Google DeepMind has released Genie 3, its latest general-purpose world model that can create interactive 3D environments from just a text prompt. “Given a text prompt, Genie 3 can generate dynamic worlds that you can navigate in real time at 24 frames per second, retaining consistency for a few minutes at a resolution of 720p,” reads the DeepMind blog.
A Step Closer to AGI
Powered by decade-old proprietary research in simulated environments, DeepMind has built AI world models that can simulate different real-world scenarios. These models can be used as stepping stones to AGI (Artificial General Intelligence). Unlike Genie 1 and Genie 2, the latest version allows users to interact in real-time.
Applications of Genie 3: History, Animation & Natural World
- Physical World Phenomena: Users can experience real-world properties like water, lighting and other environmental scenarios.
- Natural World Simulation: From rich ecosystems to animal behviors and nuanced plant lives, the model can generate it all.
- Animation & Fiction: Lets users create highly imaginative environments with animated characters and expressive visuals.
- Transcending History & Geography: Users can explore locations and historical places.
Genie 3 Limitations
- Limited Direct Actions: Although many environmental changes can be prompted, the agent itself can only perform a narrow set of direct actions.
- Geographic Realism: Lacks the ability to recreate real-world locations with precise geographic fidelity.
- Text Rendering: Legible text is typically generated only when it’s explicitly included in the input world description.
- Multi-agent Interaction & Simulation: Accurately simulating interactions between multiple independent agents in shared spaces continues to be a challenging area of research.
- Short Interaction Span: The model currently supports only brief interactive sessions lasting a few minutes, rather than prolonged, hour-long engagements.
What’s Next
DeepMind considers Genie 3 “a significant moment for world models”, that will have a ripple effect on generative AI and AI research. The firm is currently exploring implications of the model while continuing to advance it for further applications in education, gaming, and training.
Arko Chandra
