Google Earth has gained a new AI image generation feature that can turn a text prompt into altered views built from satellite, aerial, and 3D imagery. The result is a powerful visualization tool, but also one that raises a direct question: what happens when synthetic images look close enough to reality to circulate as evidence?
The concern is not theoretical. Digital Digging’s Henk van Ess has already generated examples showing “refugees near the Mexican border” and a bomb crater near a hospital in Gaza, demonstrating how quickly a prompt can produce imagery tied to sensitive real-world subjects.
A familiar map interface now creates synthetic scenes
The feature launched globally on the web version of Google Earth earlier this week. Google’s blog post presents it as a way to visualize historical sites and real estate projects, uses that fit naturally with a tool built around place, terrain, and aerial perspective.
That same realism is what makes the feature complicated. Google Earth is widely associated with geographic reference, so an image that appears to come from that environment may carry an implied credibility for viewers who encounter it outside the original tool.
Some AI-generated examples are obviously unreal, such as an image showing the Sphinx crossed with the Statue of Liberty. But the risk comes from the less absurd cases: images that are not perfect, yet are plausible enough to be shared quickly and interpreted as documentation.
Google points to SynthID and verification tools
Google responded to Digital Digging’s AI-altered images by saying, “We take misinformation seriously – every image created with Nano Banana in Google Earth includes the SynthID digital watermark, so if someone is unsure about an image, they can ask the Gemini app or use Lens in Search to see if the image was AI-generated.”
That response puts verification at the center of Google’s defense. The company is relying on SynthID digital watermarking, along with checks through the Gemini app and Lens in Search, to help people identify images made with Nano Banana in Google Earth.
Those tools matter, but they also shift part of the burden to the viewer. A synthetic image can still travel before someone checks it. It can also be cropped, reposted, embedded in a video, or shown without the context that would prompt a viewer to test whether it was AI-generated.
Why watermarks may not be enough
The source article notes that Google’s watermark may be difficult to remove, but not impossible to miss or bypass. That distinction is important. A watermark can be a useful signal, but it does not guarantee that every person who sees an image will notice it, understand it, or have access to the right verification step at the right moment.
Digital Digging was also able to fool Hive’s AI detector with an AI-altered video from Google Earth. That example shows the limits of relying on detection alone, especially when the output is designed to resemble a familiar satellite or flyover view.
The problem is amplified by the kind of subject matter these images can depict. Border scenes, conflict zones, hospitals, and disaster-like visuals can move quickly through public conversation because they appear urgent. When a generated image enters that kind of context, even a later correction may struggle to undo the first impression.
How suspicious Google Earth images can be checked
Van Ess suggests several ways to verify suspicious images beyond Google’s own tools. These checks are not about trusting a single detector; they are about comparing evidence across sources and asking whether the image matches other available records.
- Use Google’s “@verifyai” tag in Gemini to check suspicious images.
- Compare the image with other satellite platforms, including Sentinel-2 and Landsat.
- Look at orbital data, including when the image says it was taken and by which satellite.
That approach treats AI-generated imagery as something that needs corroboration. A believable aerial image is not the same as a verified image, especially when it was produced by a prompt inside a tool capable of reshaping real-world geography into something that never happened.
The core issue is trust in visual evidence
The Google Earth feature shows the broader tension around AI image generation. The same capability that can help people imagine a historical site or preview a real estate project can also produce scenes that seem grounded in reality because they borrow the visual language of maps, satellites, and flyovers.
For users, journalists, researchers, and anyone encountering these images online, the practical lesson is simple: treat dramatic Google Earth-style visuals with caution unless they can be verified. SynthID, Gemini, Lens in Search, Sentinel-2, Landsat, and orbital data can all help, but none of them remove the need for skepticism.
Google Earth’s Nano Banana 2 feature may be useful, but its most sensitive outputs will demand careful handling. When an AI tool can make a false scene look like a view from above, the future of visual trust depends not only on labels and watermarks, but on whether people know to check before they believe.