JESUS LEADS THE SEARCH IN THE LATENT SPACE
Artificial intelligence is used to show how the appearance of Jesus is perceived in our modern world. It reveals our collective fantasies and the social construction of our perception.
We try to get as close as possible to an image corresponding to Jesus. The latent space of a GAN is explored with OpenAI's CLIP.
A GENETIC APPROACH
A StyleGAN model trained on faces (FFHQ) generates hundreds of images. CLIP performs zero-shot prediction and returns the image that corresponds the most to the text "Jesus". The corresponding latent vector is kept.
The same process is repeated at each iteration by mutating the best latent vector into several similar vectors.
OTHER RESULTS
"Satan" at different iterations
"Personification of God"
This project was published in January 2021. It is one of the first artistic works using CLIP. A study of the sociocultural bias of CLIP can be found on my Twitter.