Microsoft Reveals Muse

From TwogPedia
News/Microsoft Reveals Muse /
Revision as of 14:51, 20 February 2025 by Andura (talk | contribs) (Created page with "{{News |seo_title=Microsoft Reveals Muse |seo_keywords=Microsoft |image=Microsoft.png |tags=Microsoft |date=2025-02-20T14:49:10.000Z |sources={{NewsSource |source=Bloomberg |url=https://www.bloomberg.com/news/articles/2025-02-19/microsoft-unveils-generative-ai-that-can-create-video-game-scenes |article=News/Microsoft Reveals Muse }} |author=Andura |category=Business |content=Microsoft has announced the development of a new generative AI model capa...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Written by Dante Uzel on 20 February 2025 14:49

Microsoft.png

Microsoft


Microsoft has announced the development of a new generative AI model capable of producing "complex gameplay sequences." The company’s Research Game Intelligence team introduced the World and Human Action Model (WHAM), also called Muse, which can generate both game visuals and controller actions, or either one on its own.

However, it’s important to note that this technology isn't actually building full video games just yet. According to Microsoft, Muse was trained using data from Ninja Theory, the developer behind Hellblade, and is currently capable of generating visuals at a resolution of 300 x 180 pixels. The model required 1 million training updates before achieving stable output with correct behavior and mechanics.

More:Scopely to Acquire Niantic's Portfolio

Muse has been trained on over 1 billion images and controller actions, which corresponds to more than seven years of continuous human gameplay. Microsoft emphasized that this data was "ethically" sourced from Bleeding Edge, Ninja Theory's four-player multiplayer game. Previously, Microsoft had used player data from this game for human-like navigation experiments, but in 2022, it began exploring how that data could be used to fuel a machine learning model.

Tags: Microsoft

Sources:
Bloomberg.png
Categories: