The board on Canal Street had a button at hand height and a nozzle above it. Press the button and a fine mist of hypochlorous acid landed on your face, in June, on one of the grubbiest pavements in Manhattan. Tower28 sprayed more than 188,000 people over two weeks, and that figure is the argument: once campaign imagery costs almost nothing to generate, the scarce advertising asset is the one that has to physically exist in a single place.

Zalando told Reuters that generative production cut its imagery turnaround from six or eight weeks down to three or four days and reduced costs by 90%, with roughly 70% of its fourth-quarter 2024 editorial campaign images generated rather than shot. H&M said last year it would build digital twins of thirty of its models. What comes out of these pipelines is competent and endless — a woman with good skin, a bottle, light that never fell on anything. Any brand can have that picture by Thursday, which is exactly why it stopped being worth anything.

What no model renders is the queue: strangers standing in the heat on Canal Street, waiting a turn to be sprayed.

Amy Liu has worked in beauty for 24 years and told the Glossy Beauty Podcast she has never seen the category more competitive. Her answer cost around $200,000: $150,000 to the outdoor agency MilkMoney and another $50,000 on promotion, staffing and the surrounding campaign. Against 188,000 sprays that works out to roughly $1.06 a face, and each face got the formula rather than a photograph of it. A four-ounce bottle of SOS Daily Rescue Spray retails at $28. MilkMoney sited the board near Sephora doors and put “available at Sephora” on the artwork, which makes the sample the first step of a very short funnel.

A few blocks away at Sixth and Grand, Tower28 ran a second board with no nozzle on it. Same brand, same summer, no mechanism. The agency’s own case study credits the interactive one with stopping pedestrians who then filmed themselves and shared it.

Fragrance settled this in the early 1980s. Giorgio Beverly Hills, launched in 1981, was one of the first scents advertised with the newly developed strips glued into national magazines, and it worked because the page carried a dose of the product rather than a picture of it. The sample beat the depiction 45 years ago, and nothing since has disproved it. What changed is the other side of the ledger, where depiction now costs nothing, which leaves the sample as the only part of an ad still carrying a price.

Call it experiential out-of-home, a category that usually means a stunt with a hashtag attached. That reading misses the mechanism. The value sits with the witnesses: 188,000 people who can attest the thing happened. A generated clip of a woman being misted on Canal Street takes an afternoon to produce and is worth about that much. The $200,000 bought public certainty that it occurred.

Constraint came back to the artwork, which is the part worth watching. A board carrying a reservoir and a mister has to be laid out around plumbing rather than a grid. Liu describes space limits, storage for the product, and a liability question that forced a location change at the last minute. She kept video of the spray mechanism because she was worried it would fire too hard or not hard enough. Generated imagery meets no resistance of that kind, and the output shows it: centred subjects, weightless product, backgrounds that resolve to nowhere in particular.

The sharpest moment in Liu’s account is the vendor’s advice. Just put water in it, they told her, and no one will know the difference. She refused. The board’s whole value rested on the mist being the actual formula. A convincing substitute would have hollowed it out.