Game development is a huge effort, and the industry has always looked for ways to streamline workflow to save on costs, resources, or manpower. Recently, Microsoft has introduced its latest generative AI model that might sound a little bizarre on paper: generating short clips of gameplay for “gameplay ideation”.
Microsoft Muse AI (WHAM)
The technology that sits what Microsoft calls “Muse” is WHAM, or ‘World and Human Action Model’ in full. It can be used to “generate game visuals, controller actions, or both,” according to the company’s blog post. One of Xbox’s own studios Ninja Theory was called up to the task, and the whole demonstration involves using the real gameplay footage of its game Bleeding Edge, trained many times over so it can respond to controller inputs.
The model is trained on 100 NVIDIA H100 GPUs, which can generate a whopping… 9 seconds of gameplay in a rather abysmal 300 x 180 resolution, with just 10 frames per second. Still, the model allows users to add elements into the “game” halfway – which makes it a sound idea to function as a rapid prototyping tool in the game’s concept phase, hence the “ideation”.
In the far future, Microsoft is envisioning this feature to revive old games that are no longer playable on modern systems due to compatibility reasons. “We are exploring the potential for Muse to take older back catalog games from our studios and optimize them for any device,” wrote Fatima Kardar, Corporate VP of Gaming AI under Xbox. That’s a big contrast from GOG’s approach, which is way more straightforward and perhaps much more practical.
In the near future, however, Microsoft will open up the feature via Copilot Labs for players to try out this technology – if you don’t mind the very primitive performance and the rather smudgy pixels that it currently offers, of course.
Pokdepinion: Interesting use case.
