Generative AI developers at Google’s DeepMind are taking on the world’s highest-grossing entertainment industry with the launch of a product called “Genie.” A new artificial intelligence system from Google researchers can create playable worlds similar to those in video games based on images and text prompts. Genie has not yet opened to the public, it was announced in a newspaper on Friday.
What training did Sora receive? Creative people want answers.
Similar to Sora, Genie’s creators call it a “world model,” but unlike Sora, it’s an “actionable-controllable world model.” Trained on over 200,000 hours of 2D platforming games publicly available on the internet, Genie takes tips, sketches, and other images and creates virtual worlds for users to play in – creating assets from scratch and generating pixels based on player actions.
Tweet may have been deleted
What’s most impressive about Genie is the AI’s understanding of physics through hundreds of hours of unsupervised training. This allows Genie to understand different aspects of game mechanics, such as player controls, actions and movement. In addition to developing 2D platform games, it could also potentially play a role in the field of robotics, helping to train robots how to navigate their environment.
While Genie’s high image quality and beautiful presentation are nowhere near Sora’s, it’s an impressive generative AI. It also raises questions about the future job security of video game developers, as well as concerns about gaming marketplaces like Steam, which are already overwhelmed by junk games and asset flipping.