Google Earth is the latest battleground for AI image generation. The service, which offers satellite and aerial imagery of the planet, briefly gained a new feature that allowed users to reimagine real landscapes with the help of Gemini's Nano Banana 2 model. That feature is now gone, at least for the moment. The announcement came on Wednesday, July 30, 2026, and the rollback was confirmed on Thursday, July 31. There is no official word yet on when the feature might return, but Google has promised that it is working on stricter policies and safeguards.

• Google announced the integration of Gemini's Nano Banana 2 image generation into Google Earth on Wednesday, July 30, 2026.
• The feature was rolled back within 24 hours due to policy violations.
• Google says geospatial professionals used the feature, but other users shared generated imagery that appeared to violate policies.
• Generated images did not appear in the main Google Earth experience and were watermarked as AI-generated.
• Google plans to bring the feature back after implementing stronger guardrails.

A Quick Reversal

According to Google's official news account on X, the decision to pause the feature was driven by observations that some users were sharing screenshots of generated imagery that appear to violate Google's policies. This response came after Google had rolled out the feature to everyone, hoping to broaden the creative possibilities of Google Earth. The company also noted that geospatial professionals had already begun to experiment with the tool and were using it for what appeared to be legitimate and valuable use cases, including environmental planning, architectural visualization, and educational demonstrations.

"We know that people uniquely trust Google Earth for a reliable view of the world," the statement read. "We've seen geospatial professionals using this feature for a range of useful purposes, however we've also seen people sharing screenshots of generated imagery that appear to violate our policies. So we're rolling back this feature in Google Earth while we work on implementing stronger guardrails. It's important to note that generated images didn't appear in the main Google Earth experience for others to see and were watermarked as AI generated."

Google has not shared anything more specific about the policy violations it uncovered, nor what exactly it intends for these stronger guardrails to block. The company changing its mind about features is hardly unheard of for Google, but the extremely compressed timetable we are looking at here is definitely a little out of the ordinary.

What Is Nano Banana 2?

For those who are not familiar, Nano Banana 2 is the second generation of Google's image generation model that runs on the Gemini platform. The first Nano Banana, introduced earlier in 2026, was praised for its ability to generate highly detailed and context-aware images. It could take a text prompt and produce a photorealistic image that matched the description with surprising accuracy. Nano Banana 2 builds on this foundation, adding enhanced spatial awareness, better text rendering, and improved photorealistic quality. This makes it especially suited for tasks like editing satellite imagery, because it can maintain the underlying geography while adding or removing visual elements in a realistic way.

Why Google Earth?

Google Earth is a mapping and research tool that has been around for more than two decades. It allows users to explore the globe via satellite, aerial, and street-level imagery. It is used in schools, government agencies, and even in emergency response. Adding a generative AI feature to Google Earth could make it easier to visualize future scenarios, such as sea level rise, new building designs, or the impact of deforestation. For example, an urban planner could use the feature to see how a new park might look from above, or a filmmaker could scout a location and then generate a changed version to match a script. The opportunities are vast, but so are the risks.

Policy Violations and Trust

The specific policy violations that led to the rollback have not been revealed. Google's AI generation policies, which apply to the Gemini family of models, prohibit creating content that is sexually explicit, violent, hateful, or that depicts real people without consent. The policies also restrict content that is designed to deceive, including misleading news images, realistic fake documents, or false representations of public events. Users may have generated images that, while not pornographic or violent, were so convincingly realistic that they could be confused with official Google Earth data. That would be particularly dangerous. Someone could create an image of a flood or hurricane in a region that is actually unaffected, leading to public panic or false reporting.

Moreover, Google Earth is a platform that many people trust for planning and navigation. If users see a picture that looks like a Google Earth screenshot, they may incorrectly assume it is accurate. Even with a watermark, the image can be cropped or altered. Researchers have repeatedly shown that watermarks are ineffective deterrents when screenshots are shared on social media, where they can be zoomed in on, cropped, or simply re-saved without the watermark. This is likely one of the challenges Google is trying to solve with its supposedly stricter guardrails.

The Future of AI Imagery

The rapid backtracking raises important questions about the future of AI-generated content in mapping applications. Should companies like Google allow users to manipulate satellite imagery at all? And if they do, how can they maintain the distinction between fiction and reality? These questions are not new, but they are becoming more pressing as AI models become more capable. Google has previously integrated AI in Google Maps for features like augmented reality navigation and photorealistic aerial views, but these are carefully controlled and based on actual imagery. The Nano Banana experiment was different because it gave every user the power to create entirely synthetic imagery from scratch. That is a profound step.

Google's plan to bring the feature back after implementing stronger guardrails suggests that the company does not intend to abandon the idea. It may instead limit the feature to particular types of usage, such as in a separate workspace or with additional verification for professional accounts. It might also block certain prompts or geographic regions. Some experts have suggested that AI-generated imagery should include cryptographic metadata, so that a simple scan can identify the image as synthetic. Others have argued for strict labeling and user verification. Google has not yet given a timeline, and its track record suggests that the feature could be reintroduced in a limited form after a few weeks or months.

The entire episode is a reminder that powerful AI tools can be both creative and dangerous. Google, which has positioned itself as a responsible leader in AI, must balance innovation and safety. The company has been criticized in the past for releasing AI features that later had to be tweaked after embarrassing incidents, such as inaccurate image descriptions or biased responses. The Google Earth integration was designed to make the service more fun and compelling, but it overlooked the basic truth that some things are better left unmodified. The world we see through Google Earth is real, and that reality is a source of value. If anyone can change it, that value is diminished.

For now, users who missed out on the brief window of Nano Banana in Google Earth will have to wait. It is unclear whether the feature will return in the same form, or whether Google will decide to keep it locked down. The company continues to work on stronger guardrails, and hopefully the next version will find a way to give users creative freedom without undermining the trust that makes Google Earth indispensable.

Source: Android Authority News