Google Earth’s newest generative-AI feature promises a fresh way to view historic sites and city growth. At the same time, it opens a Pandora’s box of digital misinformation. By letting users reshape satellite and 3D imagery with text prompts, the platform makes “reality-warping” content easy to produce.

The Power and Peril of Nano Banana 2

Google rolled out the feature worldwide on the web version of Google Earth, running on the “Nano Banana 2” model. Google’s docs market it as a tool for planners and historians to picture real-estate projects or ancient ruins. In practice, the tool proves far more volatile.

Digital creator Henk van Ess of Digital Digging has already shown how the model can spawn highly sensitive, incendiary images. With a few simple prompts he generated a scene of “refugees near the Mexican border” and a bomb crater beside a hospital in Gaza. Those images anchor themselves in real geography, unlike obvious hallucinations such as a Sphinx fused with the Statue of Liberty, and therefore pose a greater risk of spreading geopolitical falsehoods.

The Battle Over Digital Watermarking

Google responded by piling on technical safeguards. Every Nano Banana image now carries a SynthID digital watermark. Google tells users to scan a suspicious picture with the Gemini app or Google Lens to flag AI-generated content, and even to add an “@verifyai” tag in Gemini for help.

But the safeguards have not proven bulletproof. Van Ess fooled Hive’s AI detector with an AI-altered video sourced from Google Earth, showing that the watermark alone cannot stop determined actors. The episode underscores an escalating arms race between generative models and the detectors meant to police them.

Why This Matters for the AI Ecosystem

The Google Earth controversy spotlights a looming “reality gap” for the AI industry. As models shift from stylized art to manipulating “ground truth” data—satellite imagery, for example—the stakes jump from aesthetic to existential.

Developers and tech leaders should take this as a warning. When AI powers tools that people trust for objective truth, a digital watermark offers only one layer of defense. Van Ess and other experts argue that the only reliable antidote is cross-platform verification: compare Google Earth outputs with trusted orbital sources such as Sentinel-2 or Landsat. As generative AI embeds itself deeper into mapping and navigation, telling a synthetic render from a real capture will become a core skill for digital literacy.

Key Takeaways

  • High-Stakes Misinformation: Nano Banana 2 can produce hyper-realistic, sensitive imagery that reshapes public perception of geopolitical conflicts.
  • Limitations of Watermarking: Google’s SynthID watermark helps, but demos show AI detectors can still be bypassed, so technical safeguards fall short.
  • The Need for Multi-Source Verification: Users must cross-check satellite images against secondary orbital data from platforms like Sentinel-2 or Landsat to confirm authenticity.

Bottom line

Google’s Nano Banana 2 gives legitimate users powerful visualisation tools, but it also hands anyone a brush to paint false geopolitical narratives on a planet-scale canvas. An invisible watermark hints at synthetic origins; it does not guarantee them. Until detection improves and multi-source checks become routine, viewers must question any striking satellite image that appears out of the blue.