana 2, in Google Earth, its world exploration software. The company touted the integration as a way to create infographics about locations around the world or generate historical landscapes based on modern imagery. Users were also encouraged to create their “dream homes” against the backdrop of real-life landscapes—so if you were planning an addition to your home or dreaming of a lakeside cottage, you could visualize it using Nano Banana 2 in Google Earth.
This news follows Google’s trend of integrating its AI platforms into virtually all of its products and services (for example, last week the company added new Gemini capabilities to Google Docs for Workspace). From the company’s perspective, adding the Nano Banana 2 to Google Earth likely made perfect sense. However, as it turns out, Gemini wasn’t designed specifically for this product.
Google Earth has already abandoned the Nano Banana 2.
Just one day after launching the Nano Banana 2 integration, Google canceled it . An updated version of its original announcement stated, “We know people trust Google Earth as a trusted source of information about the world. We’ve seen geospatial professionals use this feature for many useful purposes; however, we’ve also seen people sharing screenshots of generated imagery that appear to violate our policies. Therefore, we’re retiring this feature in Google Earth while we work to implement stronger protections. Importantly, the generated imagery was not visible in the main version of Google Earth for other users to view and was labeled as AI-generated.”
This statement inaccurately describes the images users created with this tool. In my view, this went beyond “violating” Google’s rules: the internet, as expected, used the Nano Banana 2 to create inappropriate and insensitive images based on real locations. Perhaps unsurprisingly, one of the most popular images was of a plane crashing into the new World Trade Center in Manhattan . Other images showed the destroyed Eiffel Tower or the Egyptian pyramids collapsing into a sinkhole. While many will likely understand that this was done as a joke, many more will assume the worst upon seeing these “official” Google Earth images online.
But beyond the jokes, the potential for misinformation and falsehoods to spread was alarming. For example, one user demonstrated how he used the tool to place refugees shoulder to shoulder near the Mexican border . One can imagine such an image going viral, tricking viewers into believing Google Earth was still a legitimate satellite image. Google likely has some restrictions preventing the generation of large quantities of images, but these restrictions apparently weren’t enough. And since anyone, whether they paid for Google’s AI features or not, could use the Nano Banana 2 with Google Earth, all they had to do was open Google Earth and tinker with a malicious query until the Nano Banana 2 complied.
Generative AI has no place in a tool like Google Earth.
You don’t have to be an AI skeptic to understand why it was a bad idea from the start—but then again, few AI companies are skeptical before implementing generative AI features into their products. Tools like Google Earth are designed for learning. You use them to learn about the world, its geography, and what different cities and landscapes look like. Personally, I love spending time spinning a digital globe to learn about a new place, and I often place a pin on a random spot to see what it looks like in Street View. The reason I do this is because I believe Google Earth is an accurate and reliable source of information about the world; if it weren’t, why would I do it?
So it seems counterintuitive to me that Google is implementing a tool that necessarily generates inaccurate and false representations of the world. One of its advantages was displaying a place on a map as it looked in the past, but that’s not what Nano Banana 2 did. Instead, the AI used its training data to create what it thought was a historical representation of the area. If you asked it to show you what Paris looked like in 1726, it would likely overlay generic buildings and structures on the map rather than depicting what Paris actually looked like three hundred years ago. The result is an illusion of history; entertaining perhaps, but certainly not educational—no better than scrolling through AI-generated historical nonsense on social media.
The same applies to AI “infographics.” Companies like Google love to tout the potential of AI as an educational tool, and at first glance, Google Earth seems like a great idea for creating educational graphics explaining a location on a map. But the image you create isn’t verified by anyone, certainly not an expert: the image Nano Banana 2 produces could be full of half-truths and hallucinations, and worse, it won’t contain any references to verify the AI’s work. Regular users may remember the information they “learn” from these infographics, regardless of how accurate the data is.
At its worst, this feature was a disinformation machine. If you wanted to convince people that something bad had happened in a certain part of the world, you could simply type in your query, and the Nano Banana 2 would produce a realistic satellite image of your idea for easy sharing on social media. At a time when AI is spreading disinformation at breakneck speed, this only adds fuel to an already raging fire. Companies like Google should be finding ways to separate disinformation from real data, not facilitate its spread.
This isn’t an argument against AI tools in general, but it is an argument against implementing generative AI in an educational tool. What benefit Google believes AI brings by distorting the truth with false statements is beyond giving them another example to cite when boasting about their “successes” in AI.