By the early hours of August 1, 2026, Beijing time, Google had rolled back its image generation feature in Google Earth on July 31. The company updated its original product announcement, stating that screenshots of generated images shared by users were suspected to violate their policies and would be handled after stronger controls are added. Google did not disclose a restoration date or specify whether the rollback was a pause in testing, closure of access points, or regional withdrawal. It is confirmed that the web-based feature announced globally on July 30 was officially rolled back the next day.
Generation Process and Official Use Cases
Google introduced the operational process on July 30, using the Google Earth web interface as an entry point. Users first zoom into a location, select 'Create Image,' then input the scene they wish to see; Nano Banana 2 subsequently combines satellite, aerial, and three-dimensional images of that location to generate customized visuals. The company describes it as creating concept art within real-world geographic contexts rather than modifying the foundational map seen by everyone. At the time of the announcement, this feature was said to be available globally on web platforms without mentioning mobile app versions.
The official announcement listed five types of scenarios: depicting Pompeii ruins as they appeared in AD 79, creating historical information maps for the Statue of Liberty, envisioning a commercial district on an empty plot in Tokyo, adding lakeside cottages to vacant areas in real landscapes, and transforming Google's Mountain View campus into a futuristic city. Google categorized these examples under historical display, learning, real estate planning, pre-construction visualization, and location transformation. They serve as prompts and generation targets for product demonstrations; the announcement did not provide details on actual project applications.
Output Boundaries and Controversial Cases
Google supplemented two output boundaries in its update on July 31. First, generated images will not be part of what other users see in the main Google Earth experience; second, these images bear an AI-generated watermark. According to this definition, personal generation results or subsequent screenshots should be distinguished from publicly shared satellite, aerial, and three-dimensional base maps provided by Google Earth. The company did not explain how watermarks would appear after cropping, saving, or taking screenshots, nor did it publish test results for identifying the source of generated content.
The Verge reported testing cases that quickly escalated the controversy. Henk van Ess from Digital Digging used text prompts to generate images showing refugees near the Mexican border and craters near a hospital in Gaza. These images are synthetic outcomes based on deliberately set prompts, not evidence of actual events at these locations. TechCrunch summarized the risk as overlaying generated false images onto real Google Earth map contexts.
Official Rollback and One-Day Timeline
In its update, Google acknowledged that users have a special trust in Google Earth providing a reliable view of the world. The company also noted that geospatial professionals had used this feature for various purposes but found screenshots of generated images shared by some users suspected to violate policies, leading to the decision to roll back and strengthen controls. Watermarks, non-inclusion of generated images into public main experiences, and rollback decisions coexist: the first two limit display ranges within the product, while the latter addresses issues exposed when generation content is spread outside the product. Related uses and violations are determined by Google.
The timeline focuses on two calendar days. The original text date on the Google product page was July 30, with the rollback update dated July 31; TechCrunch's related item was publicly released at 19:47:28 UTC on July 31, and The Verge's rollback entry was at 19:13:29 on the same day. Both media outlets summarized the interval between launch and rollback as 'one day,' but existing public information does not provide precise times for when access points were opened and closed, so it cannot be converted into exact operational hours.
Differences in Description and Future Nodes
Media outlets and the official side used different descriptions for the same output. TechCrunch's summary referred to them as generated false images, The Verge's title used 'deepfake tool,' while Google described them as custom maps using real geographic imagery. The first two focus on how synthetic visuals might be misinterpreted as visual evidence when taken out of product context, whereas the official description focuses on in-product generation mechanisms and display boundaries. Google explicitly denied that generated images would enter others' main experiences but did not deny users could share screenshots.
The only verifiable future node is Google's announcement of stronger control measures or reopening related access points. The company has yet to explain which prompts new controls will restrict, how sensitive locations and real-world conflict scenarios will be handled, how intercepted content will be reviewed, nor provided testing standards, restoration regions, or timelines. Public materials have not disclosed the number of users, total images generated, suspected policy-violating screenshot numbers, or their spread range. Therefore, at this stage, it can only be confirmed that the product has been rolled back; it cannot be determined whether it will resume later nor assess how effective watermarks or access point isolation are in preventing misuse.