Ecologies of Noise

2025-ongoing

Ecologies of Noise follows images of endangered ecosystems as they are recursively processed through an image-to-image AI system. Each output is fed back as the next input until the landscape transforms into a minimal color field. These final frames are then painted onto discarded phones, tablets, monitors, and TVs, turning screens from interfaces into material supports.

Ecologies of Noise is my latest body of work that brings together media I’ve used over time in my practice, like painting, photography, AI, and coding. It is a research-based project where the final work takes on a sculptural form, by painting the resulting images onto screens that I physically arrange.

I start by gathering photographs of nature that show some of the most endangered ecosystems affected by global warming, which I find online under Creative Commons licenses. Each photo becomes an input to an image-to-image AI generator. The AI produces an output, and my custom-coded software takes that output and feeds it back as the next input, repeating the same loop hundreds of times. Early on, the landscape begins to shift through the process as details drop out and the scene becomes gradually unrecognizable. As iterations progress, the image collapses into a minimal color field. I call these final frames “end images.” These color fields are not abstractions in the modernist sense of purification, they are the exhausted remains of a process that has consumed its referent.

The loop is a direct visualization of the effects of model collapse, a term used in computer science to describe how AI outputs can degrade when they keep feeding on synthetic material, with each new output turning flatter, more repetitive, and drifting back toward the initial state of AI-generated images,  which is noise.

As a second step, I take those “end images” and paint them onto screens, TVs, monitors, tablets, and phones, sourced from a recycling company. Underneath their smooth surfaces sits a material paradox. Screens are often spoken of as dematerialization, yet they are among the most materially dense cultural forms we have: mineral, chemical, logistical, electrical. Devices engineered for refresh, update, and endless substitution are converted into fixed panels carrying a single final frame. I use these screens purely as physical supports in a salvage-based practice as if  in a meta-technological scenario shifting the focus from their use as an interface to their materiality.

By painting a single final image on each screen I am also canceling their functionality and in a metaphorical way, addressing my own feeling of screen fatigue while simultaneously shifting the object’s status from interface to artifact. Our most intimate aesthetic experiences are now brokered by devices whose primary logic is capture, extraction, and circulation, I'm subverting that by using a manual and historically charged way of image production which is painting.

The work nods to the assumption that the more we rely on AI to mediate cultural production, the less powerful it may become because of model collapse, and eventually we may fall into a precarious stage where technology has passed its peak era, our skills have been atrophied and the natural environment has also suffered irreversible damage. What does this vicious cycle of exhaustion look like when cultural technology begins to cannibalize itself at scale?

For more on the research and process, see below.

Research audio

Ecologies of Noise has been a research-based project resulting in many pages of notes, ideas and comments that resulted from my longstanding interest in AI’s impact on cultural flattening as well as the effects of technology’s hardware on the environment. This podcast has been created by feeding NotebookLM with all my research material and then edited down by me.

Oil painting on screens

Ecologies of Noise, Assemblage 1, 2025, oil painting on screens, dimensions : 60 x 50 cm

Ecologies of Noise, Assemblage 2, 2025, oil painting on screens, dimensions : 67 x 61 cm

Ecologies of Noise, Assemblage 3, 2025, oil painting on screens, dimensions : 67 x 53 cm

Ecologies of Noise, Assemblage 4, 2025, oil painting on screens, dimensions : 52 x 60 cm

Ecologies of Noise, Assemblage 5, 2025, oil painting on screens, dimensions : 60 x 52 cm

Ecologies of Noise, Assemblage 6, 2025, oil painting on screens, dimensions : 70 x 53 cm

Ecologies of Noise, Assemblage 7, 2025, oil painting on screens, dimensions : 66 x 93 cm

Other works

Installation proposals

Installation proposal of prints showing the 500 iterations of feedback loops starting from photos of endangered landscapes. Each print is 1 x 5 m.

Installation proposal of prints showing the 500 iterations of feedback loops starting from photos of endangered landscapes. Print dimension 1 x 5 m.

Installation proposal of oil paintings on screens

Installation proposal of looped video of “end images”

Various “End Images”

End images are the kinds of images that are produced during the recursive feedback loop process described above. These images have lost any representational reference to the original input image and become a kind of image-residue. These color fields are not abstractions in the modernist sense of purification, they are the exhausted remains of a process that has consumed its referent.

Research and Process

Input images: photos of endangered ecosystems found online under Creative Commons License

While running the recursive feedback process, an unexpected but consistent thermodynamic color shift in the images emerged. As visual information degrades through successive generations, the images often move from warmer and more complex states towards cooler, darker and increasingly simplified ones. There seems to be a correlation in the generated images between the amount of information they hold and the prevailing color of the images. After a number of feedback loop iterations the images become abstract colour fields that can initially recall modernist abstraction, although they arrive there through a kind of exhaustion and not by pursuing some kind of purity.  These "end images" as I call them are the residue of an image that has lost its connection to its referent through endless statistical averaging. They are the average of the average of the average.  This color shift also touches on Hito Steyerl's "heat images" concept. Steyerl mentions that AI images are heat images because they need energy to be produced, starting from noise, pixels are arranged into structures that we recognize as images, a process that consumes energy and emits heat. My process builds further on this idea showing that as this structure of recognizable forms degrades in a kind of entropic manner, the system "cools down" in color and probably in energy consumption, moving toward it's attractor state. The fact that these experiments nearly always end in very dark or black images also made me think that AI systems, just like any other physical system, also obey the same rules of thermodynamics. 

Example of input image no.6 (from the image above) run by a custom coded feedback loop for a hundred iterations through an image-to-image AI

Example of input image no.9 run by a custom coded feedback loop for a hundred iterations through an image-to-image AI

Example of input image no. 14 run by a custom coded feedback loop for a hundred iterations through an image-to-image AI

Print installation at the group show Collapse: data.models.worlds, curated by Daphne Dragona and Katerina Goutziouli, Athens, Greece

500 recursive iterations starting from the image below