engadget.com web signal

Google Wires Nano Banana Image Generation Into Google Earth

TL;DR

  • Google added a button to Google Earth on the web that pipes the user's current view into its Nano Banana image model.
  • Demos include reimagining Pompeii before Vesuvius, mocking up real-estate projects on empty lots, and generating infographics on landmarks like the Statue of Liberty.
  • The feature launched globally on July 30, 2026, with a Refine Image option, but users cannot reposition the map once a prompt is being written.

The interesting bit of Google's latest Earth update is not that it can now generate AI images. It is that the generator is wired directly to whatever patch of the planet is on your screen. Google added a button to Google Earth on the web that pipes the current view into its Nano Banana image model, Engadget reported, with the feature going live globally on July 30, 2026.

Google's demos cover a wide spread. A teacher can zoom to Pompeii and ask for a hyper-realistic street scene of the city before Vesuvius buried it, turning the modern-day ruins into a bustling, colorful reconstruction. A real-estate developer can look at an empty urban lot and generate a rendering of a proposed building before breaking ground. Somebody hovering over the Statue of Liberty can conjure an infographic labelled with height and construction details. After each generation, a Refine Image button lets you nudge the output before saving it into a project.

Why this matters if you are not a Google Earth power user: it collapses a workflow that used to require jumping into Gemini, pasting a screenshot, describing the location and hoping the model got the geography right. When the map view is itself the prompt, the model already knows what it is drawing. That is a quietly powerful pattern, and it is one only a company with Google's spatial corpus can offer at this scale.

The honest caveat, which Engadget itself raises, is that hyper-realistic does not mean accurate. There were no satellite cameras 2,000 years ago, so any ancient scene the model produces is a plausible confection, not a source, and the same warning applies to those real-estate mockups if a buyer mistakes them for a design document. Launch quirks are real too: you cannot adjust the map view once you have started writing a prompt, so any tweak means quitting the edit, repositioning, and starting again.

What the reporting does not yet cover is whether generated images carry SynthID or any provenance watermark, what moderation applies to sensitive sites, or when the feature might reach the Earth mobile apps. Even so, this is the kind of integration competitors without Google's map data will find genuinely hard to copy, and that is probably the bigger story than any single generated Pompeii.