谷歌在其地球网页版中推出了基于Nano Banana 2模型的图像生成功能，但这项新功能在发布不到一天后被紧急撤回[1]。该功能允许用户通过点击"Create image"按钮，在真实地理位置上根据文本提示生成AI定制化图像[1]。
Nano Banana 2采用了"地理空间约束"技术，融合了Gemini多模态能力与搜索增强功能[1]。生成单张图像需耗时长达两分钟[1]。该模型在Arena排行榜上排名第五，落后于OpenAI的GPT-image-2[1]。谷歌利用二十年积累的卫星影像、航拍照片及覆盖40多国的3D地形建模作为差异化优势[1]。
功能的技术底层由三部分组成：物理底图实时捕捉、世界知识库实时检索，以及高保真细节与精确文本渲染[1]。原计划中，这一功能支持历史重现、虚拟导游、房地产规划和建筑可视化等应用场景[1]。谷歌承诺在加强防护措施后重新发布该功能[1]。
Google has withdrawn an artificial intelligence image generation capability from Google Earth's web version shortly after its introduction [1]. The feature, which integrated the Nano Banana 2 model, allowed users to generate customized images at real geographic locations by clicking a "Create image" button and entering text prompts [1].
The technology leveraged Google's proprietary "geospatial constraint" system, which combined real-time physical base maps, live world knowledge retrieval, and high-fidelity detail rendering with precise text generation [1]. Each image could take up to two minutes to produce [1]. The feature promised applications ranging from historical recreation and virtual tours to real estate planning and architectural visualization [1]. Google's competitive advantage stemmed from two decades of accumulated satellite imagery, aerial photography, and 3D terrain modeling covering over 40 countries [1].
The function was removed less than a day after launch [1]. Google has committed to implementing enhanced safeguards before reintroducing the capability [1]. At the time of withdrawal, Nano Banana 2 ranked fifth on the Arena leaderboard, behind OpenAI's GPT-image-2 [1].
