Step aside OpenAI's Sora, Google's Genie can generate video games from prompts

Researchers at Google DeepMind have introduced “Genie,” an artificial intelligence model capable of generating playable 2D platformer video games from various prompts. Genie represents a significant advancement in AI-powered content creation in the research phase.

Unlike previous models focused solely on world generation, Genie allows users to interact with the created environments. This is achieved by training the model on a massive dataset of publicly available 2D platformer gameplay videos exceeding 200,000 hours. Genie learns to understand character movement, control, and actions within the game world through unsupervised learning.

The model accepts various inputs, including text descriptions, sketches, or existing images. Utilizing this information, Genie generates the necessary game assets and predicts future frames based on user actions, creating a functional and playable experience.

While Genie is not currently available for public use, the underlying technology and its approach to unsupervised learning hold significant potential for various applications. This includes further advancements in AI-powered content generation and the development of more interactive and engaging user experiences.

This comes after OpenAI revealed Sora, an AI video generator with super-realism, last week. I admit, Genie is nowhere near Sora regarding quality, but I believe building a game would be much more complex than building a video. Microsoft admitted that Sora would come to Copilot.

The research team, led by Tim Rocktäschel, emphasizes the importance of the underlying technology and the novel approaches developed during Genie’s creation. These advancements contribute significantly to AI research, paving the way for future innovations.

More here.