FULL STORY

Google Earth's AI Image Feature Pulled Over Abuse

Google Earth launched an AI image generation feature that was quickly pulled within 24 hours due to widespread concerns over geographic deepfakes and abuse.

2026-07-30 ~ 2026-08-03 · 3 episodes · 23 posts

Episode 1 · Google Earth AI Image Generation Pulled After 36 Hours Due to Abuse (2026-07-30, 17 posts)

On July 30, Google launched an AI image generation feature in the web version of Google Earth, powered by Nano Banana 2 (i.e., Gemini 2.0 Flash image generation). Users could zoom into an area and enter text prompts to creatively transform real street views and city buildings using satellite and 3D map data. However, due to severely ineffective safety guardrails, the feature was widely abused to create realistic fake geographic scenes, leading Google to take it down after about 36 hours.

Confirmed

  • Google officially announced the AI image generation feature in Google Earth, driven by the Nano Banana 2 model, combining satellite, aerial, and 3D imagery data to generate custom images.
  • Multiple users tested its creative potential: @rseroter transformed a corner of downtown San Diego into a creative scene, @bilawalsidhu demonstrated visual reskinning of houses and city buildings, even time-travel transformations, and @emmanuelvivier noted applications in historical visualization, urban planning, and real estate.
  • To address potential forgery risks, images generated based on real geographic coordinates were accompanied by SynthID watermarks and C2PA credentials.
  • Due to failed guardrails, users could easily generate policy-violating fake images. Journalist @henkvaness tested that with a single sentence, one could fabricate realistic scenes such as refugees at the US-Mexico border, an Iranian nuclear plant, a bombed hospital in Gaza, or even explosion craters on real maps.
  • Facing the proliferation of fake images and panic, Google rolled back the feature about a day after launch. Google stated that these generated images were not publicly shown to other users in the main app, but there was a recent spread of generated screenshots violating platform policies. They plan to consider restoring the feature after implementing stricter safety guardrails.

Unconfirmed

  • Researcher @FelixSimon expressed doubts about the feature's effectiveness in preventing misinformation, suggesting that existing mechanisms like SynthID watermarks and C2PA credentials may not be sufficient to mitigate the risk of misleading forged images.

Why it matters

  • This feature significantly lowered the barrier to tampering with geographic imagery, allowing users to place fictional events and structures directly onto real satellite images. The generated fake images blend seamlessly with real geographic environments, and if abused, could easily lead to misleading propaganda, posing a serious challenge to the authenticity and credibility of geographic information.

Episode 2 · Google Earth Launches AI Image Generation (2026-07-31, 2 posts)

Google Earth introduced a new AI image generation feature that transforms real-world street views into artistic styles like comic books. However, users currently cannot directly download the generated images.

Episode 3 · Google Earth Pulls AI Image Feature Over Deepfake Fears (2026-08-01, 4 posts)

Google Earth briefly launched an AI image generation feature integrating the "Nano Banana 2" model, allowing users to modify satellite images via text prompts. The feature was taken down within 24 hours due to serious concerns over geographic deepfakes and spatial data security.