Google Pulls Nano Banana 2 From Google Earth One Day After Launch Over Compliance Risk

Google integrated its latest image-generation model, Nano Banana 2, into the web version of Google Earth, allowing users to use text prompts to generate modified fictional scenes on real satellite maps. Official examples included restoring a present-day ruin to the ancient Roman street scene of Pompeii as it appeared in AD 78, generating a Statue of Liberty infographic with historical information, reshaping a vacant lot in Tokyo into a commercial district with open public space, adding a modern cottage beside a lake, and transforming Google's Mountain View campus into a futuristic sci-fi city. The feature was not available for direct use in Street View.

The feature was withdrawn after just one day online. Google said it had noticed some professionals using the feature for beneficial purposes, but had also detected screenshots of generated images shared by users that were suspected of violating its usage policy. It decided to suspend and retract the feature to establish a stricter compliance and security protection mechanism.

Nano Banana 2 uses a generation method based on 'geospatial constraints.' The underlying input is not an ordinary flat image, but a composite constraint matrix composed of the original satellite or aerial image of the current viewport, 3D elevation and terrain meshes, and spatial camera parameters. Generation tasks are strictly anchored to real geographic data: terrain and building bases cannot be altered, perspective and spatial relationships must align, and newly generated scenes must be embedded in real 3D structures rather than floating in midair. The model can render five subjects and fourteen objects consistently, but because of the heavy computation, generating each image takes about two minutes.

Beyond image generation, Nano Banana 2 also integrates Gemini's multimodal and retrieval-augmented capabilities. When a user requests a landmark infographic, the model retrieves historical facts such as cultural background and building dimensions through Google Search and a geographic knowledge base, then generates a visualized chart within seconds. The output supports high-fidelity images from 2K to 4K and has strong text-rendering ability, allowing clearly typeset signs or captions to appear in the image. During generation, users can step forward or backward to view changes, or add prompts for iterative revisions. The final image can be saved as a Placemark in a personal project layer.