Google’s artificial intelligence that creates interactive worlds from text + video

Google With introduction Project Geniehas taken another long step in the path of achieving artificial comprehensive intelligence (AGI). This artificial intelligence-based tool, which is available to users with an AI Ultra subscription in the US, allows users to write just a few lines of text. Interactive and experiential worlds to create

Genie 3 is actually a general-purpose “world model” that simulates diverse and interactive environments. Unlike previous Google DeepMind models that were designed for specific environments such as chess or the Go game, this model is built to understand the diversity of the real world and predict how the environment will evolve based on user actions.

Building interactive worlds with Google’s new artificial intelligence tool

In this research prototype, you first describe your target environment; For example, you specify how you plan to explore the world (walking, flying, driving, etc.) and whether your perspective is first-person or third-person.

After determining the character (human, animal or even an object), the model Nano Banana Pro It provides a default image of your world. This feature allows you to check the appearance of the built world and edit it if necessary before fully entering it. Once approved, you’ll enter a 60-second experience by selecting the Create world button.

Quality built worlds 720p And the frame rate is 20 to 24 fps. The amazing thing is that as you move, Genie 3 creates the next paths in real time based on your actions.

Project Genie also has a feature called Remix Worlds that allows users to take existing worlds or other people’s works in the gallery and produce a new version of them by changing the prompts. It is also possible to download videos of these worlds.

To show the power of this model, Google has published several videos based on text commands (Prompts).

However, Google has explicitly pointed out some limitations of its model. Including that the generated environments may not always be completely realistic or not follow the laws of physics 100%. In addition, controlling the characters is sometimes difficult or associated with latency, and the duration of each session is currently limited to 60 seconds.

Google’s goal in releasing this technology is to better understand how users use global models in AI research. But on a larger scale, this technology is part of Google DeepMind Labs’ mission to achieve AGI.

Currently, the tool is only available to users over the age of 18 in the US with a Google Premium subscription, but the company promises to make it available to more people soon.