The latent space is a metaphor to understand the possibilities of a GAN to generate images. It is a question of conceiving all the possible images that trained networks can generate as a system of multidimensional coordinates that we could walk. To create this virtual and multidimensional space from real and concrete spaces is a reflexive game about the very same metaphor of latent space.
In this case, the training dataset has been the famous tricycle pedaling through the corridors of the hotel in The Shining (1980, Stanley Kubrick). If these corridors, and the labyrinth outside the hotel, are metaphors of mental labyrinths that the characters of the movie suffer, in this case the network is a kind of virtual double of that same space.
As in a nightmare after watching the film, we pass obsessively through a space that is and is not the filmed one and that at every step unfolds, splits, fork and multiplies in a maze of multiple dimensions from which we can not get out.