Editor choice


Google DeepMind AI generates video games from text

In a groundbreaking development that promises to democratize the world of video game creation, Google DeepMind has unveiled "Genie" – an artificial intelligence system capable of generating entire video games based on simple text instructions or a single image. This revolutionary technology is poised to shatter the barriers that have long restricted game development to highly specialized professionals, ushering in a new era of accessibility and creativity.



Announced on February 26, the Genie project represents a collaborative effort between DeepMind, an Alphabet company at the forefront of artificial intelligence development, and the University of British Columbia in Canada. The tool, aptly named "Genie" – a nod to its ability to grant wishes in the form of interactive gaming environments – is a testament to the remarkable strides being made in the field of generative AI.


At the heart of Genie lies a powerful combination of innovative technologies that enable the creation of video games with unprecedented ease. Boasting an impressive 11 billion parameters, the AI model employs a "tokenizer" system that converts video images into "tokens" – units of information that can be processed and manipulated by the AI. This transformation allows Genie to reduce complex visual data into simpler elements, enabling it to study and understand action sequences within video games.


Moreover, Genie relies on a hidden action model that allows it to analyze images and comprehend the potential actions that may occur within a given context. This cutting-edge model empowers Genie to generate action sequences for the games it creates, breathing life into the virtual worlds it constructs.


Together, these technologies endow Genie with the remarkable ability to create entire game environments – complete with background images, scenery elements, game objects, and characters – based on a single textual prompt or image. The implications of such a capability are far-reaching, opening up a world of possibilities for aspiring game developers, educators, and creative minds alike.


Genie's training process is equally impressive. The AI system was initially trained on a staggering 200,000 hours of video footage from classic 2D video games, sourced from publicly available resources. To ensure relevance and quality, this vast dataset was filtered down to 30,000 hours of carefully curated video material. Remarkably, Genie's training took place without direct human supervision, yet the AI managed to grasp the fundamental principles of video games, comprehending how characters move, jump, and interact with their virtual environments.


While Genie's current output – featuring a resolution of 160 x 90 pixels and a frame rate of 10 frames per second – may fall short of modern industry standards, it is important to note that this is merely a research project in its nascent stages. As the technology continues to evolve and mature, it holds the promise of revolutionizing the video game creation process, empowering individuals with limited programming expertise to bring their creative visions to life.


The implications of Genie extend far beyond the realm of entertainment. By democratizing the creation of interactive environments, this groundbreaking AI system could pave the way for new educational tools, immersive training simulations, and even innovative approaches to problem-solving and data visualization.


As the world continues to marvel at the rapid advancements in generative AI, with tools capable of writing texts, creating images, and generating audio and video content, Google DeepMind's Genie project stands as a shining example of how artificial intelligence is pushing the boundaries of what was once thought possible. With the power to create entire video games on demand, Genie promises to unlock a new era of creativity, accessibility, and innovation in the ever-evolving digital landscape. 

However, it is worth noting that Genie is not yet a finished product. At this stage, it's just a research project.

Share with friends:

Write and read comments can only authorized users