aurèce vettier

The piece I’m presenting is called “a family gathered in a field watching a massive shooting star and its trail of dust above them” (AV-2024-U-387), and is part of a series called “le travail des rêves”. My approach at aurèce vettier combines intimate, carefully curated data with powerful algorithms, projected into the real world using artisanal processes and skills such as painting, bronze sculpture, and tapestry. These highly personal data and models enable me to develop a stable aesthetic that I use to tell all kinds of stories. Since 2018, an important component of my work uses GANs to propose an imaginary nature of near-impossible shapes. In late 2021, I began working with CLIP+GAN models, which allow me to generate images from text.

aurèce vettier, “le travail des rêves”, exhibition at Bigaignon Paris in Feb 2024, as part of Bright Moments – Photo Romain Darnaud

These CLIP+GANs are trained on photos from my childhood to the present. I have a very extensive film archive thanks to my father and my past as a photographer. I use these models to visually represent my dreams and the emotions they bring me, thus achieving a form of introspection. Circular Ruins (2022) was the first exhibition of this practice. For “le travail des rêves”, exhibiting in Bright Moments Paris (2024), I re-trained the model with more recent photos, notably taken with my smartphone. I use this custom text-to-image model to represent and interpret dreams, as a tool for introspection. When I create a prompt based on my dreams, the machine must return an image, a proposal. For some dreams, the prompt only consists of one word. With others, the dream was almost impossible to describe because it was not visual, but composed solely of pure emotion. If I’m not satisfied with the machine’s proposal, I ask for another one. When I feel that the proposed image represents my dream, I make a subjective and intuitive call to stop asking and move on. The surprise comes when this subjective intuition is affirmed in the physical world days or months later, by chance of conversation, eye contact, or some kind of epiphany. I am surprised by the accuracy with which this model represents the impalpable, the intimate. Above all, it shows a filigree of thought that has always been there, in the data set, in my childhood photos up to the present day. le travail des rêves confirms the validity of this process as a method of introspection.

Notre Dame en feu (attempt), Oil painting on canvas from an AI-generated image, 40 cm x 50 cm, AV-2024-U-514

This series captures some of my roughest years as well as absolutely intense, wonderful adventures and encounters. You will find shooting stars, flocks of yellow butterflies sent to me by loved ones from the afterlife, the crossing of vast forests, and objects that are emblematic for me, such as the sacred icon in the basilica on the island of Tinos. When I look at this body of work, it’s the precise story I wanted to tell at this very moment. When the series was finished, I experienced a blissful moment in which I understood that I had nothing to add or remove. As these works leave my studio, the point is no longer to talk about me, but to let others appropriate these visuals, by interpolating and interweaving the present and missing information with fragments of their own lives. I use AI as a catalyst too, or more precisely, as a percolator. AI models have absorbed immense amounts of data and it is possible to produce any kind of image, video, or sound in high fidelity. However, just because you have a box with many colored pencils, doesn’t mean you have to use all those colors in your work. I’m especially enthusiastic about AI’s capacity to digest intimate data and explore latent spaces, rather than the all-out generation of visuals. With le travail des rêves, I am constructing pareidolia –the tendency to perceive a specific, often meaningful image in a random or ambiguous visual pattern– rather than exact representation. The computing power I summon remains low, and the algorithms I use are intentionally old. An upscaling algorithm improves resolution but doesn’t add any detail to the images. With low resolution, it’s impossible to know precisely what these images represent, yet our brains and experiences allow familiar shapes to emerge, in the manner of interpolation.

Recently, I read Simulacres et Simulation by Jean Baudrillard, in which the philosopher describes four successive phases of the image: One, it is the reflection of a profound reality; two, it masks and distorts this reality; three, it masks the absence of a profound reality; and four, it has no relation to reality, becoming its own simulacrum. These AI-generated dreamscapes combine the four phases, perhaps going beyond pareidolia because interpolation is required both from the emitter and observer.

aurèce vettier, “le travail des rêves”, exhibition at Bigaignon Paris in Feb 2024, as part of Bright Moments – Photo Romain Darnaud

The piece I’m presenting for CVPR, called “a family gathered in a field watching a massive shooting star and its trail of dust above them” (AV-2024-U-387): I dreamt this based on a real-life experience after I had just lost a member of my family. She was there, and she was gone at the same time. Now, try to count the family members gathered in the field. It’s impossible; however, we see a family emerging from this soup of pixels—this is a pareidolia. The image is a reflection of something I’ve experienced, yet it inevitably distorts reality. In the end, it may be its own simulacrum.

a family gathered in a field watching a massive shooting star and its trail of dust above them (attempt), Oil painting on canvas from an AI-generated image, 40 cm x 50 cm, AV-2024-U-489