Google Earth Now Lets You Reimagine Any Place With AI
Google added Nano Banana image generation to Google Earth, letting anyone visualize real locations in the past, future, or reimagined for free.
What is the new Nano Banana feature in Google Earth?
Google has built its Nano Banana image generation model directly into Google Earth, letting anyone turn a 3D map view into an AI generated image with a text prompt. Instead of screenshotting a location and pasting it into a separate image tool, users can now click a “create image” button right inside Google Earth, describe how they want the scene changed, and get a rendered result based on that exact camera angle. It works on real, current locations anywhere in the app’s 3D view, and Google has made it free to use.
TL;DR
- Nano Banana is now built into Google Earth, so users can generate AI images from any 3D map view without leaving the app or using a separate tool.
- The feature works by taking a screenshot of the current 3D view and feeding it to the image model along with a text prompt, then returning a reimagined version of that same scene.
- Google’s own examples show the tool recreating historical scenes (like Pompeii around 78 AD), turning landmarks into infographics, and previewing real estate or development projects before they’re built.
- Testing the feature on Petco Park in San Diego produced a coherent “100 years in the future” version of the stadium, bridge, and surrounding buildings, all generated from a single prompt.
- The tool is currently free to access through Google Earth, with no separate subscription mentioned.
- This launch is part of a broader push to put generative image tools inside everyday mapping and productivity software, rather than keeping them siloed in standalone AI apps.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
How does the Google Earth AI image tool actually work?
The mechanics are simple even though the output can look impressive. A user navigates Google Earth’s 3D view to a specific spot, whether that’s a stadium, a neighborhood, or a landmark, and adjusts the camera angle. A “create image” button then opens a prompt box where the user types a description of the change they want: a different time period, a redesigned building, an infographic style, or a speculative future version of the same skyline.
Behind the scenes, Google Earth captures the current 3D view as an image and passes it to Nano Banana, its image generation model, along with the user’s text prompt. The model generates a new image based on both the visual reference and the instructions. Because the base image already reflects the real geometry of the location (the bridge, the field, the surrounding buildings), the output tends to keep those structural elements recognizable even as it reimagines textures, materials, or added features.
This is functionally the same workflow people were already doing manually (screenshot a map, drop it into an image generator, write a prompt). Google’s contribution here is convenience: it removes the copy-paste step and ties the feature to a “create image” button inside the map itself.
What can you actually generate with it?
Google’s published examples cover a few categories:
Historical reconstruction. One demo takes a present-day view of Pompeii’s ruins and generates a hyperrealistic rendering of what the site may have looked like in 78 AD, before the eruption that buried it.
Infographic styling. A shot of the Statue of Liberty gets converted into an infographic-style image, turning a straightforward 3D view into something closer to an educational graphic.
Real estate and development previews. The tool can render what a section of a city might look like after a proposed development, or how a specific house might look once built in its actual surrounding environment. This is aimed at things like visualizing projects before construction starts.
Speculative “future” scenes. Users can also just ask for creative reimaginings, like what a location will look like decades from now, with no real-world reference point at all.
In a hands-on test using Petco Park in San Diego (home of the Padres), prompting “reimagine this area 100 years in the future” produced a scene where the stadium field was covered by a dome-like structure (while the seating stayed open), the nearby Coronado Bridge gained some kind of elevated transit line, and several buildings took on plant-covered, greened-over facades. Not every element changed. Some existing structures in the generated image stayed essentially identical to the present day, suggesting the model anchors heavily on the original photo rather than redesigning the whole scene from scratch.
Is this actually useful, or just a novelty?
Other agents start typing. Remy starts asking.
Scoping, trade-offs, edge cases — the real work. Before a line of code.
Both, depending on the use case. For casual users, this is mostly a fun way to explore “what if” scenarios: what a childhood neighborhood might look like in a century, what a historical landmark looked like before modern development, or what a city skyline might become under some speculative theme.
For more practical use, the real estate and urban planning angle is where it has clearer value. Being able to generate a visual preview of a proposed building or development, placed directly into its actual surrounding context (real streets, real neighboring structures, real terrain), gives a faster and cheaper way to communicate a concept than commissioning custom renderings. It won’t replace professional architectural visualization for serious projects, but for early-stage pitches or public communication about a development, it’s a low-effort way to generate a plausible visual.
The historical reconstruction angle also has some educational value: teachers, students, or just curious users can get a rough visual sense of what a real location looked like in a different era, grounded in the actual geography of the place rather than a generic illustration.
How does this fit into the bigger AI news cycle?
This Google Earth update landed in a stretch of AI news dominated by bigger, noisier releases, including Anthropic’s Claude Opus 5. Opus 5 drew a mixed reception from developers: strong benchmark numbers in categories like coding and agentic tasks, but widespread complaints about verbosity, inconsistent quality, and a “personality” that made everyday use frustrating, especially compared to its predecessor Opus 4.5. One area where Opus 5 stood out was generating visually impressive browser-based games using tools like Three.js, producing environments with far more polished graphics than earlier models, even when gameplay mechanics (controls, combat, physics) lagged behind the visuals.
Against that backdrop, the Google Earth integration is a smaller, more contained release: no benchmark debates, no rate limit complaints, just a free creative tool added to software many people already use. It reflects a pattern seen across the industry this year: generative image and video models increasingly show up embedded inside existing products (maps, browsers, productivity suites) rather than requiring people to open a separate AI app.
Frequently Asked Questions
Is the Nano Banana feature in Google Earth free?
Yes. Based on demonstrated use, the feature is accessible inside Google Earth without an additional paid subscription for the image generation itself.
Do I need the Google Earth app, or does this work in a browser?
The demonstrated workflow used Google Earth’s 3D view with a “create image” button appearing directly in the interface. It works wherever that 3D view and button are available in Google Earth.
Can it generate images of any location on Earth?
It works on any location viewable in Google Earth’s 3D mode, since the tool generates its image based on a screenshot of the current view combined with your prompt.
How accurate are the historical or future recreations?
They are AI generated interpretations, not verified historical or architectural reconstructions. They’re useful for visualization and imagination, not for factual or engineering accuracy.
How is this different from just using Nano Banana on its own?
Plans first. Then code.
Remy writes the spec, manages the build, and ships the app.
Functionally, the output is similar to manually screenshotting a Google Earth view and running it through Nano Banana separately. The integration mainly removes that manual step by building the prompt and generation flow directly into Google Earth.
