Second Nature: Generative AI Textures
Apps & Workflows · Judson Studios · 2024
A pipeline that turns a single photograph into a depth-aware texture, upscaled to 8K and cast into glass. The first product line from a 128-year-old atelier.
Second Nature starts with a photograph. Bark, stone, water, the surface of a river. A model reads the image as a depth map, lifts every pixel into three dimensions, returns the texture at 8K. From there it gets cast into glass. What you end up with is something between a sample and a translation: a piece of nature you can run your hand over. It's also the first product line from Judson Studios in 128 years.
Brief
Judson Studios is one of the oldest stained glass ateliers in the country, founded in 1897 and best known for ecclesiastical work, the kind of cathedral windows that get hand-cut and leaded by people who learned the craft from people who learned it from people. The brief for Second Nature was to develop the studio's first product line, but the real question I was sitting with was whether an AI pipeline could earn its place in a workflow built on patience, without flattening what made the studio worth working with in the first place.

Process
The pipeline starts with a single photograph. Bark, stone, water, anything with a rich surface. Depth estimator models were used to estimate how far each pixel sat from the camera, which gave me a depth map. That gets upscaled to 8K, enough resolution that the eye stops seeing pixels and starts seeing material. From there the depth map is read as a height field and converted into a 3D model, which gets CNC-milled into a sheet of ceramic fiber board. That sheet becomes the negative the glass is cast against in the kiln. The tile that comes out holds the texture of the original photograph, in glass.
.webp)
.webp)
.webp)
.webp)

Watch: Second Nature: Generative AI Textures
Watch: Second Nature: Generative AI Textures
For Judson, Second Nature is the studio's first product line. For the pipeline, it was the first time the model met a kiln.
