Microsoft has showcased a new generative AI model called Muse, which has been developed by Microsoft Game Studios in collaboration with Ninja Theory. The plan is to release the model in open source to help others explore how to use it in various situations for development and research.
The demo that Microsoft showcased included generated gameplay. The company says it was trained using Ninja Theory’s 2020 shooter, Bleeding Edge. The demo claims that the user guided the Muse AI model, and the gameplay was generated on the fly.
The blog post includes a bunch of clips if you would like to see them in action. They are low resolution, low frame rate, and quite messy. Microsoft says the model currently renders at 300 x 180, but this is only the training resolution used to speed up the process. However, there’s clearly some potential here.
昔のハードでしか動かない名作ゲームが「AIの力」で現行機でも遊べるように?Xboxがゲームに特化した生成AIモデル「Muse」の機能を披露https://t.co/uZc6VN2Mpx
マルチプレイ対応のバトルアリーナゲーム『Bleeding Edge』をもとに研究、公正性や信頼性など6つのAI原則に基づいて開発が進められる pic.twitter.com/OGq5YozShf
— 電ファミニコゲーマー (@denfaminicogame) February 19, 2025
“I’m incredibly proud of our teams and the milestone we have achieved, not only by showing the rich structure of the game world that a model like Muse can learn, as you see in the video demo below, but also, and even more importantly, by demonstrating how to develop research insights to support creative uses of generative AI models.”
The post explains how the model went from 10k training updates to 1 million training updates. It started with the ability to move around the world, and over time, the model was able to produce flying mechanics. Of course, this took over 1 million training updates to get to and was conditioned on 1 second of gameplay.
Those of you who understand AI training and generative AI tech will enjoy reading through the process. If it doesn’t interest you, basically, 1 second of real gameplay was used on an AI model over a million times to teach it how to generate its own gameplay without the need for source material. Microsoft claims that at its current milestone, the team behind the project finds it difficult to tell the actual gameplay from Bleeding Edge apart from the generated AI videos.
Source: Microsoft