WHAMM! Real-time World Modelling — Microsoft Research

Free

Microsoft Research project on real-time world modelling of interactive environments, foundational to AI gaming experiences.

GamesFreeFree tier

#copilot#research#world-model#WHAMM#AI-gaming

Type

Saas

Company

Microsoft Research

Visit Website

About WHAMM! Real-time World Modelling — Microsoft Research

WHAMM (World and Human Action MaskGIT Model) is a real-time world modeling system from Microsoft Research that extends the Muse family of world models for video games. It generates interactive video gameplay at over 10 frames per second in response to keyboard or controller inputs, enabling playable AI-driven game environments. WHAMM was trained on just one week of curated gameplay data (compared to seven years for earlier WHAM models) and has been successfully transferred to the fast-paced first-person shooter Quake II, as well as previously demoed on Bleeding Edge. The model is available as an interactive experience in Copilot Labs, allowing users to explore real-time AI-generated game worlds.

Key Features

Generates video at over 10 frames per second in real time

Interactive playable model responding to keyboard and controller inputs

Part of Microsoft's Muse family of world models for video games

Requires only one week of curated gameplay data for training (down from 7 years)

Successfully transferred to fast-paced games like Quake II and Bleeding Edge

Available as an interactive experience in Copilot Labs

Pros & Cons

Pros

Enables real-time interaction with AI-generated game worlds
Dramatically reduced training data needs compared to previous models
Backed by Microsoft Research and integrated with Copilot Labs
Demonstrates transferability to different game genres (first-person shooter, multiplayer brawler)

Cons

Currently limited to specific games (Quake II and Bleeding Edge)
Still a research project; general availability and broader game support not announced
Requires substantial computational resources for real-time generation

Best For

Interactive AI-driven gameplay experiencesReal-time world modelling for video gamesResearch into generative world models and human action predictionGame design prototyping and testing

Alternatives to WHAMM! Real-time World Modelling — Microsoft Research

G3DAI {Jedi}

Rapidly develop game 3D models, generate detailed landscapes from text, and craft exceptional interior design assets.

Joypanner

Encourages children to explore with tailored AI activities and games for problem-solving and creativity.

ImgenArena.ai

A fun platform for evaluating the quality of AI-generated images. Vote for the best match between prompts and images

Chatmate AI

Chat directly with AIs that look very human in your messaging applications

Hidden Door

An AI-driven RPG adventure game that generates creative stories

NVIDIA RTX Remix

Remaster retro games with ray tracing and AI-enhanced textures effortlessly

FAQ

What does WHAMM stand for?

World and Human Action MaskGIT Model (pronounced WHAM; the M is silent, as an intentional joke).

How fast is WHAMM compared to the earlier WHAM model?

WHAMM generates images at over 10 frames per second, enabling real-time video, while WHAM-1.6B could generate about 1 image per second.

How was the training data reduced so significantly?

By using more intentional data collection and curation, only one week of gameplay data was required for WHAMM training, compared to seven years for WHAM-1.6B.

Which games can WHAMM model?

WHAMM has been trained on Quake II (a fast-paced first-person shooter) and previously on Bleeding Edge (a multiplayer brawler).

How can I try WHAMM?

An interactive real-time gameplay experience powered by WHAMM is available in Copilot Labs.