Skip to content

GameNGen

A game engine driven entirely by neural models

Visit official site

What GameNGen is

The engine is fully driven by neural models, supporting real-time interaction with complex environments and stable quality over long trajectories.

What you can do with it

  • Simulate a classic game in real time
  • Predict the next frame from past actions and observations

What to watch out for

  • Training data comes from recorded agent play sessions
  • The next frame depends on past actions and observations

Pros & cons

✓ What we like

  • Long trajectories keep quality through conditioned augmentation
  • A single TPU is stated to run the simulation

! What to watch out for

  • Training data comes from recorded agent play sessions
  • The next frame depends on past actions and observations

FAQ

What speed is stated?

Over 20 frames per second on a single TPU.

What is the training process?

An RL agent records play sessions and a diffusion model predicts the next frame.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review