New GAN-based models can now recreate complex game environments.
Have a look at GANTheftAuto, a fork of Nvidia's GameGAN which was trained to emulate dynamic game environments. Basically, the neural network studied the original game and then recreated the game world adding some basic movements, physics, and other elements. Previously, researchers used the model to recreate simpler games with GameGAN like Pacman but with the latest research, they decided to generate one of the most complex environments in games.
"GANTheftAuto focuses mainly on the Grand Theft Auto 5 (GTA5) game but contains other environments as well," noted the team.
The visuals are not perfect, of course, but the team continues to improve the model by adding new features. What do you think about the model? How could it be used in actual game development?
You can find the model's source code on GitHub. Also, don't forget to join our new Telegram channel, our Discord, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.