Google showcases its latest AI model, Genie 3. The model can design interactive worlds as you explore them.
Google has released a demo of Genie 3, the latest AI model developed from the DeepMind lab. The model is a combination of the previous Genie generation and the AI video generator Veo. However, Genie 3 is not limited to short video clips and can design and adapt virtual worlds in real-time.
The demo video shown by Google DeepMind resembles a VR game advertisement without context. Genie 3 designs virtual worlds based on a text prompt that you can explore interactively. These live simulations can range from skiing in the mountains, traveling back to a historical period, to everyday situations like painting your house.
“It goes far beyond the limited models that existed before. Genie 3 is not confined to a specific environment. It can generate both photorealistic and imaginary worlds, and everything in between”, says DeepMind researcher Shlomi Fruchter to TechCrunch.
Interactive World
The virtual worlds move with you. With every step you take, the model expands the world in real-time. You can use text prompts to specify what you want to add to the world or which environment you want to travel to. Genie 3 remembers every change and action you take, so they remain visible when you return. The model is designed to first “look back” at previous steps before it can generate the next step.
Genie 3 generates images at a resolution of 720p and 24 fps. Simulations are limited to a few minutes, but this is still a step forward compared to current image generation models.
Move 37
According to Google, the potential applications of Genie 3 are numerous. Think of gaming in virtual reality without needing a specialized headset, but it goes much further. Google is convinced that the simulations will also be useful for scientific research, education, and developing digital twins for agriculture and manufacturing.
AI agents are trendy, and thus Google emphasizes that companies can prepare their agents for the real world through Genie 3. At DeepMind, they even speak of the “Move 37” moment for AI agents, referring to when a Google AI model defeated the world champion in a game of Go.
read also
Sam Altman: “AI Agents will become better than your most experienced employees”
For now, Google is keeping Genie 3 behind closed doors for the general public. A preview is available for research purposes. It is not yet clear if and when Genie 3 will become more widely available.