Google Pulls Nano Banana 2 AI Image Tool From Earth Just One Day After Launch Over Fabricated Imagery

Google pulled a new artificial intelligence image-generation feature from Google Earth on Friday, just one day after launching it globally, after users demonstrated that the tool could be used to fabricate realistic-looking satellite-style imagery superimposed over real, named locations.

The feature, called “Create Image,” used Google’s Nano Banana 2 model, technically known as Gemini 3.1 Flash Image, to generate AI images grounded in actual satellite, aerial and 3D terrain data for any location a user selected within Google Earth’s web platform. Google framed the launch as a creative tool that could let users visualize how a historical landmark once looked, brainstorm urban planning concepts for a vacant lot, or preview real estate development ideas before construction began, distinguishing it from general-purpose AI image generators that start only from a blank text prompt rather than a real, mapped location.

The feature rolled out globally at no additional charge on Thursday, available exclusively through Google Earth’s web version at earth.google.com rather than through the mobile apps, which together account for more than 500 million downloads on the Google Play Store alone. Generated images were watermarked, both visibly and with an embedded invisible marker, and could only be saved within Google Earth projects, without the ability to be directly exported or shared outside the platform, according to Google’s developer documentation.

Those safeguards proved insufficient to prevent misuse once the tool became widely accessible. Digital investigator Henk van Ess demonstrated the feature’s potential for generating misleading content by producing images depicting refugees near the Mexican border, a nuclear facility in Iran, a fatal crash scene in Amsterdam, and a bomb crater near a hospital in Gaza, all superimposed onto real, identifiable map locations within Google Earth’s interface.

The specific interaction model underlying the tool, allowing users to type a text description and have the resulting AI-generated image dropped directly onto a real satellite map location, drew particularly sharp criticism given Google Earth’s established reputation as a widely trusted source of geospatial reference imagery. Journalists routinely use the platform to geolocate conflict footage and verify the authenticity of on-the-ground reporting, human rights investigators use it to corroborate witness accounts of specific events, and academics rely on it as a baseline dataset for geographic and historical research. Layering a generative AI tool directly inside that same interface effectively collapsed the visual distinction between genuine, recorded satellite imagery and synthetic, AI-fabricated imagery within a single trusted platform, a concern critics raised almost immediately after the feature’s rollout.

Google confirmed it was rolling back the feature after becoming aware of screenshots showing generated imagery that the company said appeared to violate its policies. The company said it plans to rebuild the tool with stronger safeguards before any future relaunch, though it has not provided a specific timeline for when a revised version might become available again.

Prior to the rollback, Google had highlighted a range of intended use cases for the tool through its official launch materials, including visualizing historical scenes at real-world landmarks, generating easy-to-read historical infographics tied to specific locations, and exploring conceptual redesigns of public spaces or vacant lots. The company positioned the feature as adding a new creative layer to Google Earth, transforming the platform from a tool that primarily shows users the world as it currently exists into one that also lets them imagine what a given location could become in the future.

Nano Banana 2 sits within a broader family of Google image-generation models of varying capability tiers. It occupies a mid-tier position between the lighter-weight Nano Banana 2 Lite, which can generate standalone images in roughly four seconds through Google’s developer API at a cost of $0.034 per image, and the more capable Nano Banana Pro, which Google has reserved for more complex, professional-grade image rendering tasks.

The rapid rollback comes just days after Google had separately touted the growing role of artificial intelligence within its broader product ecosystem, crediting AI tools with helping identify and fix significantly more bugs within its Chrome browser in June than the company had resolved across the prior two years combined, according to reporting on the company’s recent AI-related announcements.

The Google Earth incident adds to a broader, ongoing debate across the technology industry about how to responsibly deploy increasingly capable generative AI image tools within platforms that carry an established reputation for accuracy and trustworthiness. Watermarking and export restrictions, the two primary safeguards Google built into the initial version of the Nano Banana 2 Earth integration, proved insufficient on their own to prevent the creation of convincing, potentially misleading fabricated imagery tied to real, sensitive locations, underscoring the difficulty technology companies continue to face in balancing creative AI features against the risk of enabling visual misinformation.

With the feature now pulled less than 24 hours after its global debut, Google has not specified what additional guardrails it intends to implement before considering a future relaunch, leaving open questions about whether the underlying tension between creative flexibility and misinformation risk can be fully resolved for a feature integrated directly into one of the world’s most widely used and trusted geospatial reference platforms.