Google Earth's New AI Image Generator Can Fabricate Real-Looking Satellite Photos
Pick any real coordinates and generate a photorealistic image grounded in that location's actual map data. An investigator immediately used it to fake a nuclear site in Iran.
Google Earth can now generate fake satellite images of real places. It watermarks them.
Google Earth added AI image generation on July 30, 2026, powered by a Gemini model called Nano Banana 2. Pick any real coordinates on the map, type a prompt, and it generates a photorealistic image grounded in that location's actual satellite and aerial data.
What it actually does
The feature runs in the web version of Google Earth. A user zooms to a real place, clicks create image, and gets output rendered onto Google's genuine basemap for that spot, matching real lighting and terrain. Google frames it as a visualization tool for education, real estate, architecture, and historical reconstruction.
How fast someone found the problem
Open-source investigator Henk van Ess tested it the same day and generated a fabricated nuclear installation at real coordinates in Iran, along with other misleading scenes layered onto real locations. A tool built on top of Google Earth's two decades of documentary credibility makes a fake look more trustworthy than one made in a generic image generator.
The watermark, and its limit
Google's developer documentation confirms every generated image carries two layers: a visible watermark distinguishing it from real map imagery, and an invisible SynthID watermark checkable through the Gemini app or Lens in Search. Generated images cannot be exported directly, they save only as attachments inside an Earth project. Van Ess's underlying point is that the watermark and the tool that reads it both belong to the same company that makes the image generator.
Why a build studio cares
Any product we build that generates or displays AI content inherits this same design question: who controls the proof that something is synthetic, and what happens to that proof once the image leaves the app it was made in.
Next step: read Henk van Ess's original findings or Google's own announcement. If AI-generated content authenticity is part of your product's design, write to us at hello@gattyworks.com.