Google Earth Integrates NanoBanana2 for Real-World Terrain Editing, Pulled After Going Viral
新智元 · wechat · 2026-08-01
Google has integrated its NanoBanana2 image generation model into the web version of Google Earth. Users can input prompts directly on the satellite map to generate high-fidelity 3D buildings or scenes within real geographic environments, such as restoring historical ruins or previewing urban development.
The feature leverages Gemini's multimodal and retrieval capabilities, using "geospatial grounding" to ensure generated content strictly adheres to real terrain and perspective. However, due to hyper-realistic outputs and users wildly abusing the tool, Google urgently rolled back the feature shortly after launch, stating it would reinforce safety guardrails before re-releasing.
Critics noted that while innovative, the tool lacks geometric rigor and doesn't support Street View, with some professionals dismissing it as polished "AI slop." Nonetheless, backed by Google's massive proprietary geographic data, this marks a shift in AI image generation from mere aesthetics to spatially credible "earth editing."
Related event: Google Earth's AI Image Feature Pulled After Abuse Concerns(10 posts)→
More from Multimodal
- Grok Imagine Video 1.5 lands on Runway — runwayml · 2026-08-01
- New ComfyUI Node Automatically Fixes LoRA Color Casts — Forsaken-Mouse-5071 · 2026-08-01
- AI-Generated Music Goes Viral: Fans Dub the Style 'Clanker Gospel' — repligate · 2026-08-01
- Grok's Imagine Video 1.5 Upgrades Hit API with Text-to-Video and Native 1080p — chaitu · 2026-08-01
- Seedance 2.5 Demo: Seamless Switching Between Four Languages — johncoogan · 2026-08-01
- Testing Seedance 2.5: 30-Second Direct Generation Achieves Cinematic Quality — AlchainHust · 2026-08-01