The cyber version of "words become law" has reached a whole new level!
Yesterday, Google officially integrated its latest image generation model, Nano Banana 2, into the web version of Google Earth.
Who could have imagined that upon waking up, the new feature was broken, and Google urgently rolled it back!
The result was too realistic; after strong opposition from experts, Google urgently withdrew the new feature, then re-released it after "enhancing protective measures."
New gameplay for Nano Banana 2: a visual feast for netizens
You open Google Earth in your browser and skillfully zoom down to the old street where you played as a child. You don’t open complex Photoshop software, nor do you copy a screenshot into another AI chat tool.
You simply clicked the brand-new, softly glowing Create Image button in the top-right corner.
First, let's look at a few examples provided by Google.
Rewind time and relive history
Help students not just read about history, but truly see the past.
The teacher can display a historical site in class and then input: "Generate a hyper-realistic image depicting the original appearance of the city of Pompeii in 78 AD."
Modern ruins instantly transform into a vibrant, colorful ancient Roman street scene, allowing students to visually experience the world of two thousand years ago.
Virtual Guide: Explore and Deepen Your Understanding Along the Way
Create customized infographics to gain deeper insights into a location without leaving Google Earth.
Create an easy-to-understand infographic of the Statue of Liberty, highlighting key historical facts.
Behind the scenes, Gemini retrieves relevant historical data, and Nano Banana generates corresponding visual charts within seconds.
"Dream City Renovator": Create Professional-Level Real Estate Planning Solutions
Real estate projects become easier to present when clients can clearly see the final outcome.
Architects and urban planners can now bring their designs to life with a single prompt: “Reimagine this vacant land in Tokyo as a vibrant retail district featuring open public spaces.”
The once barren concrete lot will be replaced with a high-quality 3D-rendered modern urban oasis, helping clients visualize future possibilities.
Visualize the project outcome in advance before construction begins.
Whether it’s a backyard studio you’ve always dreamed of or a lakeside home, it’s often hard to visualize what the finished result will look like.
Now, you can quickly turn it into a real-time preview.
Zoom in on an open area, then enter: "Add a modern lakeside cabin built using locally sourced sustainable materials."
The system will generate a photorealistic rendering that seamlessly integrates the future residence into the natural terrain.
Give your favorite place a future makeover
Sometimes, letting your imagination run free is its own kind of joy.
If you'd like to see what a place might look like a hundred years from now, enter: "Transform the Google Mountain View campus into a futuristic sci-fi utopia featuring glowing walkways, glass ecological domes, flying transit pods, and trees growing around the buildings."
Then, the entire area will gradually transform into a cinematic cyberpunk metropolis.
Players exclaim, "Oh My God"
Tech journalist David Gewirt spent hours exploring Google Earth, captivated by the ability to travel the world without leaving his home. Although the Street View feature was somewhat invasive, it fascinated him then and continues to do so today.
Upon hearing the news, he immediately tested the new feature, which "led to a lot of laughs."
Independence Hall is a U.S. National Historic Landmark and a UNESCO World Heritage Site. The U.S. Declaration of Independence was adopted here on July 4, 1776.
Undoubtedly, this holds significant historical and cultural value for the United States.
But David Gewirt has a wild idea: “Imagine this is set in a post-apocalyptic dystopian future 500 years from now.”
One moment, it was a grand palace; the next, it became a post-apocalyptic wasteland.
One of the most iconic Philadelphia traditions is the Mummers Parade, the oldest annual folk parade in U.S. history and a New Year’s tradition since 1901.
This made him wonder: Philadelphia in the future needs a post-apocalyptic dystopian Mummers Parade.
Then he added, "Now imagine it being taken over by zombies, evil clowns, and giant alien mechs."
Zombies, evil clowns, and giant alien mechs, oh my!
Zombies, evil clowns, and giant alien mechs are tense, the atmosphere is off—but one more hint would make it perfect. He told the AI, “Make them all happy.”
Now, he believes: "This is a credible future Philadelphia, especially on January 1st."
Please note that he also reminded: "Although this can lead to some interesting experiments, it is clearly not yet a reliable or practical tool."
Currently, the new feature is not fully supported for direct use in Street View, which diminishes its practicality and appeal—regrettably.
Some media harshly criticized it: for professional developers or architects, this artwork, which lacks geometric precision and is merely pieced together by chance, is nothing more than refined “AI slop.”
Unexpectedly, the feature was ruined by netizens in less than a day—Google has urgently rolled it back!
Technical Breakdown
From "something from nothing" to "geospatial constraints"
In early AI image generation logic, the AI was like playing a game of pixels.
You give it a prompt, and it pieces together a image from a vast probability distribution—it doesn’t care about gravity, nor whether the building in the image actually exists in reality.
However, the integration of Nano Banana 2 into Google Earth has completely changed the game. This is a novel approach known as "geospatial grounding."
At the technical level, it leverages Gemini's powerful multimodal and retrieval-augmented (Search Grounding) capabilities:
First is the real-time capture of the physical map.
The underlying input received by Nano Banana 2 is not a standard overhead image, but a composite constraint matrix consisting of the current viewport's satellite/aerial raw imagery, 3D elevation and terrain mesh, and spatial camera parameters.
It is presumed that its generation task is strictly anchored:
Terrain is immutable: the folds of rolling hills, the baseline of lake shores, and the precise location and scale of building foundations must strictly conform to geographic reality.
Perspective and spatial relationships must align: newly generated buildings, vegetation, or scene elements must be seamlessly integrated into the real 3D terrain and existing structures, not floating in mid-air.
Consistency: Nano Banana 2 supports ultra-high consistency rendering for up to 5 subjects and 14 objects, ensuring no noticeable visual collapse when switching perspectives in new cities or communities.
This is conditioned generation under real geographic constraints, with each Google Earth image taking up to two minutes to generate.
Second, real-time retrieval from the global knowledge base.
Nano Banana 2 is not just an image generator—it is connected to Google’s vast search engine and geographic knowledge base. When you attempt to “reconstruct history” or “label landmarks,” it actively retrieves the cultural context and geographic knowledge associated with that landmark.
For example, the user requests the generation of an infographic of the Statue of Liberty.
The model not only generates images but also calls upon Gemini’s knowledge base to automatically fill in details such as construction year, materials used, building dimensions, and historical information.
However, whether the information retrieval is entirely accurate is another issue that requires careful consideration.
AI can generate professionally looking infographics, but that doesn't mean all the information is absolutely reliable.
Third, high-fidelity details and precise text.
Nano Banana 2 supports high-fidelity output at 2K and even 4K resolutions, and features exceptional text rendering capabilities. This means you can directly see clearly laid-out, artifact-free text signs or informational labels in generated scenes, and even "hand-draw" informative infographics with accurate historical facts directly onto landmarks.
On an operational level, Google has compressed this complex technology to its essence: generation supports forward and backward navigation to view changes, clicking “Refine image” lets you add prompts for further iterations, and the interface provides an intuitive before/after experience.
This directly eliminates the cumbersome workflow of “screenshotting—exporting—inputting into another AI—reassembling,” enabling a seamless, native fusion of “creativity” and “real-world geography.”
After you're satisfied, click "Save to Project" to permanently pin the image as a Placemark on your personal project layer, preserving the camera perspective. Anyone who visits this location will see your "Another Possibility."
Does Virtual Earth help Google Gemini break through?
The significance of Nano Banana entering Google Earth may far exceed that of a new feature.
The AI image generator market is currently in fierce competition.
OpenAI's GPT-Image-2 leads the Arena leaderboard, with Nano Banana 2 Lite in fifth place. Adobe Firefly, Midjourney, and others are all competing for the same user base.
Google took a completely different path.
Its edge isn't algorithms—it's data. Two decades of accumulated satellite imagery and aerial photographs, combined with 3D terrain modeling of hundreds of cities across more than 40 countries—this proprietary spatial dataset is a moat no pure text model can ever reach.
Attach Nano Banana 2 to this dataset to create an entirely new category: AI visualization grounded in real geographic coordinates with spatial credibility.
It’s not about comparing who draws more beautifully than Midjourney, but who draws more accurately—so accurately that it includes latitude, longitude, elevation, and physical perspective.
Kevin Kelly once predicted that the physical world would give rise to a strictly corresponding "mirror world" in the digital realm. Today, Google has lightly applied a "digital skin, arbitrarily re writable by prompts" to this mirror world using Nano Banana 2.
Google has directly escalated the battlefield from "drawing beautifully" to "transforming the Earth." Others can't even get a ticket to this fight.
Reference materials:
https://blog.google/products-and-platforms/products/earth/nano-banana-google-earth-image-generation/
https://www.zdnet.com/article/google-earth-added-nano-banana-and-i-immediately-reimagined-philly-with-zombies-and-evil-clowns/
https://www.techtimes.com/articles/322272/20260730/google-earth-gets-ai-image-generator-grounded-satellite-data-web-only.htm
https://x.com/googleearth/status/2082818165503902043
https://x.com/Google/status/2082834786310123787
Edited by: David
This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation.