Generative AI in Fashion: What It Does, What It Does Not, and Where to Start

Generative AI in fashion is the use of models that produce new images, text or 3D from a description or a reference, applied to the work of making and selling clothes. It is not one tool. It is a family of jobs: designing a garment that does not exist yet, photographing it on a model without a shoot, animating it, writing its specification for a factory, and putting it on a product page. The phrase covers all of them, which is why it is hard to answer the only question that matters: what does it replace today, and what does it not.

What generative AI already does in fashion, job by job

• Design. A garment, a print, an accessory, generated from a written brief or from a sketch. The value is the number of directions you can look at before committing: ten silhouettes in an afternoon instead of three.
• Product imagery. A product photo becomes an on-model image, a flat lay, a ghost mannequin or a try-on. This is the use with the clearest economics: a shoot costs a day and a budget, a generated image costs a credit.
• Video. A still becomes a walk, a turntable, a close-up, a short ad. The same asset feeds a product page, a story and a reel.
• Specification. Photos of a piece become a tech pack: an annotated flat sketch, a size chart with grading, a bill of materials. This is where generative AI stops being a picture tool and starts touching production.
• Copy and merchandising. Product descriptions, alt text, categories, all generated from the product itself.
• Styling and fit assistance. Outfit suggestions, size guidance, virtual try-on in the shop.

What it does not do

Three limits are worth stating plainly, because most articles skip them.

It does not produce a pattern. A generated image is not a set of cut pieces, and no image model knows how the fabric behaves on a body. That belongs to 3D construction software, and the two are complementary rather than competing.

It does not guarantee repeatability on its own. The same brief twice gives two garments unless the tool is built to hold a reference: the same model, the same fabric, the same silhouette across a catalogue. Repeatability is a product feature, not a property of the underlying model.

It does not remove the decision. The gain is in the number of options and the speed of the round trip; choosing which one becomes a product is still the designer's job, and that is the part that has value.

What the numbers say

McKinsey put the potential contribution of generative AI at 150 to 275 billion dollars of operating profit for apparel, fashion and luxury over three to five years, most of it in product development and marketing rather than in a single spectacular use. Brands that have used it publicly, from Adidas on colourways to Collina Strada on a whole collection, have used it in the same place: upstream, to widen the set of ideas, and downstream, to produce the images a launch needs.

How a brand starts, in practice

The honest starting point is not design, it is imagery, because the result is immediately comparable to what you pay for today. Take ten products that have a flat photo and no model shot. Generate the model shots. Compare them to your last shoot on three criteria: does the garment look like the garment, does the person look real, and would you put it on a product page. That test takes an afternoon and settles the question better than any article.

From there the chain extends on its own: the images that work become videos, the pieces that get chosen become tech packs, and the new ideas start from a description in the design studio.