Google DeepMind 将其 Genie 3 世界模型与多年积累的街景图像数据相结合，推出了一项新功能。用户现在可以在地图上指定一个真实地点，系统便能生成一个基于该地点的、可供行走和探索的 AI 生成世界。这一整合不仅为创意演示提供了强大工具，更重要的是，海量的街景数据成为了训练 AI 代理和机器人的重要战略资源，为人工智能在模拟现实环境中的交互与导航能力开辟了新路径。
原文 · 未翻译
Google pairs its Genie world model with Street View to create explorable AI worlds based on real places
Key Points
Google Deepmind connects its Genie world model to Street View imagery, letting users create interactive, AI-generated environments based on real places.
Google's massive Street View database gives it an edge no competitor can match. The system is primarily designed as a realistic training ground for AI agents, robots, and self-driving cars.
The feature launches as an experimental prototype for Google AI Ultra subscribers, is limited to U.S. locations for now, and still shows visible graphical rough edges.
Google Deepmind is connecting its Genie world model to Street View imagery, letting users generate interactive, AI-built environments that start from an actual location.
Users drop a pin on a map, pick an optional style like "Ocean World," "Desert Sands," "Stone Age," or "B&W film," and describe a character. Genie 3 then builds a walkable world whose starting point is tied to real Street View footage. Google says the feature relies on "Maps Imagery Grounding," an interface developers already use to create AI-powered visuals from Street View data.
The company shows off the tech with demos like a flooded Golden Gate Bridge and the Fort Worth Stockyards in Texas styled after the 1920s.
Google's map data gives it an edge no competitor can match
The real story here is the data. Google has spent years collecting Street View imagery of roads, building interiors, waterways, and remote areas. That library now serves as both training material and an anchor for generative worlds.
Google isn't pitching Genie mainly as a consumer product, though. The world model is meant to give AI agents and robots an environment where they can navigate, reason, and learn. Deepmind's agent SIMA 2 already uses Genie as a training ground. Waymo, meanwhile, uses it to simulate realistic street scenarios for self-driving cars. The Street View connection means these training runs can now be pinned to specific real-world locations.
Google DeepMind 将其 Genie 3 世界模型与多年积累的街景图像数据相结合，推出了一项新功能。用户现在可以在地图上指定一个真实地点，系统便能生成一个基于该地点的、可供行走和探索的 AI 生成世界。这一整合不仅为创意演示提供了强大工具，更重要的是，海量的街景数据成为了训练 AI 代理和机器人的重要战略资源，为人工智能在模拟现实环境中的交互与导航能力开辟了新路径。
原文 · 保持原样，未翻译
Google pairs its Genie world model with Street View to create explorable AI worlds based on real places
Key Points
Google Deepmind connects its Genie world model to Street View imagery, letting users create interactive, AI-generated environments based on real places.
Google's massive Street View database gives it an edge no competitor can match. The system is primarily designed as a realistic training ground for AI agents, robots, and self-driving cars.
The feature launches as an experimental prototype for Google AI Ultra subscribers, is limited to U.S. locations for now, and still shows visible graphical rough edges.
Early demos look like a mix of tech showcase and game preview
Bilawal Sidhu, a former Google product manager for AR/VR and 3D Maps, posted a string of test prompts that show what this looks like in practice. He races a Google Maps-branded Formula 1 car down the Las Vegas Strip, rides past the Palace of Fine Arts as a squirrel on a scooter, steers a boat across Austin's Lady Bird Lake, and walks through the White House using indoor Street View data.
Woot! You can now simulate real world places by grounding Genie 3 experiences with Street View imagery. Google sitting on the mother lode of real world data, and is starting to put it to work!AdDEC_D_Incontent-2 Let's dive into some prompts & locations I tested... pic.twitter.com/TlXSWNx08hAd - Bilawal Sidhu (@bilawalsidhu) May 19, 2026
Woot! You can now simulate real world places by grounding Genie 3 experiences with Street View imagery.
Google sitting on the mother lode of real world data, and is starting to put it to work!
Let's dive into some prompts & locations I tested... pic.twitter.com/TlXSWNx08h
Bilawal Sidhu (@bilawalsidhu) May 19, 2026
The examples show the range but also the rough edges that are still visible: soft textures, unstable geometry, and surreal transitions.
Who gets access, and what's still missing
Project Genie with Street View grounding is rolling out globally to Google AI Ultra subscribers ($200 per month, 18+). The real-world location feature only works for U.S. spots for now, with plans to expand.
Deepmind itself calls this an experimental research prototype and says the team is working on a better version.
Google Deepmind is connecting its Genie world model to Street View imagery, letting users generate interactive, AI-built environments that start from an actual location.
Users drop a pin on a map, pick an optional style like "Ocean World," "Desert Sands," "Stone Age," or "B&W film," and describe a character. Genie 3 then builds a walkable world whose starting point is tied to real Street View footage. Google says the feature relies on "Maps Imagery Grounding," an interface developers already use to create AI-powered visuals from Street View data.
The company shows off the tech with demos like a flooded Golden Gate Bridge and the Fort Worth Stockyards in Texas styled after the 1920s.
Google's map data gives it an edge no competitor can match
The real story here is the data. Google has spent years collecting Street View imagery of roads, building interiors, waterways, and remote areas. That library now serves as both training material and an anchor for generative worlds.
Google isn't pitching Genie mainly as a consumer product, though. The world model is meant to give AI agents and robots an environment where they can navigate, reason, and learn. Deepmind's agent SIMA 2 already uses Genie as a training ground. Waymo, meanwhile, uses it to simulate realistic street scenarios for self-driving cars. The Street View connection means these training runs can now be pinned to specific real-world locations.
Early demos look like a mix of tech showcase and game preview
Bilawal Sidhu, a former Google product manager for AR/VR and 3D Maps, posted a string of test prompts that show what this looks like in practice. He races a Google Maps-branded Formula 1 car down the Las Vegas Strip, rides past the Palace of Fine Arts as a squirrel on a scooter, steers a boat across Austin's Lady Bird Lake, and walks through the White House using indoor Street View data.
Woot! You can now simulate real world places by grounding Genie 3 experiences with Street View imagery. Google sitting on the mother lode of real world data, and is starting to put it to work!AdDEC_D_Incontent-2 Let's dive into some prompts & locations I tested... pic.twitter.com/TlXSWNx08hAd - Bilawal Sidhu (@bilawalsidhu) May 19, 2026
Woot! You can now simulate real world places by grounding Genie 3 experiences with Street View imagery.
Google sitting on the mother lode of real world data, and is starting to put it to work!
Let's dive into some prompts & locations I tested... pic.twitter.com/TlXSWNx08h
Bilawal Sidhu (@bilawalsidhu) May 19, 2026
The examples show the range but also the rough edges that are still visible: soft textures, unstable geometry, and surreal transitions.
Who gets access, and what's still missing
Project Genie with Street View grounding is rolling out globally to Google AI Ultra subscribers ($200 per month, 18+). The real-world location feature only works for U.S. spots for now, with plans to expand.
Deepmind itself calls this an experimental research prototype and says the team is working on a better version.