llms.blog
AI

Google pulls AI image tool from Earth one day after launch

# Google pulls AI image tool from Earth one day after launch Google removed the AI image-generation feature from Google Earth on July 31, 2026, roughly 24 hours after switching it on for every user of the browser version. The company said it is "rolling back this feature in Google Earth while we work on implementing stronger guardrails." ## What shipped The feature launched on July 30, 2026. Bryan Horowitz, a product manager on Google Earth, described it as a "create image" button in Earth on

axc
axc
2 min
Google pulls AI image tool from Earth one day after launch

# Google pulls AI image tool from Earth one day after launch

Google removed the AI image-generation feature from Google Earth on July 31, 2026, roughly 24 hours after switching it on for every user of the browser version. The company said it is "rolling back this feature in Google Earth while we work on implementing stronger guardrails."

## What shipped

The feature launched on July 30, 2026. Bryan Horowitz, a product manager on Google Earth, described it as a "create image" button in Earth on the web: zoom to any location, type a prompt, and Nano Banana 2 returns a photorealistic image built from Google's satellite, aerial, and 3D imagery of that exact spot. Availability was global on day one, with no waitlist, no application, and no enterprise-only tier.

The launch post pitched five uses: a classroom reconstruction of Pompeii, an infographic about a landmark, a real-estate concept for an empty Tokyo lot, a cabin rendered into a real lakefront, and a science-fiction remake of Google's Mountain View campus. Its instruction to users was to type whatever they wanted to see.

## What went wrong

Google's statement puts the trigger in distribution rather than in the model. The company said it had seen geospatial professionals using the feature for useful purposes and had also seen people sharing screenshots of generated imagery that appeared to violate its policies. The generated images never appeared in the main Google Earth experience for other users, and they were watermarked as AI-generated. Both controls govern what happens inside Earth. The screenshot is the path the images actually took out of it.

Henk van Ess, who publishes the Digital Digging newsletter, posted a walkthrough hours after the launch showing what he had generated on top of Google's own imagery: refugees near the Mexican border, a nuclear plant in Iran, a fatal crash on an Amsterdam street, a bomb crater beside a hospital in Gaza. He said none of his prompts were refused. NPR ran its own tests and produced Iran's Kharg Island on fire and a flooded US Capitol complex.

## Why the rollback matters

Google Earth's institutional value is that it is the thing other images get checked against. Its archive is dated, its coverage is broad, and newsrooms, human-rights researchers, and open-source investigators treat it as the reference layer when a photograph appears claiming to show a strike, a camp, or a flood. The rollback statement opens on trust rather than capability. The product being defended is the reference, not the generator.

Jake Godin, a senior researcher at Bellingcat, told NPR the change was one of speed rather than of kind. Fabricated satellite imagery circulated well before this launch, including an AI-altered image of a US Navy base in Bahrain that ran in Iranian media earlier in 2026. Producing one previously meant screenshotting Earth, moving the file into a separate generator, prompting it, then cropping the result. The Earth integration compressed that sequence into a single prompt inside a trusted application.

## Sources

- [Google statement on X](https://x.com/NewsFromGoogle/status/2083249962150760610) - [Google Earth launch post](https://blog.google/products-and-platforms/products/earth/nano-banana-google-earth-image-generation/) - [NPR coverage](https://www.npr.org/2026/07/31/nx-s1-5914652/google-adds-ai-to-satellite-images-raising-fears-of-deepfakes-in-the-sky) - [Digital Digging walkthrough](https://www.digitaldigging.org/p/how-to-plant-a-nuclear-plant-in-iran) - [Unite.AI coverage](https://www.unite.ai/google-pulls-earths-ai-image-tool-a-day-after-launch/)

AILLM
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Disaggregated Prefill and Decode in Production LLM Serving: Architecture, Network KV Cache Migration, Chunked Prefill Trade-Offs, and Asymmetric Hardware Economics

    Large language model serving systems have historically treated transformer execution as a homogeneous sequence of forward passes over a single unified GPU pool. Under continuous batching engines, incoming requests execute their prompt evaluation (prefill) and autoregressive token generation (decode) on the exact same accelerators, co-locating both phases within shared iteration batches. While continuous batching improves GPU compute utilization compared to static batching, co-locating prefill a

    1 min

  • Direct Preference Optimization: Mathematical Derivation, Implicit Reward Formulation, and the Mechanics of RL-Free Alignment

    Direct Preference Optimization: Mathematical Derivation, Implicit Reward Formulation, and the Mechanics of RL-Free Alignment Aligning autoregressive large language models with human preferences has traditionally relied on Reinforcement Learning from Human Feedback (RLHF). In the standard formulation popularized by InstructGPT and related post-training regimes, alignment requires a multi-stage pipeline: supervised fine-tuning (SFT), training a separate reward model on pairwise comparison data, a

    1 min

  • Hugging Face Introduces gr.Workflow to Turn AI Pipelines into Visual Graphs and REST APIs

    Hugging Face has released gr.Workflow, a native extension to the Gradio framework designed to convert multi-stage artificial intelligence pipelines into interactive node graphs, visual user interfaces, and deployable REST APIs. Modern machine learning applications increasingly rely on compound pipelines that chain heterogeneous models: generating text via large language models, feeding prompts into diffusion systems, processing outputs through background removal or audio synthesis models, and a

    1 min