2026
generative video
Dirty Window continues untitled [ ] and works with Google Street View. The sites are places held in my memory. A memory names a place, and that place has been photographed for years by a machine that was not there for the memory and knows nothing of it. I train a model on that record to look back onto the site that trained it.
The model is trained from scratch on a closed set of Street View images from a single memory-site, with no visual reference outside that site’s record. Street View has photographed the site repeatedly over years. I gather images from this record and approve them one by one. I keep only those that belong to the place as I hold it. What remains becomes the entire training set.
Through training, qualities that recur across the images begin to form a texture of the site. This texture is made of light, angles, weather, surfaces, and traces of the camera’s passage. The many times the camera has passed through the site coexist within it.
Each Street View frame from the site is then refracted through this texture. I layer the model’s output over the original frame. The frame is seen through the composite rather than replaced by it. The model learns the site’s details and the camera’s flaws together, and the blurs, seams, and repetitions of automated capture return as part of the place. The window is dirty because it is made of the site’s own residue.
still from generative video
