The gallery's environments were generated using VQGAN and CLIP.
According to the developers, the gallery was created using a combination of VQGAN, a neural network capable of generating images that look similar to others, and CLIP, OpenAI's neural network that is able to determine how well a caption matches an image, as well as Unreal Engine 5. The engine was used to render 360° images at various locations, the images were then fed to the networks with unique prompts, and after that, the AI-generated paintovers were re-projected to their original locations.
You can learn more and download Gallery PsyGan here. Also, don't forget to join our Reddit page and our Telegram channel, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.