DeepMind has announced Genie 2, an artificial intelligence model capable of creating playable 3D environments from single images and text prompts. The model, unveiled on December 4, 2024, represents an advancement over its predecessor Genie 1, which was limited to 2D worlds.
According to DeepMind, Genie 2 can generate interactive environments that respond to keyboard and mouse inputs, allowing users or AI agents to control characters within the generated worlds. The model demonstrates capabilities including object interactions, character animations, physics simulations, and lighting effects, with environments remaining consistent for up to one minute.
The company positions Genie 2 primarily as a research and prototyping tool, particularly for training and evaluating AI agents. The model was trained on video datasets and uses an autoregressive latent diffusion approach, though DeepMind has not disclosed specific details about the training data sources. While the technology shows promise, current limitations include the brief duration of generated environments and potential intellectual property considerations regarding training data.
Sources: Google DeepMind, TechCrunch