Google Earth launches Nano Banana image generation for realistic place redesign and historical reconstruction
Google ● Covered by 7 sources
Google Earth let you AI-generate photorealistic scenes on real map locations. It pulled the feature a day later after people started sharing rule-breaking fakes.
Based on reporting by Google — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
For about 24 hours, Google Earth let anyone with a browser rewrite reality. The feature, built on Nano Banana 2, let users zoom into any spot on the map, hit "create image," and type a prompt — turning a parking lot into a lakefront cabin, or the ruins of Pompeii into a bustling 78 A.D. street scene. Google's pitch was practical: architects mocking up real estate projects, teachers showing history classes what a site looked like centuries ago, homeowners previewing a backyard studio before pouring concrete.
The examples Google shipped alongside the launch were genuinely clever. Type "reimagine this empty lot in Tokyo as a vibrant shopping district" and Nano Banana would render a 3D urban scene sitting right on top of the actual terrain. Ask for an infographic about the Statue of Liberty and Gemini would pull historical facts while Nano Banana built the graphic in seconds. There was even a version for pure fantasy — turning the Mountain View campus into a glowing sci-fi utopia with flying pods and glass biodomes.
But Google Earth isn't a random image generator. It's the tool millions of people treat as ground truth for what a place actually looks like, and that reputation turned out to be the problem. Within a day of launch, Google noticed screenshots circulating that appeared to break its content policies — despite the generated images carrying an AI watermark and never appearing in the shared, public version of Google Earth that other users see.
So Google rolled the feature back on July 31st, just one day after announcing it, saying it needs stronger guardrails before trying again. No firm timeline for a relaunch, no detail on what exactly the offending images showed. Just a quiet admission that pairing satellite-grade realism with generative AI created a misuse problem faster than the safety net could catch it.
My take — AI-written commentary, not fact-checked reporting
This was entirely predictable, and I'd bet the guardrails discussion happened after the fact rather than before. Slapping a watermark on an image does nothing once someone crops a screenshot — that's not a safety feature, it's a fig leaf. Google Earth's whole value proposition is that people believe what they see there, so bolting a fantasy generator onto it without serious friction was always going to blow up fast; it just blew up faster than usual.
Read more about this at: Google