Google DeepMind has introduced its latest generative model, Genie, setting a new standard in the world of video game creation.
This innovative AI model heralds a significant leap forward, demonstrating the ability to generate playable video games from mere text descriptions, hand-drawn sketches, or photos reminiscent of classic 2D platformers such as Super Mario Bros.
However, these creations come with a twist: the games operate at a singular frame per second, a stark contrast to the 30 to 60 fps typical of contemporary video games.
Our new AI model Genie can create playable worlds in the style of 2D platformers – all from a single image prompt, sketch or text description. ✏️
As a foundation world model, Genie could also help us train AI agents.
Here’s how. ↓ https://t.co/UX6oEkhkgi
— Google DeepMind (@GoogleDeepMind) April 2, 2024
Matthew Guzdial, a pioneering AI researcher from the University of Alberta, expressed admiration for the project, acknowledging its potential to redefine game development.
Guzdial, familiar with the terrain through his previous work on a similar game generator, recognizes the novelty of Genie’s approach. Genie’s training involved over 30,000 hours of video footage from a plethora of 2D platform games sourced from the internet, a method that, while not unprecedented, is uniquely applied in this context.
People seemed to like the new idea:
Such a cool concept! Impressive!
— Adolfo Asorlin 👨🏼🚀 (@adolfoasorlin) April 2, 2024
Contrasting with prior models such as Nvidia’s GameGAN, which required labor-intensive tagging of video footage with input actions, Genie simplifies the process by learning directly from video footage.
This technique not only streamlines training but opens up vast amounts of existing online video for potential data use.
Genie distinguishes itself by generating new game frames based on player actions, such as jumping or moving left, thereby creating a dynamic, interactive experience from scratch.
Most of the reviews on the internet are positive:
I guess in future there maybe a time where we can make Call of duty style game on snap of finger.
— Pranav (@pranavishnoi) April 2, 2024
The potential for Genie extends beyond its current capabilities, with promises of future versions achieving standard gaming speeds. Tim Rocktäschel, a lead research scientist at Google DeepMind, highlights the technological affinities between Genie and advanced large language models, hinting at rapid future advancements in processing and efficiency.
While Genie remains a research project within Google DeepMind and is not slated for public release, its implications for both game creation and AI development are vast.
People do have some controversial remarks to make:
I don’t have the slightest interest in touching anything AI from Google at this point. Maybe that’ll change, but there are many other significant offerings to explore.
— paliaskepsi (@paliaskepsi) April 2, 2024
The model’s capacity to replicate visual features typical of 2D platformers, such as parallax effects, underscores its sophistication and adaptability. Beyond gaming, Genie’s technology shows promise in robotics and other areas where understanding and replicating physical actions from visual data can be transformative.
Google DeepMind’s exploration into AI-generated virtual environments and task-solving bots, as exemplified by projects like XLand, further underscores the broad potential of AI in creating complex, interactive systems that can learn and adapt through reinforcement learning.
As AI continues to evolve, projects like Genie not only showcase the current state of technological achievement but also open new horizons for creativity, problem-solving, and the way we think about the integration of artificial intelligence in our lives.
For more of such news, visit our AI news at allaboutai.com.
