Real-time world and human action model
WHAM-RT
A real-time gameplay model that generates visuals in response to controller input.
01 / Overview
Interactive gameplay prediction, controller-conditioned generation, real-time world-and-action modeling.
A real-time gameplay model that generates visuals in response to controller input.
Check input response, visual coherence, and whether the scene remains recognizable during continuous control.
Keep the Quake II technical demo separate from the earlier Bleeding Edge WHAM research release.
02 / Strengths
Where it stands out
- Provides a direct public demonstration of action-conditioned gameplay generation.
- Shows why a world model needs to account for both environment state and player actions.
- Uses a recognizable game domain that makes control failures easy to notice.
03 / Boundaries
What not to overclaim
- The demo is tied to a narrow gameplay domain and should not be generalized to arbitrary games.
- It does not generate a complete editable project, executable game, or production-ready level.
- Copilot Labs eligibility and availability can change.
A constrained gameplay technical demo, not a general text-to-game creator.
04 / Access & output
How the model reaches people
Playable through the official WHAM-RT Copilot Labs link; the earlier WHAM model and demonstrator are available separately through Microsoft research channels.
- Platforms
- Copilot Labs technical demo · Microsoft Research
- Outputs
- Interactive generated gameplay visuals
- Inputs & compatibility
- Controller or game input · Supported Copilot Labs access
05 / Practical path
How to evaluate it
- 01
Open the WHAM-RT page and follow the current Copilot Labs access link.
- 02
Use repeated directional inputs to judge response rather than visual similarity alone.
- 03
Read the WHAM family page before transferring claims from WHAM to WHAM-RT.
Primary evidence