Lost Compressions is a time-based project built from unviewed and low-visibility YouTube videos sourced via Petittube. The work examines how digital memories persist not through intention or care, but through infrastructural inertia——files remain stored because deletion is costly, not because they are remembered.

I process the footage through a stable diffusion model I trained based on corrupted video files. This process physically alters and destabilizes the pixels from the original footage. The algorithm recalculates every component and transforms them into blurred, half-remembered, or perhaps completely reimagined fragments. I try to replicate the way my own memories deteriorate over time: faces lose clarity, details slip away, moments are pushed towards abstraction.

The project reframes digital storage as a stratified system shaped by access, labor, and infrastructure. By simulating memory decay within an environment designed for permanence, Lost Compressions questions who gets to preserve, who gets forgotten, and how, perhaps, uneven archival systems shape what survives.