Google expands AI image generation in Google Earth with “Nano Banana” for web users

Alphabet’s Google is rolling out a new AI image-generation capability inside Google Earth that lets users create custom views of real places by typing natural-language prompts. In a blog post dated July 30, Google described the feature as “Nano Banana 2” image generation inside Google Earth web, saying it can generate images grounded in the real-world context captured by Google Earth’s satellite, aerial, and 3D imagery.

The company said the experience is designed to be simple: users zoom in to a location on Google Earth for the web, tap “create image,” and then describe what they want to see. Google did not provide technical details about how the prompts are processed beyond stating that Nano Banana creates “concepts grounded in the real world” using Earth imagery as a reference.

Google framed the feature as useful for education and planning, offering examples aimed at students, teachers, and technical users. For instance, it said teachers could prompt the system to render a historic site in a specific time period, such as transforming the “Pompeii ruins” into a bustling Roman-era street scene. Google also described using Gemini, its AI assistant, to retrieve relevant historical information so that the generated infographic can include key facts when users ask for an easy-to-understand graphic.

For real estate and design audiences, Google said the tool can help architects and urban planners communicate visual concepts. It gave an example in which a user reimagines an empty lot in Tokyo as a vibrant shopping and retail district with open spaces, with Google saying barren concrete is replaced by a high-quality 3D rendering that can be used to help clients visualize potential outcomes.

The company also pitched the tool for individual residential visualization, describing prompts that generate a concept for a future home positioned in the real landscape. In its example, a user types a request for a “modern lakefront cabin” using “sustainably sourced local materials,” and Google says the system produces a photorealistic rendering of a cabin placed in the scene.

Google further illustrated the feature’s range with a creative prompt about the Google Mountain View campus transforming into a futuristic “sci-fi utopia” with elements such as glowing walkways, glass biodomes, flying transport pods, and surrounding trees. Google described the result as the area morphing into a “cyberpunk-inspired metropolis,” presenting the capability as a blend of imagination and geographic anchoring.

According to Google, the feature is available globally today for Google Earth web users. That matters because it brings the capability directly into a widely used consumer mapping product, potentially lowering the barrier for people who do not want to build or operate separate AI image-generation tools outside the mapping context.

In business terms, Google’s move targets a recurring challenge in visualization. Whether for development pitches, museum and education use cases, or everyday ideation, translating a concept into a believable image is often time-consuming and requires specialized design work. Google’s approach, as described, shifts more of that effort into an interactive “type and render” workflow tied to real geographic imagery.

Still, the announcement leaves open several questions that typically accompany AI image-generation rollouts. Google did not explain how it handles safety and accuracy for historical reconstructions, how consistently images match complex prompts, or what guardrails limit potentially misleading depictions. It also did not specify whether users can export, share, or store generated images beyond saying the capability is available through Google Earth web.

Next, the practical test for users and developers will be how well the tool captures both spatial grounding and prompt intent across different geographies and image complexities. If Google continues expanding Nano Banana 2 within Earth or adds new capabilities such as improved sharing workflows or deeper data retrieval for educational content, it could further strengthen Google Earth’s role as a visualization platform rather than just a map. Other likely watch items include how Google sets policies around generated imagery in contexts like history and planning, where credibility and context can materially affect how outputs are interpreted.