Google shuts down AI satellite image editing tool after deepfake concerns surface within 24 hours

The Nano Banana 2 feature was removed one day after users generated fabricated images of sensitive locations and demonstrated that its watermarks could be overlooked by detection tools.

Google has rolled back an artificial intelligence feature that allowed users to transform Google Earth imagery through text prompts, one day after launching the tool worldwide.

The feature combined Nano Banana 2 with Google Earth’s satellite, aerial, and 3D imagery. Users could select a location, describe what they wanted to see, and generate a photorealistic version of the altered landscape.

Google promoted the tool for uses including historical visualizations, property development concepts, architecture projects, and urban planning. It became available globally through the web version of Google Earth on Thursday.

The company reversed the launch on Friday after users shared fabricated images involving politically and geographically sensitive locations.

Online research specialist Henk van Ess demonstrated that the tool could add refugees near the US and Mexico border, depict a nuclear facility in Iran, and create artificial damage near a hospital in Gaza. He said the prompts were processed without being rejected or softened.

Google initially defended the feature by pointing to safeguards intended to identify generated content and prevent harmful creations.

Images produced through the tool included Google’s SynthID watermark, which could be detected using the Gemini app or Google Lens. However, van Ess reported that a screen recording of an altered Google Earth image received a 1% AI generated score and a 0% deepfake score from Hive’s detection system.

The result raised concerns that fabricated imagery could circulate through screenshots, recordings, or reposts without viewers noticing the original watermark.

Google said it was removing the feature while developing stronger safeguards, acknowledging that some shared screenshots appeared to violate its policies.

The company stressed that generated images were not added to the main Google Earth experience and were not visible to other users browsing the platform. The altered images existed only as watermarked creations generated by the individual user.

The brief rollout highlighted the risks of placing generative AI tools inside services commonly used to verify real world locations and events.

Google Earth imagery is frequently used by journalists, researchers, and open source intelligence specialists to compare images from conflict zones, disaster areas, and locations that are difficult to access.

Allowing users to generate fabricated scenes from genuine geographical imagery could make false content appear more credible by preserving the lighting, perspective, terrain, and coordinates of the original location.

Google has not indicated when the feature might return. The company said it would restore the tool only after implementing stronger guardrails.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.