CHROMATICA
WHAT YOU ARE LOOKING AT
Every particle is one real colour, extracted by k-means clustering from the
photograph of one real painting. Nothing here is decorative: no colour has
been chosen, corrected, or invented. There are no cells, rows or columns —
only colour against black.
WHY A COLOUR SITS WHERE IT SITS
Two axes, and both of them are measurements. Which two is the one thing you
choose, and the field morphs between them rather than redrawing, because it
is the same colours being read a second way.
Chronology, the view it opens on. Left to right is the year the
painting is dated to, linearly from the first work to the last. Bottom to top
is L*, its lightness. So the collection is unrolled in time: you can follow a
colour arriving, spreading and thinning across six centuries instead of
seeing it piled into a single cloud. The 19th century stands as a wall and
the 14th as a few sparks — that is what these museums own, shown rather than
evened out into fairness. The strip below the field is the same year axis, so
the wall stands over its own histogram.
Chromatic plane, the second view. Left to right is a*, bottom to top
is b*: the CIE L*a*b* plane itself. Neutrals fall in the middle, saturated
colours reach the rim, hue runs around as angle. Colours that look alike land
near each other because that is what the space means, and the regions in the
cloud are not drawn — they are what the measurement does on its own. Here
lightness is given no axis: against near-black it is already visible as the
particle's own brightness, and a dim cloud is a dark century.
DENSITY
A particle is pushed out from its exact coordinate in proportion to how much
of the shown collection's measured colour falls in that part of the field —
counted in whichever of the two views is on screen, since crowding is only
real where the particles are actually standing. So a heavily used region
swells and thickens, and a rare one stays a few sparks. Area and solidity are the statistic; it is share within what is
shown, not raw count, so changing what is shown changes the shape of the
cloud instead of merely inflating it wherever the museum owns more paintings.
Particle size follows the cluster's share of its own painting, largest
first.
THE WHOLE SPAN, AND THE TIMELAPSE
By default every painting is in the field at once: this is the colour of the
whole collection, 1300 to 1910. The timelapse is a second reading, not the
main one. It weights each painting by how far its date is from the year you
are on, so the field becomes a moving window rather than a bin, and the
window widens where the collection thins — coverage swings about sixtyfold
between decades, and three paintings is not a period's palette, it is three
paintings. The strip along the bottom shows works per year, so the thin
stretches stay visible instead of being flattened. This is a survey of what
four museums happen to hold and have released, not of art history.
FINDING SOMETHING IN IT
Two questions can be put to the field, and both answer the same way: the
works that answer keep a thin ring on the field and are named in a list
beside it, in date order, and Enter walks them one at a time. Nothing is
removed — everything else only dims — so an answer is always read against
the collection it came out of rather than against black.
By name. Type a title or an artist. All the words have to match, so
a second one narrows; case and accents are folded, because four catalogues
do not spell diacritics the same way and nobody types them anyway.
By colour. Open a painting by clicking one of its particles and ask
for colours like that one. Every other painting is ranked by how near its
closest measured colour comes to the one you clicked — straight-line
distance in CIE L*a*b*, lightness included — and the nearest two dozen come
back. This is the question this dataset answers better than any other, and
no amount of typing could reach it: it finds works that have a colour in
common with each other and nothing else, across schools and across
centuries. The list states the colour it was given and how far the last of
the works it returned actually sits from it, in ΔE, since a small number
there means the collection is full of that colour and a large one means it
barely holds it at all.
THE EDGE AT 1910
The field stops around 1910, and it stops rather than looping. That is a
copyright boundary, not an aesthetic one: paintings after it are almost never
in the public domain, so the twentieth century cannot be shown honestly here.
METHOD
SOURCES
Images and metadata come from the open-access programmes of these
collections, all of them CC0. Attribution is provided on every work: click
a particle to see the painting it was measured from and the museum that
holds it.
A handful of paintings are held in more than one of these catalogues.
Where the artist, title and date agree, the work is counted once.
Built by @userluke_ ·
source on GitHub ↗