Google has pulled the Nano Banana image generation feature out of Google Earth less than 24 hours after switching it on, following a wave of demonstrations that the tool could fabricate convincing satellite imagery of events that never happened. The feature went live on 30 July and was gone by 31 July.
What Nano Banana in Google Earth actually did
Nano Banana 2 is the Gemini image model that Google has been pushing across its consumer apps, including the Gemini app on Android. Inside Google Earth on the web, where the feature rolled out globally on 30 July, it let anyone type a prompt and have the model redraw a slice of the map from overhead — producing something that looked like a satellite pass, but wasn’t one. Within hours, users were posting generated overhead scenes — Engadget cited fabricated blast craters and protest scenes — layered onto real, identifiable places.
That is a different category of problem from a generated portrait or a fake product shot. Overhead imagery is treated as evidence. Newsrooms, open-source investigators and humanitarian monitors all lean on satellite pictures precisely because they are hard to stage, and Google Earth is one of the most widely trusted places ordinary people go to look at the planet. Google’s own statement conceded the point.
“We know that people uniquely trust Google Earth for a reliable view of the world. We’ve seen geospatial professionals using this feature for a range of useful purposes, however we’ve also seen people sharing screenshots of generated imagery that appear to violate our policies. So we’re rolling back this feature in Google Earth while we work on implementing stronger guardrails.”
The statement was posted to Google’s @NewsFromGoogle account on X.
Why the SynthID watermark did not save it
Google’s first line of defence, before the rollback, was that every image generated in Earth carried a SynthID watermark, so anyone in doubt could run it through the Gemini app or Lens in Search and be told it was AI-generated. Google also said generated images never appeared in the shared Google Earth view for other people to stumble across.
Both things can be true and still miss what went wrong. SynthID is an invisible signal embedded in the image file. The material that spread was screenshots — recompressed, cropped, reposted to platforms that strip metadata and re-encode everything on upload. A watermark that survives a clean download but not a screenshot is not a safeguard against misinformation; it is a safeguard against a version of misinformation that nobody actually practises. Google’s own wording gives this away: the company says it saw “people sharing screenshots of generated imagery,” which is exactly the distribution path the watermark was weakest against.
It also puts the burden in the wrong place. The check only works if the person seeing the image already suspects it is fake and knows to go ask Gemini. That is a high bar for a picture that arrives in a group chat looking like a news photo.
What this means for Android users
Nano Banana 2 is not confined to Google Earth. The same model sits behind image generation in the Gemini app on Android and is steadily being wired into more Google surfaces, which makes Earth a useful test of how the guardrails hold up when a generative feature is dropped into a product people treat as factual rather than creative. The answer, this week, was: not for a full day.
The practical takeaway is unglamorous but worth repeating. Treat an aerial or satellite image circulating on social media the way you would already treat a screenshot of a text message — as a claim, not proof. If it matters, check whether an established mapping or imagery provider shows the same thing at the same coordinates.
Google has not said when or whether the feature will return to Google Earth, only that it is working on stronger guardrails. The rest of Google Earth, including the Android app, is unaffected.
