UPDATED, see Nano Banana removed
Tonight I typed just one sentence into Google Earth and put refugees near the Mexican border. Then I planted a nuclear plant in Iran. Then I put a fatal crash on a street in Amsterdam. Google’s own satellite imagery underneath all three. What on earth is Google doing?

Google spent twenty years building the reference the world checks against.Today it added a button that makes things up.

Image generation went live in Google Earth on the web this afternoon, powered by Nano Banana 2, worldwide, for everyone, with no waitlist and no application.

You zoom to any coordinates on the planet, click create image, and type what you want to see. The model takes Google’s satellite, aerial and 3D imagery of that exact spot as its starting point and hands back a photorealistic picture. If it isn’t quite right, you refine it.

Google describes the feature as creating “concepts grounded in the real world,” which is the problem stated as a selling point. Grounded means the invented thing is welded to genuine coordinates and drawn on genuine imagery, often in the same colours and the same light and at the same angle as the picture beside it. A forgery built that way doesn’t have to be convincing on its own. It inherits the credibility of the map it was born on.

The announcement runs to 487 words, written by Bryan Horowitz, Product Manager, Google Earth. Horowitz gives the planet a six-word instruction: type whatever you want to see. That is the whole gate. Elsewhere in the same post he suggests it’s fun to let your imagination run wild, and the piece signs off by inviting you to pick a spot on the map and start bringing your ideas to life.

Somebody is going to bring an idea to life before morning. It will not be a community garden, trust me.

The people who want this are not hard to picture. A government official wants a strike to look bigger than it was. Two factions want the same hospital to look flattened or intact, depending on which of them is holding it this week. A troll wants forty thousand reposts before lunch. This morning all three needed a screenshot, a second browser tab and a little patience. Tonight they need a sentence.

Google responded to this article and said:

We take misinformation seriously - every image created with Nano Banana in Google Earth includes the SynthID digital watermark, so if someone is unsure about an image, they can ask the Gemini app or use Lens in Search to see if the image was Al-generated. In addition, we prevent image creation on harmful topics and are continually updating our protections.

Two claims in there are checkable, so I checked both. Refugees at a border, a nuclear plant in Iran, a fatal crash on an Amsterdam street, a hospital with a bomb crater in Gaza. Nothing was refused, nothing was softened, and nothing suggested I try a different prompt.

The other half of the answer is worth reading slowly. The watermark is Google’s, the app that reads it is Google’s, and the search that reads it is Google’s. Every road out of the problem runs back through the company that built it. I understand that Google has Google products to run Google checks, but planet Earth is bigger than Google. What happens if you take a photo of the AI generated picture and feed that to VerifyAI (SynthId)? This:

So it’s not recognized.

I puta clips on X and asked nobody to check it. Hive, one of the biggest names in AI detection outside Google, scanned the post automatically and published its verdict underneath: AI Generated Video 1%, Deepfake 0%, AI Generated Speech 0%, AI Generated Music 36%.

The clip had come out of Google’s model minutes earlier, and its audio track is silence — I measured it at minus 91 decibels, first frame to last. The only thing Hive found suspicious was music that does not exist.

Hive has a fair defence, so let me make it for them. What I posted was a screen recording, and a recording of a screen is a real recording. But that defence is the problem rather than the excuse, because fakes do not travel as clean files with their credentials intact. They travel as screen recordings, re-encodes, screenshots of screenshots, filmed off somebody’s phone in a hurry. The one case where the detector has an excuse is the only case that ever happens. The watermark was there the whole time. Outside Google, it reads as authentic.

If you have never verified anything in your life, the stakes are simpler than they sound. When a photograph turns up claiming to show a bombed hospital or a refugee camp, somebody has to decide whether it is true before a newspaper prints it. What they have is a reference: a picture of that same place, taken from above or from the street, that everybody agrees is real. They put the claim next to the reference and look at the gap.

Google built that reference. Street View passed 10 million miles of road in 2019 and now holds more than 280 billion images across over 110 countries. Google Earth turned twenty this year. Between them they are a photographic record of the physical world, and every frame of it is dated. If you know when a picture was taken, you can prove when something appeared.

In July 2014 the Russian Ministry of Defence held a press conference and produced satellite images about the downing of MH17. Bellingcat compared them against the dated archive in Google Earth, found the landscape didn’t match the dates claimed, and showed the images were fabricated. They published the walkthrough under the headline Who to Trust, Google or the Russian MoD?

In 2015 that was a rhetorical question.

Volkel

There is a village in North Brabant, ninety minutes from my desk, called Volkel, and it has an air base that stores American B61 nuclear bombs — a fact the Dutch government has never confirmed, which surfaced through a leaked diplomatic cable and was later described by a former Dutch prime minister as absolutely pointless. On Google’s maps, Volkel spent years as a rash of coloured pixels. So did the ammunition depot at Staphorst, a Patriot missile site in Limburg, and several royal palaces, because the Dutch approach to hiding a military installation was to paint a large abstract artwork on top of it. The photographer Mishka Henner collected them into a gallery and sold them as art, which tells you how well the hiding worked.

Notice what that entire twenty-year argument assumed. You only censor a map that works. Every blur, every polygon, every diplomatic request was an admission that Google Earth was accurate enough to be dangerous. Governments were never worried the map would lie. They were worried it would tell the truth.

Google’s own position never wavered: it does not blur satellite imagery by choice, and sites arrive pre-obscured because a government or a supplier demands it as a condition of flying over. So to take one real building out of Google Earth, you need a state. To put a nuclear plant into Iran, I needed a sentence.

That asymmetry is not meant to be funny. For twenty years the danger was a map that revealed something true, and there was a procedure for it — a law, a ministry, a condition on the overflight permit. Tonight the danger is a map that shows something which was never there, and for that there is no procedure at all.

The cars nobody moved

In February 2025 a Google Earth satellite pass photographed a parking lot at the US Navy’s Fifth Fleet headquarters in Manama, Bahrain — ordinary cars, ordinary rows. A year later those cars were on X, in a photograph of the same base supposedly flattened by an Iranian drone strike. Same cars, same rows, same spaces, vehicle for vehicle.

Somebody generated a war zone and forgot to move the Toyotas.

The Tehran Times posted it as a before-and-after of a US radar site in Qatar. It was not Qatar, it was Bahrain, and the “before” was a genuine Google Earth capture from 10 February 2025. AFP and BBC Verify both ran the “after” through Google’s watermark detector and got a hit, though not before it passed 950,000 views. Building it had taken six steps — screenshot Earth, open Gemini, upload, prompt, download, crop — which is six chances to get bored, and the forger got bored at the parking lot. Tonight Manama can be changed in seconds.

So can you still trust Google Earth?

Nothing was taken away today. Every satellite pass Google has ever flown is still there, still dated, still the thing that caught the Russian Ministry of Defence. The generated picture is a separate object that somebody has to deliberately ask for. All of that is true, and Google Earth remains full of genuine imagery.

It just doesn’t help the person looking at a JPEG on X at two in the morning. That person is not inside Google Earth. They are holding something that came out of it, and the thing that came out no longer says which half of the product it came from. The archive is intact; the output is not distinguishable. Only one of the two travels.

And there is a further problem Google doesn’t count. An official can now look at a genuine photograph of a genuine atrocity and say: AI. He doesn’t need the tool for that. He needs everyone to know the tool exists..

How to fight this?

Upload a suspicious satellite image to Gemini, add @verifyai. Google’s watermark is there, you have an answer in seconds. If it isn’t, you have learned one thing only: Google didn’t make it. Google has renamed this shortcut before, so if it fails, just ask in plain words whether the image was created with Google AI.

Then settle the question somewhere that cannot generate anything. Sentinel-2 is free and refreshes every five or six days, and there is Landsat, Planet, Airbus, Vantor besides. You want a sensor with an orbit, a collection time and a resolution you can check — then a second one, because as Todd Myers at the National Geospatial-Intelligence Agency put it long before any of this existed, for every collect you need a duplicate collect from different sources.

The dullest check is the strongest. If an image claims a specific satellite at a specific hour, published orbital elements will tell you whether anything was up there at all. No model can fake that, because it has nothing to do with the pixels.

Which is what broke Bahrain in the end. Not a detector. A person, counting cars.

Disclosure

I build ImageWhisperer. Google technology is part of it, not the whole of it. No single score is ever the finding.

Sources

• Google, Transform any place with Nano Banana in Google Earth, Bryan Horowitz, 30 July 2026

• Google Maps Platform, Street View Insights is now Generally Available — 280 billion images, 110+ countries, March 2026

• TechRadar, Google says its Street View imagery now covers 10 million miles

• Bellingcat, Who to Trust, Google or the Russian MoD? A Guide to Verifying Google Earth Satellite Image Dates

• Bellingcat, MH17 — The Open Source Evidence

• CNN, Google Maps: When censorship is beautiful — Mishka Henner, Dutch Landscapes

• Live Science, 15 Secretive Places You Can Now See on Google Earth — Volkel Air Base

• Google, We’re creating a new satellite imagery map to help protect Brazil’s forests

• Forbes, Google Earth Users Get Powerful New AI Image Upgrade Today, 30 July 2026

• Engadget, Nano Banana image generation comes to Google Earth, 30 July 2026

• Dan Clark, How AI fakes are turning satellite images into war misinformation, Financial Times, 4 March 2026

• AFP / CEDMO, Fake AI satellite imagery spurs US-Iran war disinformation

• DW fact check, Fake satellite images distort Middle East conflict from above

• European Commission, Transparency obligations under Article 50 of the AI Act

• TechRadar, You can now ask the Gemini app if an image was made by AI

• History & Archaeology Online, Did the Earthquake of 62AD lead to Pompeii’s Decline or Renewal?