The Moon is a Mirror

Artist's Text — Oren Eliav This was the first time I worked with synthetic images. I used an early method called GAN (generative adversarial network). Unlike today's text-to-image models, GAN networks operate through internal competition: one part of the system generates an image, while another evaluates whether it looks "real." Through this back and forth, the system teaches itself to produce convincing images. It also operates according to a kind of "genetic" logic: images can merge and evolve, with each visual "offspring" inheriting features from its predecessors. The paintings were arranged in numbered order across the four walls of the gallery, creating a visual progression that followed how images develop inside a GAN. I wrote a poem called "The Moon is a Mirror," composed partly from the titles of the paintings, addressing the viewer directly. I envisioned the installation as a metaphorical neural network, or as a visual equation built from birds and branches, stars, mirrors, and the moon. What will seeing look like in a hundred years? Will we still be able to distinguish between what happens in reality and what is fabricated? I generated images that at first glance looked like photographs of birds among branches. While painting them, that is, observing them closely, I got a sense of what it might feel like to see through non-human eyes. The mind that created these images had never seen a bird and had never held a branch. When painters depict birds, they start with the body, the beak, the wings and feathers; when they depict branches, they usually follow the logic of nature, a main branch splitting into smaller ones. Instead, the model engaged in a kind of "machine impressionism," as if trying to find the most efficient way to arrange small triangular shapes on a flat square to give the impression of reality, exactly what many painters try to do.