He never sat for this portrait.
What you see is a frontal face held in grey mist, the hair dissolving at its edges into a halo of other haircuts, the eyes not quite matching (each one weighted a fraction off from the other, a ghost of his asymmetry surviving the average). The mouth is closed. The hold is direct. Ziggy, the Thin White Duke, Aladdin Sane, Major Tom, Halloween Jack, Blackstar: none of them is centred, and all of them are here.
Three hundred and ten photographs, each held at equal 1/310 weight, aligned on eye midpoint and mouth and averaged in perceptually uniform colour space. No single image wins. What emerges is a figure that none of them alone contains: Bowie as a sum of how he has been pictured.
The method
This is via parametric authorship (a rule-set declared, a source-stock selected, the software executing what the rule-set demands). The ethical position sits in the 1/N. No source dominates, no persona crowned over the others. Ziggy is not elevated above Blackstar; neither is elevated above a passport-photograph moment between records. A small piece of code does the pressing, which is arguably the least interesting sentence in this paragraph. What matters more is the compression itself, and what survived it: the quiet.
The script is layer-average.py, written in Python on OpenCV, MediaPipe (for face-landmark detection), and NumPy. Landmark alignment is done on eye-midpoint and mouth (the geometry of attention, more or less). The averaging is then done in a perceptually uniform colour space, so what holds the centre is a face that emerges rather than a smear with face-like features in it.
Sometimes you can hear an instrument’s resonance most clearly when nothing in particular is being played on it (the wood, the cavity, the temperament between notes). Averaging is a kind of machine for that. What sits in the average is not a single face. It is the shape the room makes when no one in the room is louder than anyone else.
Bowie as exemplary subject
Bowie is the exemplary subject for this. A man whose constant reinvention means the source stock spans an extraordinary range of visual identities (and the refusal of fixity itself is part of the source). Averaging is a machine for finding what remains when every version is given the same room. With Bowie, the machine finds a face you feel you recognise but that never existed: a holographic character, the version that was always underneath the versions.
The choice here was not the choice of persona. It was the choice of letting them all stand at the same volume and listening for the resonance.
Two scales
Three hundred and ten into one, for the large print (570 mm, unique). One into ten, for the CD jewel-case edition (12 × 12 cm). Many into one, one into many.
The edition keeps a familiar price (R3,333.33, the same figure carried forward from earlier editions in this practice). The large print inaugurates a new one (R33,333.33). Same image at two scales, an old price and a new held alongside each other, a tensor rather than a break. The pricing structure echoes the work’s logic.
The lineage runs back. Layered averaging began as a practice with the Composers I–V series (Resolution Gallery, Johannesburg, 2012) and continued through the bodyTime() double-exposure life-drawing series (2017–). The software is new; the ethic is not. Equal weighting at 1/N has been the through-line for a decade-and-a-half: hand-blended exposures first, then digital compositing, now Python on OpenCV. The artefact changes. What does not change is that no source dominates.
And yet
What the method does not claim: that this is who Bowie was. The work is honest about being a representation of representations, a figure assembled from how he has been pictured. The averaging produces something that did not exist. The tribute is in keeping the not-existing visible.
What the method also does not claim: that the equal-weighting is automatically ethical. Averaging photographs of a real person (even one who is publicly archived) is its own kind of pressing. The dignity-conservation move sits in the 1/N, and the 1/N does its work, and the question does not therefore disappear.
There is the artist-question too. A man making composite portraits of another man does what he can with his looking. The hold is direct, the press is unstaged, but the hand is unmistakably here.
It is posture as much as aesthetic, something kinda queer, kinda kinky, kinda cute among the versions of a man who was never only one thing. And yet the face looks back in a way that always seems to be folding something. It never existed, and it returns a look anyway, which is, kind of, what a tribute is for. It’s also kind of alien, isn’t it, a man recognisably shining across time, like a star, man.
Notes
The work shows at The Viewing Room Art Gallery, St. Lorient, Pretoria (23 May – 15 June 2026), as part of From Ziggy Stardust to Blackstar: A Tribute to the Genius of David Bowie, curated by Gordon Froud. Both pieces are available; during the gallery period and the three months of gallery aftercare that follow, enquiries are best made via the gallery.
Source script: tools/layer-average.py. Lineage: Composers I–V (2012), bodyTime() (2017–).
Concepts referenced
The frameworks this work sits inside (further reading in the README repository):
- Parametric authorship — the operational lens. Meaning held in the mapping from rules to outputs rather than in any single output; the work is the rule-set as much as the print.
- Hyperstrate — the protocolled field in which sovereign peers attest to claims. The 1/N as a quorum, not a queue.
- CARDS — needs construct (competence, autonomy, relatedness, dignity, safety). The dignity-conservation move sits here.
- SRII — axiological framework integrating these principles into the engineering surface.
- Voice signature — the register this prose tries to keep within.