Microsoft has unveiled a new AI model named Muse, designed to generate game visuals and controller inputs. This development stems from the company’s training set created using the game Bleeding Edge, a 4 vs. 4 brawler that saw limited success.
Despite its commercial failure, Bleeding Edge provided Microsoft with a unique opportunity. The game’s end-user license agreement allowed the company to record gameplay data, which it utilized to train Muse. The AI model aims to assist game designers in prototyping ideas without the need for extensive programming and asset creation.
Applications of Muse
Muse can significantly streamline the game development process. For instance, a designer at Blizzard could use it to test a new character for Overwatch 2. This approach could save time and reduce development costs, which often escalate during the iterative phase of game creation.
One of the model’s standout features is its capability of persistency. This allows Muse to adapt to user modifications in real-time, enhancing the realism of generated gameplay. For example, the model can simulate player reactions to newly introduced power-ups, creating a more immersive experience.
Future Prospects
Microsoft is already leveraging Muse for creating a real-time playable AI model based on other first-party games. Additionally, the company is exploring how this technology can help revitalize older games that are currently limited by outdated hardware.
While Muse is touted as a groundbreaking generative AI model, it is important to note that world models are not entirely new. The company Decart previously introduced a model named Oasis capable of generating Minecraft levels.
Currently, Muse generates visuals at a low resolution of 300 x 180 pixels and approximately 10 frames per second. Microsoft plans to release Muse’s weights and sample data for researchers to investigate its capabilities further.
For more details, visit the original article on Engadget.