Marimekko sent SS27 out in blue, white, pink and red, green in the accents, no black anywhere in its top shades. Season-wide, blue and pink together took 20 percent of Copenhagen’s palette. Read that as a merchandising decision rather than a mood: saturated colour is the cheapest proof left that a person chose the rack, and it is cheap only because the tools drafting the rest of the season’s imagery will not pick it unprompted.

The count comes from an AI-assisted tagging pass across eleven designers, published anonymously in The Data Fashion Brief on 17 August, then within days under Carmen Martinez Ferrer’s byline in FashionUnited. Stine Goya put 17 percent of its own output in green against a season average of 4 percent. Caro Editions ran pink, white and blue. Collina Strada layered a sheer multi-panelled tunic in pastel pink, green and blue over lace and distressed denim cargo shorts. A classifier sorting by garment type, fit, fabric and colour has just finished counting the one attribute machines are worst at proposing.

The bias is documented, though not in fashion. Farzan Farnia, Mohammad Jalali and Azim Ospanov compared generated samples with real test data across benchmark image datasets in February, and found the real data scoring consistently higher on the Vendi and RKE diversity measures. Their account of why is structural: a finite training sample already understates the diversity of the real distribution, so a generator trained to match that sample inherits the shortfall. None of that work concerns knitwear. But a merchandising tool fed four seasons of quiet-luxury product photography will drift the same way, toward the centre of what it was shown.

Ask a generator for elevated minimal knitwear and it returns the mean of every elevated minimal knitwear it has seen, which is beige.

Copenhagen did not abandon neutrals. It made them expensive. The season’s vanilla story arrived loaded with handwork: By Malene Birger’s all-over shaggy fringe on a strappy camisole, Henrik Vibskov’s shredded and frayed panels on a tiered midi, Mr. Larkin’s collarless jacquard tunic closed with frogging. Each of those looks pays for its authorship in labour. Fringe, fray and jacquard are slow, and slowness photographs as intent.

Marimekko has run this argument before. Armi Ratia had banned florals from the house on the grounds that no print could hold the truth of a real flower; in 1964 Maija Isola designed a full collection of them anyway, and Ratia put Unikko into the line. Isola’s answer to the fidelity problem was to drop fidelity altogether — flat oversized petals in red and pink, dark blue stems, nothing resembling an actual poppy. Sixty-two years on, the same house walks Copenhagen with no black in its top shades. The counter to a good reproduction has never been a better one.

The consensus reading of this season is that quiet luxury lost its customer. That’s only part of it. Brunello Cucinelli closed the first half of 2026 at 749.4 million euros, up 9.5 percent as reported and 13.3 percent at constant currency; Kering, over the same period, only just returned to growth, up 1 percent after twelve straight quarters of decline. The gap between them is worth sitting with: the customer buying Cucinelli, Loro Piana or The Row was never buying beige as a trend, so a colour swing can’t touch what wasn’t trend-driven to begin with.

Beige still sells. What it no longer does is identify anyone. When the fast-fashion floor and a diffusion model both produce the same oatmeal, an oatmeal rack cannot name its author. Colour is what a buyer reaches for when the problem is attribution rather than demand.

There is a bill attached. Colour is the attribute a catalogue describes worst, which is why metadata rather than model quality keeps turning out to be the bottleneck in the try-on and outfit systems we have looked at. Copenhagen’s street style wore its turquoise in transparent linen and cotton; saturated colour on a sheer ground is hard to photograph honestly, and among the easier things to send back. Buyers locking autumn assortments this month are paying for a signal that is cheap on the design floor and expensive in the warehouse. That is what makes it a signal.