AI clothing on a model: what a fashion brand should check first

AI clothing on a model means taking a photo of a garment on a hanger or laid flat and making an image that shows the same garment worn by a model. For a clothing or textile brand it is tempting, because a new product no longer needs a model, a studio and a shoot day.

In clothing, though, mistakes are expensive. A product that looks a different colour, texture or fit in the image comes back as a return. This guide covers what to check as a brand or producer: the source photo, fabric, pattern, colour, size and the marketplace side. At the end there is a checklist to run before you publish.

The source photo decides most of the result

The AI takes the garment from your photo. It cannot get a detail right that the photo does not show, at best it guesses. Five more minutes at the shoot are cheaper than throwing out dozens of images later.

• Lay the garment flat on a plain surface or hang it, with no creases. Creases in the fabric carry into the image.
• Shoot the front and the back separately. If there is a print, zip or slit on the back, the back photo is a must.
• Use soft daylight. Yellow lamps shift the colour and flash causes glare on shiny fabric.
• Add a close-up of the fabric so the knit, weave, lace or print is sharp.
• Take a separate detail shot of labels, buttons, embroidery and logos.
• Send a separate photo for each colour. Shoot the real colour instead of having one colour turned into another.

Fabric texture and drape

Fabric is where AI struggles most. A knit can flatten, linen can look like smooth cotton, satin can come out matte. If a heavy coat looks like a thin jacket on the body, the buyer will notice when it arrives.

Check the drape too. Heavy fabric falls, light fabric moves. If a chiffon dress looks stiff or denim ripples like silk, do not use that image. If fabric movement is what sells the product, as with evening wear or scarves, plan for at least one real shoot or video.

Pattern and print

On a patterned product look at three things. Scale: are the flowers, checks or stripes the same size as on the real product. Repeat: does the pattern repeat in the same order as the original. Seams: how do the lines of a checked shirt meet at the placket and the collar.

Text and logos are a separate risk. One changed letter in a chest print or a misspelt brand name is a small error with a big cost. Zoom in and read the image letter by letter. Then look at it small as well, because fine stripes can shimmer at thumbnail size.

Colour accuracy

A well-known reason for clothing returns is that the product's colour differs from the image. AI can warm or cool a colour slightly to suit the light and the scene. Beige turns cream, navy drifts to black.

1. Put the image next to the original photo and compare them on the same screen.
2. Hold the real product next to the screen. Photos can be wrong too.
3. Make each colour variant from its own photo and show only that colour in each image.

Check on both a computer and a phone. Buyers will see the product on a phone screen too, and two screens can show the same colour differently.

That last point is more than good practice. Google Merchant Center's image requirements ask for a unique image for each variant, only one variant per image, and an image that shows the selected colour and material correctly.

Size, fit and measurements

A model made by AI has no real size. So a line like "the model is wearing size S" means nothing on an AI image and can mislead. Put a real size chart on the product page instead: chest, waist, length and sleeve in centimetres.

Check the fit as well. If an oversized sweatshirt looks slim in the image, or a midi skirt ends above the knee, the image tells the wrong story about the product. Show every image once to the person who knows the product best, usually the pattern maker or the production lead.

Choose the model with your buyers in mind. Showing a plus-size range on a very slim model tells the buyer nothing about how it will look on them. Keeping the same type of model across a collection also keeps the catalogue consistent.

Marketplaces and honesty

Google openly accepts images that show clothing on a model in product listings. Its image best practices page gives "apparel worn by a model" as an example and says such images can also be sent with the lifestyle_image_link attribute. Under the same rules, images made with generative AI must carry metadata saying so, and tags such as IPTC DigitalSourceType must not be removed.

We could not find a public official rule on AI model images from other marketplaces, so we do not state rules on their behalf. Read the image guide in your seller panel before you upload.

Then there is the person. If you use a photo of your own model or of someone you know, get their written permission. Never use a face that resembles a well-known person.

Checklist before you publish

Open each image at full size and go through it in this order. If an image fails a point, do not try to save it with retouching. Make it again or leave it out.

1. Is the colour the same as the original photo and the real product.
2. Is the fabric texture (knit, weave, sheen) recognisable.
3. Are the pattern's scale and repeat the same as the original.
4. Are prints, embroidery and logos correct letter by letter.
5. Are buttons, pockets, zips and seams the same in number and position.
6. Is the fit and length (oversized, midi, cropped) shown correctly.
7. Does the drape suit the weight of the fabric.
8. Is an accessory that is not sold with the product in the foreground.
9. Does the product page have a real size chart.
10. Does the file still carry its AI label and metadata.

With Bulsu Labs Studio

Bulsu Labs Studio makes product images with AI from your product photos. One photo per product is enough. You get a white or plain backdrop, and room or outdoor scenes if you want them. Every image is checked against the original product, and one where the colour, pattern or logo changed is generated again at no cost. Every file is labelled as AI-made. You may send a photo of a real person only with their permission. 50 images cost €100 and 100 images cost €175, paid once. Our guide to AI product images from one photo covers how to shoot the source photo, and product photo shoot or AI images compares the costs.