AI
TRASH
Recovered An image-to-image AI method trained on matching pairs. It learns how one visual form maps to another, then attempts that transformation on new inputs. leftovers returned to work as generative image systems.
Leftover machine vision, brought back from the bin.
AI TRASH is made from leftover An image-to-image AI method trained on matching pairs. It learns how one visual form maps to another, then attempts that transformation on new inputs. outputs. In the early years of my AI experimentation, I used pix2pix eagerly and accumulated folders of strange, partial, and failed generations. I found some of these outputs again while cleaning up and reorganizing an old computer. I was about to delete the unnecessary files when I came across the pix2pix leftovers. Luckily, before the folders disappeared, I had the idea to put these discarded outputs back to work. That idea became AI TRASH.
How the trash becomes art
After organizing 1002 discarded pix2pix outputs, I began working with AI to build a A JavaScript library for making visual art and interactive work in a web browser. Here it drives the rules that arrange selected source images into each landscape. project that could give these leftover assets a new function. Through a process of testing and adjusting, the project gradually developed into a system for generating corrupted digital canvases.
The resulting images sit somewhere between failed machine vision, damaged landscapes, and pure visual noise. Some resemble glitched environments or synthetic ruins. Others feel closer to raw corruption and digital trash.
Source Archive
Loading archive...