Google Rolls Back Earth AI Image Generator
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google Earth launched an AI image generator globally on its web version earlier this week, built on the "Nano Banana 2" feature and pitched by Google's blog as a tool for visualizing historical sites and real estate projects.
- Digital Digging's Henk van Ess demonstrated the tool could produce believable flyover images, including composites showing "refugees near the Mexican border" and a bomb crater near a hospital in Gaza.
- Google initially defended the feature, saying every Nano Banana-generated image carries a SynthID digital watermark that users can check via the Gemini app or Lens in Search.
- Digital Digging subsequently fooled Hive's AI detector with an AI-altered video from Google Earth, demonstrating that watermark-based verification alone could be circumvented.
- Van Ess recommended cross-checking suspicious images against independent satellite platforms such as Sentinel-2 or Landsat, and even verifying stated orbital metadata like capture time and satellite identity.
- On July 31st, after the article's publication, Google announced it was rolling back the feature entirely.
- While some outputs were easy to spot as fakes—such as the Sphinx crossed with the Statue of Liberty—others were realistic enough that the author conceded they could plausibly spread as misinformation.
Why it matters: Google's watermark-based defense collapsed under a single public demonstration: Digital Digging produced realistic fake flyovers and then fooled Hive's AI detector with one, forcing Google to roll back the feature the same day it was scrutinized. For any platform layering generative AI on top of trusted geospatial imagery, the episode shows that provenance tooling and third-party detectors are not yet a sufficient substitute for not shipping the capability in the first place.
