Last updated: August 2, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Google kills Earth AI generator after one day, pulling the "Create image" button from Google Earth's web app on July 31, 2026, less than 24 hours after it shipped. The company yanked it after journalists and researchers showed the Nano Banana 2-powered tool could generate convincing fake satellite imagery of real, sensitive locations with almost no pushback.
Short answer: Google pulled its Google Earth AI image generator on July 31, 2026, about 24 hours after launch, after reporters and researchers showed the Nano Banana 2-powered tool would create photorealistic fake imagery — a nuclear plant in Iran, refugees at the Mexican border, flooding at the U.S. Capitol — over real satellite maps with almost no refusals, and its SynthID watermark didn't survive a simple screenshot.
I cover AI product launches for a living, and the ones that get pulled inside a day are rarer, and more informative, than the ones that stick around. This is one of the fastest reversals I've tracked this year, and the reason it broke matters more than the reversal itself: it's not that the tool made bad-looking images, it's that it made a trusted mapping brand's imagery too easy to fake convincingly.
What actually happened
Google added a "Create image" button to Google Earth's browser version on Thursday, July 30, 2026. Google product manager Bryan Horowitz described the pitch simply: "Just zoom in to a place in Google Earth on web, tap 'create image,' and type whatever you want to see." The button ran on Nano Banana 2, Google's image model, and built its output from the same satellite, aerial, and 3D imagery Google Earth already had on file for that exact spot — so a generated image looked like it belonged there.
Within hours, journalist Henk van Ess, who publishes the Digital Digging newsletter, tested how far the tool would go. In a July 31, 2026 post, he described generating imagery of refugees near the Mexican border, a fictitious nuclear plant in Iran, a fatal crash on an Amsterdam street, and a bomb crater beside a hospital in Gaza. His verdict: "Nothing was refused." NPR’s own testing that same day produced a flooded U.S. Capitol complex and a burning oil terminal on Iran's Kharg Island. The Washington Post's Evan Hill called the misinformation potential "literally boundless."
Google pulled the feature that same day, roughly 24 hours after it shipped, per TechCrunch’s reporting. Its statement: "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. We're rolling back this feature in Google Earth while we work on implementing stronger guardrails."
| Date | What happened |
|---|---|
| Feb 26, 2026 | Google launches Nano Banana 2, sets it as the default image model in the Gemini app, Search's AI Mode and Lens, and Flow |
| Thu, Jul 30, 2026 | Google Earth (web) adds a "Create image" button powered by Nano Banana 2 |
| Fri, Jul 31, 2026 (morning) | Van Ess, NPR, and others publish fake imagery of a nuclear plant, refugee camps, a fatal crash, and a bombed hospital, all generated over real Google Earth maps |
| Fri, Jul 31, 2026 (same day) | Google pulls the feature, citing policy-violating imagery, about 24 hours after launch |
Why Google pulled it so fast
Google's defense was that generated images "didn't appear in the main Google Earth experience for others to see and were watermarked as AI generated." Both of those claims turned out to be weaker than they sounded. The images were private to the creator by design, but nothing stopped anyone from screenshotting one and posting it, which is exactly what happened. And the watermark — an invisible SynthID tag readable through the Gemini app or Google Lens — didn't hold up to that screenshot. Forbes reported on August 1, 2026 that after a generated image was simply re-photographed, Hive's AI-detection tool rated the resulting copy just 1% likely to be AI-generated.
That's the part that should worry anyone who treats Google Earth as a neutral reference layer. In my testing of AI image detectors on screenshotted AI art this year, this exact failure mode shows up constantly: any invisible watermark that depends on the original file's metadata tends to disappear the moment an image gets re-encoded, whether that's a screenshot, a recompress, or a repost. SynthID isn't unusually weak here — it's that watermarking as a category doesn't survive the way images actually spread online.
What this means if you use Google Earth or Gemini's image tools
- Nothing changes for normal Google Earth use. The "Create image" button is gone, but the underlying app and its satellite, aerial, and 3D imagery work exactly as they did before July 30.
- The model itself isn't gone. Nano Banana 2 remains the default image model in the Gemini app, in Search's AI Mode and Lens, and in Flow, per Google’s own February 26, 2026 announcement. Only the Earth mapping integration was pulled.
- Don't trust a "satellite-style" image just because it looks like Google Earth. This incident proved that a convincing fake, built from real map data, can pass a casual glance. Treat dramatic images of conflict zones or disasters with more scrutiny, especially ones with no clear original source.
- Don't count on invisible watermarks to catch AI images for you. SynthID failing on a basic screenshot means the responsibility for spotting a fake still falls on you, not on the platform's metadata.
- Expect a comeback, not a cancellation. Google specifically credited "geospatial professionals" with real, useful cases for the tool. Its own wording — rolling back "while we work on implementing stronger guardrails" — points toward a more restricted version returning, not the feature dying for good.
Example prompts you can copy
The removed tool isn't coming back in its old form, but these prompts are useful any time you run into a suspicious "satellite" or aerial-style image online, whether it came from this feature or somewhere else:
- Sanity-check an image: "I'm looking at [describe or paste the image]. Based on lighting, shadow direction, repeated textures, and edge artifacts, does this look like it could be AI-generated? What specific details would you check first?"
- Cross-check a location claim: "Someone is claiming this image shows [event] at [location]. Compare the landmarks, terrain, and structures against verified satellite imagery of those coordinates. Do they match?"
- Check for a SynthID tag: "Open this image in the Gemini app or Google Lens and tell me whether it's flagged as AI-generated."
- Draft an honest caption: "Write a one-sentence disclosure noting that this image is AI-generated for illustration and doesn't depict a real event at this location."
Common mistakes to avoid
The mistake I see most often already, less than a week out, is people assuming Nano Banana 2 itself got banned. It didn't — only the Google Earth mapping integration was pulled; the model still runs in the Gemini app, Search, and Flow.
Second is trusting Google's "private by default" framing too much. Generated images weren't published to the shared Earth view, but that never stopped anyone from screenshotting one and posting it elsewhere, which is exactly how this story broke.
Third is leaning on SynthID as your fact-check. Per Forbes' reporting, a basic screenshot dropped Hive's AI-detection confidence to 1% on a generated image. Treat any invisible watermark as a weak signal, not proof either way.
Fourth is assuming this is over. Google's own statement points at guardrails, not cancellation. Don't be surprised if a restricted version — maybe limited to verified researchers — reappears later this year.
Where Nano Banana 2 still works
| Product | Real-world map overlay | Status as of Aug 2, 2026 |
|---|---|---|
| Google Earth "Create image" | Yes (was) | Pulled July 31, 2026 |
| Gemini app | No | Live, default image model |
| Search's AI Mode and Lens | No | Live, default image model |
| Flow (Google's AI video tool) | No | Live, default image model |
If what you actually wanted was Nano Banana 2's photorealism, not the mapping trick, my Gemini guide covers how to use it in the regular Gemini app. If you're comparing it against other AI image generators, my ChatGPT image tool walkthrough is the closest apples-to-apples comparison I've tested. For a running list of which AI vendor claims actually hold up versus which don't, my AI tool ratings hub tracks stories like this one alongside the tools themselves. My free AI tools roundup and AI tool reviews page are the two I'd start with if you want alternatives without an affiliate pitch attached. For the wider news cycle this story sits in, AISagely’s AI news hub is where I log launches and rollbacks like this one as they happen.
My take
Google shipped this one faster than its own trust-and-safety review could keep up with, and pulling it within a day was the right call. The part I'd flag for anyone building or reviewing AI products: Google's two stated safeguards — private-by-default sharing and an invisible watermark — both failed against the simplest possible workaround, a screenshot. That's not a hard problem to have anticipated. The honest caveat here is that Google didn't kill the underlying capability, just the version that overlaid it on a mapping product people already trust by default. If a restricted version returns, the real test is whether the guardrails hold up to the same screenshot test that broke this one in under 24 hours.
Frequently Asked Questions
Is Google's Earth AI image generator gone for good?
Not necessarily. Google said it's "rolling back this feature in Google Earth while we work on implementing stronger guardrails," which points toward a more restricted version returning rather than a permanent shutdown. As of this writing, Google hasn't given a timeline.
Why did Google pull the Google Earth AI tool so fast?
Because journalists and researchers, including Henk van Ess and NPR, showed within hours that the tool would generate convincing fake imagery of sensitive real-world events — a nuclear plant, refugee camps, a bombed hospital — over real satellite maps with almost no refusals, and its SynthID watermark didn't survive a basic screenshot.
Can I still use Nano Banana 2 anywhere?
Yes. Only the Google Earth mapping integration was pulled. Nano Banana 2 remains the default image model in the Gemini app, in Search's AI Mode and Lens, and in Flow, Google's AI video tool, per Google's own February 26, 2026 announcement.
Were the AI-generated Google Earth images visible to other people by default?
No. Google said generated images "didn't appear in the main Google Earth experience for others to see." The problem was that anyone could screenshot a private generation and share it elsewhere, which is how the misleading examples spread.
How do I tell if a satellite-style image I'm seeing online is real?
Don't rely on an invisible watermark alone — Forbes reported SynthID failed to survive a simple screenshot in testing. Instead, cross-check the claimed location against known satellite imagery, look for lighting and shadow inconsistencies, and treat images with no clear original source skeptically, especially of active conflict zones or disasters.