GameNGen
DOOM, simulated by a neural model instead of a game engine.
The project
GameNGen is a research system that simulates the game DOOM using a neural model. Its project page demonstrates real-time interaction and studies how a learned model can stand in for a traditional engine's visual output.
Where AI comes in
A diffusion model predicts the next frame using recent frames and player actions. Training includes gameplay from an agent, connecting game interaction data to generative simulation.
For game makers
A concrete example of action-conditioned video generation. Developers can study how history length and error accumulation affect a playable simulation, without assuming it replaces general-purpose game engines.
Go to the source
GameNGen · Research projectResearch paper · arXivOriginal project page checked at the time of curation.
Keep exploring
Explore
Relight: Lantern Expanse
Fly between worlds, explore their surfaces, and relight the way home.
Hollow Crown
A browser-playable pixel adventure, with AI helping build and demonstrate the game.
Wildbrush
A family-made adventure with paintbrush combat and floating islands.