Google DeepMind revealed on Wednesday the successor to the Genie artificial intelligence (AI) model that could create an endless world of 2D gaming on Wednesday. The new AI model, known as Genie 2, is able to generate a unique controllable 3D environment based on a single image prompt. The company calls Genie 2 an AI “world model” that means it can generate environments that last up to one minute with consistent objects. The company said these generated worlds can be played by humans or used to train AI agents.
Google DeepMind unveils Genie 2 AI model
In the blog post, the company details the new AI model and its capabilities. Although its predecessor can only generate game worlds for 2D platform games, the Genie 2 AI model can generate 3D worlds and has a consistent model that can interact with. This means that humans or AI agents can walk, run, swim, climb and perform more movements in these environments.
Genie 2’s generation feature enables it to generate routes, buildings, and objects that are not visible in the input image. These elements are designed and rendered from scratch. In addition, the underlying model is able to maintain consistency in these environments. This means that the environment remains the same even when the player leaves an area and returns.
Beyond that, Genie 2 is able to produce different perspectives, such as first person view, isometric view or third person view. Additionally, the user can interact with objects in the generated world and can perform actions such as opening a door, breaking a balloon, or climbing a ladder. The model can also prompt the production of physically related effects such as water ripples, smoke, gravity, directional lighting, reflection, etc.
With an in-depth look at technical details, Genie 2 is an autoregressive potential diffusion model and has been trained in large video datasets. The transformer architecture also includes an autoencoder that can generate these worlds frame by frame.
It is worth noting that DeepMind also released an AI model earlier this year with scalable scalable multi-world proxy or SIMA, which essentially has proxy AI capabilities in the 3D world. The company said Genie 2 is able to provide a unique environment for similar AI agents and train for a variety of real-life lives.
Since the world model can generate unique environments, Google says this will remove the risk of data pollution and allow developers to correctly evaluate the capabilities of AI agents.
