AI image generation moved from novelty to budget item in fashion faster than almost anything else I have worked with. The conversation about it is almost always a conversation about cost, which is understandable and still beside the point. Treat the technology as a saving and you get pictures that are cheaper and carry less. It becomes interesting somewhere else, at the point where it lets a decision happen earlier than it used to.
- AI image generation rarely saves money, it moves the effort forward.
- It carries in the pre-stage and in the variant, not in the campaign image.
- Fabric, fit and skin remain the hard limit.
- Rights, labelling and sign-off belong before the first prompt.
Why the cost question about AI image generation misleads
The first instinct in almost every company is to compare against the shoot. A day rate on one side, compute on the other, and a convincing number appears. It does not hold, because it compares two different things. A shoot delivers a finished, approved, legally clean picture. AI image generation delivers a proposal that still has to be curated, corrected and owned.
So the effort does not disappear, it travels. It travels away from logistics and shoot days and towards selection and judgment. That is not a small shift, because selection and judgment are exactly the capabilities most thinly staffed in marketing organisations. How to organise that judgment is something I set out under evaluating creative work. Without it, the tool mainly produces volume.
Expectations are certainly high, and they are documented. In Generative AI: Unlocking the future of fashion, McKinsey estimates that generative AI could add between $150 billion and $275 billion to the operating profits of the apparel, fashion and luxury sectors within three to five years. That is a figure across the whole value chain, not across picture production alone. Read it in-house as a savings target for photography and you will miss it.
Where AI image generation actually carries in fashion today
In the pre-stage, before anything is produced
The strongest effect of AI image generation sits before production. A mood board that used to cost a week of research now takes an afternoon, and more importantly it arrives in several directions at once. That moves the moment at which a conversation about direction can happen forward in time. The value is there, not in the saved day rate.
In the variant, not in the hero image
A campaign image has to carry for months, work across markets and survive scrutiny. AI image generation is currently the wrong tool for that. For the thirty derivatives of the same image in different formats and backgrounds it is very well suited. The rule that settled in my own work is simple: the original is photographed, the family around it is computed.
In the catalogue, where volume is the problem
The least glamorous use case is the most rewarding one. Product pictures in large numbers, in a constant look, across many articles, are a volume problem and not a creative problem. Here AI image generation genuinely pays into speed without anybody having to argue about brand effect.
Where the limit runs
There are three places where the technology reliably fails in fashion, and all three are about credibility. The first is fabric. A knit that does not fall the way knit falls is noticed by people who buy knitwear, even when they cannot name what bothers them. The second is fit. Clothing sits on a body, and that fit is the actual product information. The third is skin, because that is where the eye is least forgiving.
There is also a less technical point. A brand lives on its pictures looking related to each other. Without firm direction, AI image generation produces an aesthetic that shows up everywhere at once, because many companies are operating the same tools with similar instructions. Move carelessly and you buy speed and lose distinctiveness. That is the same conflict I described under brand leadership, only with a new tool in hand.
What has to be settled before the first prompt
The most uncomfortable part comes first and still gets handled last. It comes down to four questions, and none of them is a marketing question in the narrow sense.
- Provenance. What data is the model trained on, and can that be evidenced for commercial use?
- People. Do the existing model and photographer contracts cover images being reworked as source material?
- Labelling. When is a generated picture declared as one, and who decides in a borderline case?
- Sign-off. Who carries responsibility for a picture no human took?
These four points can be settled in a few weeks if you touch them early, and they block a project for months if you touch them late. At this stage AI image generation is less a question of creation than one of governance.
What a sensible start looks like
If I were introducing AI image generation into a company today, I would not start with the tool but with a single product group. It should carry enough volume that speed becomes noticeable at all, and be uncritical enough that one failed picture stops nobody from buying. Basics work better than outerwear for this, and still life better than shots with people in them.
The second step is the uncomfortable one. Somebody has to own the selection, and their judgment has to be accepted across the house. That role cannot be handed out on the side, and it cannot be outsourced to the agency either, because it rests on brand knowledge. Only once both of those stand does the question of models and licences pay off. Reverse the order and you spend months discussing AI image generation as a technology and still end up with nobody who decides.
What I took away from it
My biggest misjudgment was assuming the tool would relieve the scarce resource. It relocated it. The scarce resource is not production, it is the decision about which of forty plausible pictures is the right one. Introduce AI image generation without naming that decision point and you do not get a faster brand, you get a louder one.
What surprised me is the side effect on collaboration. Because a proposal is now visible in minutes, buying, creative and sales suddenly argue at the picture instead of at the description. That is the real gain, and it appears in no business case. More on this in the AI in Marketing section.