Emily Oberg put her brand's e-commerce photography through an AI image generator, looked at what came back, and threw it out. Writing on her Substack, Emily Loves, the founder of the clothing and lifestyle label Sporty & Rich said she is returning to shooting with a full team — models, photographers, editors — after the generated images came back with what she called "a certain lifelessness." The economics behind the experiment were not subtle: commercial imagery is one of the brand's largest expense lines, and a single shoot, run twice a year, costs more than $200,000.
Her objections were specific rather than atmospheric. The generated images looked flat. The styling was mediocre. The pictures were too perfect in the bad sense of the word. The models had dead eyes and moved in ways that were not natural, and none of it carried feeling or emotion. The AI was asked to produce images of models wearing the latest collection, and it did produce them; it just did not produce what her team produces when the team is real people in a room.
Oberg's explanation is that taste is felt rather than learned. Image models train on data, but they have no personal experience that would shape a view of the world and the objects in it, and they lack the elusive sense that is difficult to put into words at all. Taste is how people signal to each other, she argues, and it stays largely unexplainable even between people. So the smartest system available can absorb everything ever written about taste and style and still not have the sensation itself.
The context makes the experiment sharper than it looks. Sporty & Rich was founded in 2018 and grew out of a moodboard Oberg had been assembling from images she liked. Like Glossier and Goop, it is a business built around an aesthetic — the model that Pinterest, Tumblr and Instagram made commercially viable through the 2010s. Oberg gathered carefully chosen images from scattered sources into a single mood, then converted the mood into a product line organized around one idea about health and wellbeing: health is wealth.
Which is the uncomfortable part. Collecting other people's images and recombining them into a consistent mood is, mechanically, a description of what an image model does. Oberg's claim is that the difference lies entirely in who does the collecting. That may well be true, but the test she ran cannot establish it. One brand, one collection, one round of generation, judged by the founder whose taste is itself the product being sold. Sporty & Rich has a direct commercial interest in the verdict it reached, and a finding in the other direction would have devalued the brand's core asset. This reads less like an experiment than like a founder confirming what the business requires her to believe.
What is missing is any number on the other side of the ledger. We know the human shoot costs over $200,000 and happens twice a year. We do not know what the AI images cost, how many rounds were attempted, which tools were used, or — the number that would actually settle it — whether the generated set sold worse. A brand can legitimately decide this on aesthetics alone. It just means the decision transfers to nobody else.
Oberg ends up arguing identity rather than capability: people with genuinely good taste do not follow trends, they set them, and they do not do it deliberately. They stay themselves and choose what they like rather than what an algorithm suggested. That is a claim about who a person is, not about what a model can do, and it is not the kind of claim a better model in a year would refute.
Her own open question is whether other brands will keep judging human involvement worth the cost. The shape of the answer is already in her numbers. Roughly $400,000 a year is affordable for a label whose entire equity is curation and prohibitive for one whose equity is price. The split will not fall between brands that think AI can do taste and brands that think it cannot. It will fall between brands that can charge a premium for the difference and brands that were never able to.