Leroi.
charles — Evolutionary Systematics for Art
What if a collection of artworks could organise itself—not into fixed style labels, but into families, mutations and shared visual ancestry?
charles takes an unstructured collection of images and explores how an evolutionary systematics for art might be reconstructed from the works themselves. Visual similarity becomes evidence of kinship, divergence becomes mutation and conflicting signals remain visible as hybrid lines. The aim is not another classifier or numbered cluster map, but a readable hypothesis about how a visual field may have developed.
Every feature is calculated inside charles. A DINOv2 backbone describes the works without relying on captions or inherited embeddings, while colour, composition, texture, motif and overall form remain separate feature families. Switching the lens can therefore reveal that two works are related by palette but belong to very different compositional lineages. Metadata such as date, medium or provenance is introduced later as a way to test the result rather than predetermine it.
Art rarely follows a clean family tree: references cross between artists, styles converge independently and several influences can coexist in a single work. charles therefore treats a phylogenetic network as the honest default and resolves tree-like regions only where the signal supports it. Confidence, contradictions and alternative placements are part of the reading instead of being hidden behind a definitive diagram.
The experience begins with an unordered image cosmos. As the analysis progresses, related works draw together, islands and lineages emerge and the same collection can morph between network and radial tree views. Semantic zoom moves from the whole corpus to individual works, while the prototype is intended to test future uses in curation, artist scouting and the discovery of gaps inside a portfolio.