Developed by Microsoft Research in collaboration with game studio Ninja Theory, Muse is a World and Human Action Model (WHAM) – a generative AI model of a video game that can generate game visuals, controller actions, or both. Trained exclusively on the game Bleeding Edge, researchers and game creatives can explore how these model capabilities will have potential to accelerate their creativity in the future.

Muse was trained on more than 1 billion images and controller actions, from the game Bleeding Edge, corresponding to over 7 years of continuous human gameplay.

The WHAM demonstrator app provides a visual interface for interacting with a deployment of the Muse model instance on Azure AI Foundry. Creators can load a screenshot from Bleeding Edge as an initial prompt, then use the model to generate multiple potential continuations of gameplay from this starting point. They can then explore the generated sequences and tweak them, such as changing the controller inputs or pasting game elements into the scene and predicting what will happen as a result. These features demonstrate how Muse’s capabilities could someday enable AI-supported iteration and brainstorming as part of the creative process.