Google Pulls AI Satellite Image Editor From Google Earth Less Than a Day After Launch

On 30 July 2026, Google launched a feature in Google Earth that edits satellite imagery with the Nano Banana 2 AI model. On 31 July 2026, less than a day later, it switched the feature off. The reason: a researcher from Digital Digging demonstrated how the tool could add refugees at borders and airstrike damage near hospitals to real imagery.

What the AI Editor in Google Earth Could Do

The new feature built generative AI directly into the Google Earth interface. A user selected a region and entered a text prompt, and Nano Banana 2 redrew the satellite image to match the description. Google pitched the tool as a way to explore places visually: its official demo showed Pompeii reconstructed as it might have looked before the eruption of 79 AD.

Key details of the launch:

• Launch date: 30 July 2026
• Model: Nano Banana 2 (Google's generative AI image model)
• How it works: text prompts applied to real satellite imagery
• Safeguard, per Google: a digital watermark on every output
• Built-in restrictions: blocking of "image creation on harmful topics"

Users could add, remove or alter objects on real maps. In effect, they were editing images of the Earth's surface in a public service with an audience of millions.

Why Google Shut the Feature Down the Next Day

Henk van Ess of the media outlet Digital Digging deliberately tested the tool's limits and documented the results. With a few text prompts, he generated images showing groups of refugees at the Mexican border, a bomb crater next to a hospital in the Gaza Strip, and a nuclear facility in Iran. He described his experiments in an article titled "How to Plant a Nuclear Reactor in Iran," published the same day on the Digital Digging site.

Google's initial response to the criticism was restrained: the company pointed to watermarks and built-in content filters.

"Images created with

Nano Banana 2 in Google Earth include a digital watermark; the company prevents the creation of images on harmful topics," read Google's official reply from the @NewsFromGoogle account on X.

But on 31 July 2026, Google issued a new statement and pulled the feature completely. The tool had been live for less than 24 hours. The watermark proved to be inadequate protection: when images spread on social media, it often goes unnoticed, and pictures that look like real satellite photography are taken as documentary evidence.

What This Means

The Google Earth episode exposed a fundamental risk of AI editing of geodata. Even with filters and watermarks in place, generative AI can produce convincing deepfakes of specific places on the planet, and such a tool is immediately put to use in information manipulation. The danger is especially acute at national borders, in active conflict zones and near civilian infrastructure.

Frequently Asked Questions

What is Nano Banana 2?

Nano Banana 2 is Google's generative AI model for creating and editing images. The Google Earth feature ran on it, synthesizing altered versions of satellite images from users' text prompts.

Why doesn't a watermark protect against deepfakes?

A digital watermark can help identify AI origin under close inspection, but it does not stop an image from spreading: on social media, users see the picture, not its metadata. A convincing satellite image of "refugees at a border" or a "crater next to a hospital" can circulate thousands of times before anyone stops to question its origin.