Member-only story
Microsoft Introduces Generative AI for Gaming with Project WHAM
Microsoft Research has unveiled a groundbreaking development in gaming AI technology with the introduction of WHAM (World and Human Action Model), detailed in a new paper published in Nature. This generative AI system, named Muse, represents a significant step forward in AI-powered game development, capable of generating both game visuals and controller actions from minimal input.
What is WHAM?
WHAM is a generative AI model that can create game visuals and controller actions based on brief gameplay sequences. The system developed through collaboration between Microsoft Research, Game Intelligence, and Teachable AI Experiences (TYAX) teams alongside Xbox Game Studios’ Ninja Theory, requires only 10 initial frames (approximately one second) of human gameplay to generate extended gameplay sequences.
Technical Development and Training
The model was trained using data from Ninja Theory’s online game Bleeding Edge, collecting over seven years’ worth of continuous human gameplay data, totaling more than 1 billion images and controller actions. The training process evolved from initial V100 GPU clusters to eventually utilizing H100s, allowing for significant scaling of the model’s…
