Why fashion ecommerce moved to AI imagery first

If you sell a garment in five colourways and three of them need their own images, in a category that drops new product several times a year, for a website, a marketplace, paid social, organic social and a newsletter, each with its own crop, you are producing dozens of images per style before anyone has mentioned a campaign. Multiply that by a few hundred styles a year. No studio schedule survives that, and no budget does either.

Other categories are heading for the same wall more slowly. That makes fashion worth watching, as long as you copy the right parts.

Three things fashion worked out first

1. The person and the setting are infrastructure, not creative choices. Fashion teams learned to fix the model and the environment once, then run the collection through them. The creative decision happens at the start of the season, not per image. That is what makes a drop look like a drop.

2. Variants are a production problem, not a photography problem. The second colourway does not need a new idea. It needs the same image with a different colour, correctly. Once you see variants as a production line rather than a series of shoots, the whole cost structure changes.

3. A review step is cheaper than a recall. Fashion found out early that a wrong image is not a cosmetic issue, it is a return. So the approval moment got formalised. Someone who has held the garment signs off the set before it goes out.

What transfers to your category

Most of it, and faster than people expect.

Fixed settings beat one-off scenes. A furniture brand that defines six interiors and runs the whole range through them gets the same benefit as a fashion brand with one studio. We wrote that out in one product across ten interiors.

Variants are where the money is. Sanitary ware in three finishes, windows in four frame colours, a planter in five sizes. These are the images a studio day cannot justify and a production system can.

The second market is a variant too. The same product in a setting that reads as German, Danish or Dutch is the thing your local dealer actually needs, and it is the kind of work that is impossible to shoot and straightforward to produce.

The review step transfers exactly. Whatever your category, the person who knows the product has to look before it goes out.

What does not transfer

This is the part other categories get wrong by copying fashion too literally.

Dimensional accuracy. In fashion, a scene can be invented. In sanitary ware, furniture, windows and anything installed, the proportions are the product. That is why we work from technical drawings and product references rather than from a description, and why an image that looks beautiful but shows the wrong tap height is simply a reject.

Material as the purchase reason. Where the buyer is buying the feel of the material, the hero image has to be real. Fashion has this too, but fashion compensates with volume elsewhere. A single-product brand cannot.

Context that has to be correct, not just plausible. A kitchen scene in a fashion-style campaign only has to feel right. A scene showing how a component is installed has to be right, because someone will build from it.

Speed as the main argument. Fashion is driven by the calendar. If your category replaces its range every four years, the argument for production at volume is consistency and coverage, not speed. Selling it internally as a speed story will set the wrong expectation.

What this looked like outside fashion

Xenz, part of Beterbad Sanitair, needed lifestyle visuals for its bathtub and tap catalogue on a tight deadline, with their existing supplier unable to deliver in time. The visuals also had to be suitable for high-resolution print, so technical accuracy and photorealism both had to hold.

The method was the fashion method applied to sanitary ware. First a scan of the brand identity, colour palette and visual style. Then a collection of bathroom presets built on that foundation, with contemporary materials and interior styles that fitted the range. Only then the products, produced from technical drawings and product references so that shapes, proportions and finishes were right. Complementary products from the assortment, taps and shower enclosures, were placed into the same settings, which gave the images a cross-selling job as well as a catalogue one. The whole project was finished in three days. It is written up in full on our use cases.

What matters for this article is the sequence, which is identical to a fashion drop: identity first, settings second, products through them third. What is different is step three. A garment can be invented into a scene. A bathtub has to measure correctly, and that is the constraint every non-fashion category inherits.

The order to copy it in

If you want fashion's benefit without fashion's mistakes:

1. Pick one product family, not the whole catalogue.

2. Fix the settings before you produce anything.

3. Produce the variants first, because that is the clearest win and the easiest to evaluate.

4. Formalise who signs off, and give them the real product to compare against.

5. Only then look at campaign and lifestyle work.

Teams that start at step five get a few impressive images and no change in their weekly output. Teams that start at step three notice the difference within a month.

Where this is going

The brands that get the most out of this are not the ones using the most AI. They are the ones who decided what their visual standard is and then produced against it consistently. The technology lowers the cost of producing to a standard. It does nothing about setting one.

That is also why we are sceptical of tool-led decisions. Which model or tool is best changes every few months. What your category needs from an image does not. More on that in the categories that matter for a brand team.

In closing

Fashion got here first because it had to. Your category will arrive at the same place with more time to prepare, which is an advantage if you use it to write down your standard before you scale production.

Start where fashion started: fixed settings, variants first, one person who signs off. The detail on models is in where AI models hold up and where they do not, and the consistency side is in what breaks first when a range grows. Or see how it works on your own range.

Want to test it on one product family? Try LensLab free with five credits.