Walk into any small creative agency in Shoreditch or Peckham this year and you will find the same quiet shift underway. The mood boards are generated, the retouching is assisted, and the concept visuals that used to consume a junior designer’s week now take an afternoon. What has not changed is who decides which image is any good — and that distinction explains most of what is actually happening in commercial creative work right now.
The Capability That Changed Everything
For two years, the reliable tell of an AI-generated image was text. Shop signage read like nonsense, product labels dissolved, and any layout containing words needed rebuilding by hand. The latest generation of image models solved this, and it mattered enormously for commercial work, because a huge proportion of advertising visuals contain readable words — packaging, posters, storefronts, screens.
OpenAI’s newest image model is the clearest example. Ask for a café menu board with a real menu on it, or a magazine cover with a legible cover line, and it comes back correct. That single improvement moved generative imagery from a mood-board tool to something studios use in client-facing comps.
What Studios Actually Generate
Three uses dominate. Concept visualisation comes first — showing a client three directions before committing production budget to one. Localisation is second: a campaign key visual regenerated with different copy for different markets, which used to be a manual layout job per territory. Social variants are third, where the volume demanded by platforms simply exceeds what traditional production can supply.
The commercial access route is unglamorous. Studios rarely deal with AI laboratories directly; the tools they use reach models over APIs, typically through aggregation platforms. A team can call the GPT Image 2 API alongside competing image models through one endpoint with per-image pricing published openly, which is why a five-person studio now has the same generative capability as a network agency.
What Still Gets Done by Hand
Ask any working art director and the list is consistent. Final typography is rebuilt as a proper layer, because generated letterforms drift from licensed brand fonts in ways clients notice. Anything with a real product must use real photography — a generated approximation of a client’s actual bottle or handset is a legal and reputational problem waiting to surface. And the selection itself remains entirely human: the studios producing distinctive work generate in volume and cut ruthlessly, which is the same discipline good photo editors have always applied to a contact sheet.
The Honest Economics
Generation costs pennies per image. The real cost is the curation time, and studios that measured it report the same finding: they are not spending less on creative; they are producing far more options within the same budget and spending the saved production hours on thinking. Whether that produces better work depends entirely on whether the thinking was the bottleneck in the first place.
For London’s creative sector, which has always competed on ideas rather than on production capacity, that trade looks favourable. The studios worried about this technology are the ones whose value was in execution. The ones thriving were never selling execution in the first place.
One practical note for any studio considering this: run your own test before believing anyone else’s benchmark. Take a finished piece from your portfolio, write the brief that would produce it, and count how many attempts it takes to get something you would put in front of a client. Studios report numbers between three and eight, and knowing your own figure tells you far more about whether these tools fit your aesthetic than any comparison chart will.
