Hopp til hovedinnhold
Abonnér

Google DeepMind Launches Project Genie — AI That Generates Interactive Worlds

Google DeepMind has made Project Genie available in beta to Google AI Ultra subscribers in the US. Powered by the Genie 3 model, the system allows users to create, explore, and remix interactive environments from simple text descriptions or

Av Joachim Høgby

Google DeepMind has made Project Genie available in beta to Google AI Ultra subscribers in the US. Powered by the Genie 3 model, the system allows users to create, explore, and remix interactive environments from simple text descriptions or images.

What is Project Genie?

Project Genie is a "world model" — an AI model that doesn't just generate static content, but creates fully interactive, playable environments in real-time. Users can describe a scene, and Genie builds a world they can navigate and modify.

The Genie timeline:

  • Genie 1 (2024): Generated 2D games from screenshots
  • Genie 2 (December 2024): Extended to 3D environments
  • Genie 3 (August 2025): Higher resolution, visual consistency over several minutes
  • Project Genie (now): Available to end users via AI Ultra

Practical Use Cases

While it's tempting to imagine anyone creating a GTA clone from their couch, the reality is more nuanced. The technology is promising for:

  • Training and simulation: Organizations can build interactive training scenarios
  • Rapid prototyping: Designers can sketch 3D concepts verbally
  • Games and entertainment: Indie developers can iterate faster on world design
  • Architecture and visualization: Build interactive room walkthroughs from floor plans

Competition and Context

Google positions Genie as a response to OpenAI's Sora (video) and Meta/Apple's emerging world model projects. Combined with Gemini Robotics (integrated into Boston Dynamics' Atlas), DeepMind is painting a picture of AI that doesn't just understand the world — but simulates and interacts with it.

For CIOs and product leaders, this is an early signal: interactive AI-generated simulation is approaching production-ready technology.

Mer om Google

Google and Meta let PyTorch code run natively on Google's TPU chips

Google and Meta have released TorchTPU as open source. Existing PyTorch code should run on Google's TPU chips with trivial changes, but the performance figures are the developers' own and pricing is not stated.

Gemini hacked three companies in a test: guessed passwords, then stopped

Google confirms Gemini broke out of a May Irregular test, guessed passwords and logged into three real companies. The model then stopped. The lab knew in July.

Plugin4Shell: zero-click RCE in four coding agents. Copilot unpatched

AIR: zero-click RCE in Claude Code, Codex, Copilot and Gemini CLI. Pinning does not hold. Anthropic and OpenAI patched. Copilot had no fix. Gemini CLI will not be patched.

OpenAI makes Singapore its first AI lab hub outside the U.S.

OpenAI is putting its first Applied AI Lab outside the United States in Singapore, backed by more than S$300 million and over 200 technical roles. Google is expanding a national AI partnership at the same time.

Oppskriften. Strategien. Hver dag.

AI-nyheter for ledere. Kuratert av en CIO som bygger det selv. Daglig i innboksen.