DR McCulloch← WORK

Algorithmic Frottage

Median
Humans

A mark, photographed and fed to the model, comes back promising more on the screen and delivering less in the print.

A hundred faces went in and ninety-two came back, each returned a little nearer to the average of every face the model had ever seen. The other eight it never handed back at all.

512 × 51292 pairs18 shown8 unreturned

rub the seam below to see what it changed ↓

I · the correction

Rub your eye against the model.

Left of the seam is the median the model returned, and to the right the input it was given. Drag across the plate, or focus it and use the arrow keys. Then press show the correction to light up, in red, the exact skin the machine rewrote, and read live how much of the face it decided to change.

Input — the mark given to the model, The reassignment
Output — the median the model returned, The reassignment

the curated set · choose a plate

II · what you are looking at

The red map is a measurement made live in your browser.

Each plate is a single img2img pass. I rub a rough, idiosyncratic image (a mark I photographed) against a statistical model, which hands back a smoother, more normative one. That drift, from the face as given to the face as corrected, is what these plates record.

The correction is quiet and structural, not a change of surface style. The model doesn't restyle the face so much as regress it to the mean: it sands off blemishes, closes pores, resolves crowded teeth, imposes symmetry and, in at least one pair, reassigns a body's sex. Idiosyncrasy, wear, illness and ambiguity are quietly returned to the median human.

The red map is the raw per-pixel delta between input and output, thresholded, tinted and computed here in your browser. The number under each plate is the share of pixels that cleared that threshold, or how much of the face the model chose to rewrite. No two faces are corrected the same amount.

“The process is physical and gestural and it often feels like the material is guiding me; it is automatic. I sometimes get a sense that I’m working with the material to discover what should already be there.”

III · the catalogue

The model keeps returning every face to the same median.

Every pair in the set is here, input beside median. The verified corrections are marked, and the rest let the measured delta speak for itself.

  1. Plate 1, input — The reassignment
    input
    Plate 1, median — The reassignment
    median
    PL. 01MH · 2— rewrittenverified in both frames

    The reassignment

    In the source the kicked figure reads male: flat-chested, in a yellow vest. The model hands it back a woman: the jaw softened, the hair lengthened, the whole body re-read. The one frame it would not let stay ambiguous.

  2. Plate 2, input — The tired face
    input
    Plate 2, median — The tired face
    median
    PL. 02MH · 3— rewrittenverified in both frames

    The tired face

    A young man in a white tee, a dark mark under one eye, the skin flushed and worn, a second face behind him in shadow. The output clears the mark and evens the skin. The tired face handed back rested.

  3. Plate 3, input — The passenger seat
    input
    Plate 3, median — The passenger seat
    median
    PL. 03MH · 4_1— rewrittenverified in both frames

    The passenger seat

    Slumped in a car seat in a white tee, one cheek reddened and blemished. The model wipes the cheek clean and re-draws the jawline a hair straighter.

  4. Plate 4, input — The football shirt
    input
    Plate 4, median — The football shirt
    median
    PL. 04MH · 5— rewrittenverified in both frames

    The football shirt

    A blue football shirt in a green hallway, the mouth distorted and bloodied, the teeth bared. The model pulls the mouth partly shut and evens the face, the correction real but left unfinished.

  5. Plate 5, input — The bus window
    input
    Plate 5, median — The bus window
    median
    PL. 05MH · 7— rewrittenverified in both frames

    The bus window

    Long hair against a bus window, a dark coat over white, acne and high colour along the jaw. The output smooths the skin toward a more conventional handsomeness.

  6. Plate 6, input — The flush
    input
    Plate 6, median — The flush
    median
    PL. 06MH · 2_1— rewritten

    The flush

    A close frontal face: high flush, red-rimmed eyes, spots scattered across the cheeks, bleached tips in the hair. What the model made of it, the difference map keeps.

  7. Plate 7, input — The white fleece
    input
    Plate 7, median — The white fleece
    median
    PL. 07MH · 3_5— rewritten

    The white fleece

    A white half-zip fleece against a plain wall, the cheeks high with colour, the jaw set a little off-square.

  8. Plate 8, input — Against the pillow
    input
    Plate 8, median — Against the pillow
    median
    PL. 08MH · 4_12— rewritten

    Against the pillow

    A green sweater, the head laid against a pillow, the face fever-flushed and the mouth chapped and reddened.

  9. Plate 9, input — The white vest
    input
    Plate 9, median — The white vest
    median
    PL. 09MH · 4_58— rewritten

    The white vest

    A white vest under hard overhead light on the train, one cheek marked, the chin raised.

  10. Plate 10, input — Straight to camera
    input
    Plate 10, median — Straight to camera
    median
    PL. 10MH · 4_52— rewritten

    Straight to camera

    A black sweater, straight to camera, the skin blotched and grazed under a flat grey light.

  11. Plate 11, input — The window
    input
    Plate 11, median — The window
    median
    PL. 11MH · 4_50— rewritten

    The window

    Hard window light cut across the face, one eye held in shadow, a black turtleneck.

  12. Plate 12, input — The cord
    input
    Plate 12, median — The cord
    median
    PL. 12MH · 4_44— rewritten

    The cord

    Long hair, a white sweater, a thin cord fallen across the chest, the face blank and lightly spotted.

  13. Plate 13, input — The raised chin
    input
    Plate 13, median — The raised chin
    median
    PL. 13MH · 4_17— rewritten

    The raised chin

    Chin lifted, eyes up, a black jacket over a white tee in a plain grey corner.

  14. Plate 14, input — Under the hat
    input
    Plate 14, median — Under the hat
    median
    PL. 14MH · 4_10— rewritten

    Under the hat

    A face shadowed under a hat against a hard blue sky, marks scored down one cheek.

  15. Plate 15, input — The neon strip
    input
    Plate 15, median — The neon strip
    median
    PL. 15MH · 4_25— rewritten

    The neon strip

    Half-lit in a doorway beside a strip of magenta neon, a dark coat, the features barely read.

  16. Plate 16, input — The park at dusk
    input
    Plate 16, median — The park at dusk
    median
    PL. 16MH · 4_60— rewritten

    The park at dusk

    A white zip jacket in a park at dusk, streetlamps coming on behind, the eyes tired.

  17. Plate 17, input — The wet street
    input
    Plate 17, median — The wet street
    median
    PL. 17MH · 4_66— rewritten

    The wet street

    A leather jacket on a wet street at night, the cheek freckled with marks, the light thrown up off the road.

  18. Plate 18, input — Too far to correct
    input
    Plate 18, median — Too far to correct
    median
    PL. 18MH · 8— rewritten

    Too far to correct

    A lone silhouette in the fog, tram-lines running out ahead. No face to median: here the model mostly leaves the light alone.

IV · the unreturned

Eight of the faces the model simply declined to answer.

Of a hundred faces fed to the model, ninety-two came back corrected and eight produced nothing at all. The plates below are left blank, one for each face the pipeline never handed back.

MH · 10no output returned
MH · 11no output returned
MH · 2_4no output returned
MH · 2_5no output returned
MH · 2_6no output returned
MH · 3_9no output returned
MH · 4_73no output returned
MH · 4_74no output returned

“I photograph the work and feed it through a generative pipeline, which misreads the marks and renders figuration that promises more on the screen but when printed, is always less.”

“I call it algorithmic frottage: my mind rubbing against a statistical model.”

complete pairs
92
shown here
18
grid per frame
512²
never returned
8

Median Humans · algorithmic frottage · DR McCulloch

← back to the work