A sunless summer in Shangri Lá
Video Art- A.I 2022
In this work, the weather becomes one of the artistic agents, drawing lines between the utopian literary city high up in Tibet, the Brazilian beach on the coast of Rio Grande do Sul, and the mountains of the Yosemite Park in the United States. We operate by tensing the relationship between the natural and the artificial. Amidst grains of sand and grains of pixelated noise, we used cyclic generative adversarial networks (cycle GANs) to build a dreamlike imagery set in search of the sun of Shangri Lá.
The artificial intelligence networks used in the project are trained on images from different seasons at the same location in the Yosemite Park. From this data set, they compete with each other to artificially alter any image with opposite seasons, until one convinces the other of the naturalness of its artificialization.
The model was developed based on images from Yosemite Park in the United States of America. The algorithm is trained with a set of 853 images of the park in summer and 1273 images in winter. From this set, it translates any input image to the opposite season of the year. With this, the definitions of winter and summer, despite their globality and diversity, are established based on local characteristics.
The park is frequently visited by professional and amateur alpinists who venture into climbing, recording the routes in photos, videos, and blog posts. A renowned climber from the region, Brutus of Wyde, claims to have found Shangri Lá on one of his expeditions. Brutus described and even made some drawings of the beautiful High Sierra site, but kept its location a secret.
The story of Shangri Lá, however, is older. It was originally created in 1933 by James Hilton in his novel Lost Horizon. It is an imaginary, mountainous place, supposedly in the region of Tibet, where the inhabitants never grow old as long as they never leave, oscillating between paradise and prison. The literary success inspired the founding of several places, such as the coastal municipality of Xangri Lá.
We confronted the algorithm trained with Yosemite references by presenting images of the beach at Xangri Lá on a cloudy summer day. With no references for sand dunes, the algorithm turns them into snow-capped mountains. This creates the utopian landscape of a subtropical snowy beach. Unknowingly, the system replicates the seasons of the northern hemisphere in the southern hemisphere, creating dreamlike scenes looking for the Shangri-La’s sun.