DRIFT FACTORY AI
Beauty from entropy
What happens when AI tries to copy itself, over and over? It drifts. It breaks. It makes something new.
Subscribe on YouTubeWhat is model collapse?
AI image models don't copy — they interpret. Each pass compresses the image into an abstract mathematical space, then re-imagines it from that compressed meaning. Like a game of telephone, but visual. Over a hundred rounds, the original message is gone — but what's left is the model's accent.
The image doesn't shatter into noise. It erodes into something coherent but alien — the model's own "attractor," a stable pattern it converges toward regardless of where it started. Text dies first. Faces survive longest. Every model finds a different fixed point. The degradation is the model's autobiography.
Artists have been making beauty from broken signals for 50 years — Alvin Lucier dissolving his voice into room resonance, William Basinski recording tape eating itself, VHS bootlegs building an entire aesthetic from generational loss. We're continuing the lineage with a new machine.
The full story →