Captura-de-tela-2025-02-19-221300.png

Microsoft Testa would help create games

 

The data used by the team to form the “Muse” model comes from a closed partnership in 2022 with Theory, from Xbox Game Studios

Cutouts with examples of technology use
The AI ​​model has been trained in internal hackaton (image: reproduction / microsoft)

 

Microsoft published the details of the search on The first general model of artificial intelligence of a video game which can generate game visuals, control actions or both. Technology is known as the name World and human action model (Wham) and was baptized “Muse».

The company says that focusing on capacity What models like this must effectively support human developers in the creation of games – Signaling that It is not planned to replace real breeders with the new tool. The results are in an article published in the scientific journal Nature.

The resolution is 300 × 180, still below 1080p and 4K in current games (image: reading / Microsoft)

Find out more:

Microsoft took years to finish research

  • The data used by the team to form the AI ​​model comes Partnership closed in 2022 with theory, of Xbox game studioswhose headquarters are in Cambridge (United Kingdom), where Microsoft employees are also participating in the project;
  • Were collected Information on the gameplay of bleeding edgesXbox 2020 game;
  • “Bleeding Edge” is an online match four against four in which All matches are recorded if the player agrees with the terms;
  • From there, the company has chosen to hackathon (Event that brings together professionals to develop innovative solutions in a short time)Explore new interaction and creative use paradigms that the muse could unlock»»
  • The result was a prototype called Wham demonstratorOpen for interactions with other users.
Microsoft logo in a smartphone; Behind, screens with graphics with actions
Microsoft has collected data on the SAIDING EDGE gameplay (Image: Rokas Tenys / Shutterstock)

One of the main improvements came with Higher resolution image coders: Until then, the models have generated images of 128 x 128 pixels; Now they are 300 × 180 – Always below 1080p (1920 x 1080)common among PC players, and even less than the 4K (3840 x 2160)which is already reality in many players’ configurations.

One of my favorite model capacities is the way it can be requested with gameplay sequence changes and persist recently produced elements.

For example, in a demonstration, we added a character to the original look of the game. Asking the model the modified look, we can see how the “persisted” model added and generates plausible variants of the way the gameplay sequence could have evolved from this modified starting point.

Katja Hofmann, senior research manager at Microsoft;

 

 

Add a Comment

Your email address will not be published. Required fields are marked *