PROJECT 02 / 2026 / ACTIVE / NYC
Convergent Grammar
A computational study of latent visual grammar across portraiture and adjacent image regimes.
01 / QUESTION
Do images share a grammar?
Convergent Grammar began with a narrow question about portraiture. Across centuries and cultures, across media, do painted portraits share an underlying spatial grammar, a set of compositional rules that show up regardless of artist, period, or style?
The question expanded quickly. If a grammar exists for portraits, what about landscapes, interiors, or garments? What about photographs? What about images generated by machines? The project now runs across modes of image making, looking for compositional structure that survives translation between them.
PORTRAIT / EXTRACTED GRAMMAR EXAMPLE
02 / METHOD
A pipeline for reading images as structure.
The pipeline ingests images from open collections, classifies each by genre, and extracts a spatial grammar, the geometric relationships between figures, horizons, focal points, and negative space. Grammars are stored as vector primitives and compared across the corpus.
The project’s finding so far: compositional grammar is shaped by the spatial problem a genre was built to solve, not by subject matter. A portrait grammar and a landscape grammar diverge because they answer different questions. A portrait grammar and a figure-study grammar converge because they answer the same one.
03 / SCOPE
What the project looks at now.
The current active scans cover painted portraits, painted landscapes, garments, and interiors. Photographic portraiture and machine-generated portraiture are being added as a longitudinal follow-up, tracking how portrait grammar shifts across regimes of image production.
The project is in its research phase. Outputs will be released as diagrams and, eventually, as a paper on the vertical longitudinal study.
STUDY IMAGES
