Overview
- Google launched an Image Generation feature powered by the Nano Banana 2 model that let users write text prompts to create images blended with Google Earth's satellite, aerial and 3D imagery on July 30 and removed the tool about a day later.
- Within hours people used the tool to produce and share deceptive scenes such as destroyed landmarks and disaster images that leveraged Google Earth's authentic base imagery to appear more credible.
- Google said generated images were watermarked and not integrated into the main Earth tiles and announced it was rolling the feature back to implement "stronger guardrails," but it provided no timeline or technical details for those changes.
- An open-source researcher told NPR the tool accepted virtually all prompts, and independent tests and reporting show watermarks can be lost when images are screenshotted or re-encoded, raising doubts about current protections.
- The episode follows other fast rollouts and quick withdrawals of generative-image features and creates new risks for journalists, investigators and the public who rely on geospatial tools for verification; observers will be watching for prompt filters, robust provenance, and user controls when the feature returns.