Google eliminated the AI image-generation characteristic from Google Earth on July 31, 2026, a couple of day after switching it on for each consumer of the browser model. The corporate mentioned it’s “rolling again this characteristic in Google Earth whereas we work on implementing stronger guardrails.”
The characteristic had shipped on July 30, 2026. Bryan Horowitz, a product supervisor on Google Earth, described it within the launch publish as a “create picture” button in Earth on the internet: zoom to any location, kind a immediate, and Nano Banana 2 returns a photorealistic picture constructed from Google’s satellite tv for pc, aerial and 3D imagery of that precise spot. Availability was world on day one, with no waitlist, no software and no enterprise-only tier. It was the identical distribution reflex that has carried the Gemini app to 950 million customers.
The publish pitched 5 makes use of: a classroom reconstruction of Pompeii, an infographic a couple of landmark, a real-estate idea for an empty Tokyo lot, a cabin rendered into an actual lakefront, and a science-fiction remake of Google’s personal Mountain View campus. Its instruction to customers was to kind no matter they needed to see.
Why the rollback landed in a day
Google’s assertion places the set off in distribution quite than within the mannequin. The corporate mentioned it had seen geospatial professionals utilizing the characteristic for a variety of helpful functions, and had additionally seen individuals sharing screenshots of generated imagery that appeared to violate its insurance policies. It added two design information: the generated photos by no means appeared in the principle Google Earth expertise for different customers to see, they usually have been watermarked as AI-generated. Each of these controls govern what occurs inside Earth. The screenshot is the trail the photographs truly took out of it.
Henk van Ess, who publishes the Digital Digging e-newsletter, posted a walkthrough hours after the launch displaying what he had generated on high of Google’s personal imagery: refugees close to the Mexican border, a nuclear plant in Iran, a deadly crash on an Amsterdam avenue, a bomb crater beside a hospital in Gaza. He mentioned none of his prompts have been refused. NPR ran its personal assessments and produced Iran’s Kharg Island on fireplace and a flooded US Capitol advanced.
The asset the characteristic sat on high of
Google Earth’s institutional worth is that it’s the factor different photos get checked in opposition to. Its archive is dated, its protection is broad, and newsrooms, human-rights researchers and open-source investigators deal with it because the reference layer when {a photograph} seems claiming to point out a strike, a camp or a flood. That’s the reason the rollback assertion opens on belief quite than on functionality. The product being defended is the reference, not the generator.
Jake Godin, a senior researcher at Bellingcat, advised NPR the change was certainly one of velocity quite than of variety. “It’s simply streamlining the method, which I believe goes to make it proliferate extra,” he mentioned. Fabricated satellite tv for pc imagery circulated effectively earlier than this launch, together with an AI-altered picture of a US Navy base in Bahrain that ran in Iranian media earlier in 2026. Producing one meant screenshotting Earth, shifting the file right into a separate generator, prompting it, then cropping the end result. The Earth integration compressed that sequence right into a single immediate issued contained in the reference device.
What needs to be rebuilt earlier than it returns
Google has tied the characteristic’s return to stronger guardrails. The enforcement stack it has described has two layers: SynthID, the invisible marker that identifies a picture as AI-generated to Google’s personal detection instruments, and refusal on the immediate for dangerous subjects. The primary layer is a provenance sign quite than a block. It survives a screenshot, and it studies one thing solely when the individual holding the file runs a test.
That hole is now a compliance query in addition to a product one. Google is a signatory to the EU code on labeling AI-generated content material, and labels on artificial media are changing into necessary underneath EU regulation. A characteristic that mints photorealistic imagery welded to actual coordinates is near the middle of what these obligations are written for, and a marker that solely Google’s instruments can learn is a skinny reply to them.
For the geospatial clients Google named in its personal assertion, the reversal restores the established order: Earth on the internet is again to displaying captured imagery, and the create-image button is gone from it. Google’s dedication is to return picture technology as soon as the guardrails maintain, and the constraint it now has to design in opposition to is the one the screenshots demonstrated. No matter Earth generates has to remain identifiable after it leaves Earth.
