What AI models can and cannot do for body diversity
| Question | What AI models can do | What they cannot do |
|---|---|---|
| Showing range | Show the same garment on models with different builds, ages and skin tones | Prove that the garment fits every one of those bodies |
| Fit information | Give a visual impression of length and drape | Replace a size chart, garment measurements or a real fit test |
| Representation | Help a small store show more than one look | Replace working with real, diverse people where that matters to your brand |
| Consistency | Keep the same set of models across the whole catalog | Decide for you which bodies your customers have |
The left column is real value, especially for a store that so far had only flat lays. The right column is where stores get into trouble: an image that suggests a fit the garment does not have leads to returns, and presenting generated people as a diversity achievement has drawn public criticism, as the Levi's case below shows.
Why body diversity matters in product images
Shoppers look for themselves in product images. A person who wears a larger size, is older or has a different skin tone than the usual catalog model has to guess how a garment would look on them. Some guess wrong and return the item; others do not buy at all.
For a fashion store in Southeast or Central Europe the customer base is broad: different ages, sizes and styles, often in a small market where every loyal customer counts. Showing garments on more than one kind of body tells these customers that the store had them in mind. It is also simply better product information.
The effect differs by product. A loose scarf or a bag looks much the same on any model. A fitted dress, high waisted jeans or a wrap top look very different on different bodies, and these are the products where showing more than one build helps shoppers most. Start there rather than with your whole catalog. Our article on how to reduce fashion returns covers the wider return problem.
The Levi's lesson
In March 2023, Levi's announced a test with Lalaland.ai, a studio whose AI models were described as being of every body type, age, size and skin tone. Engadget reported that Levi's framed the partnership as increasing the number and diversity of its models.
The reaction was sharp. Critics asked why the company did not hire diverse real models, as reported by The Independent. Levi's then said it was not scaling back live shoots or its commitment to working with diverse models, and that the partnership should not have been conflated with its diversity strategy.
The practical lesson: use AI models to show your products on more bodies, but do not present generated people as proof of inclusion. Describe what you do plainly, and if representation is part of your brand, keep real people in your campaigns. More cases are in our article on companies using AI fashion models.
Plan a model set instead of random faces
Diversity works best as a plan. A different body for every product looks random, and shoppers cannot compare items. A fixed set of models, used the same way across the catalog, looks deliberate.
- Look at your customers. Your order data, reviews and social followers tell you roughly who buys. Plan models that look like them.
- Choose two to four models. For example, two builds in womenswear and one in menswear. A small set is easier to keep consistent.
- Decide where each one appears. For instance, every dress is shown on both womenswear models, and basics on one.
- Keep the scene and poses the same. When the background and poses stay fixed, the body is the only thing that changes, and the comparison is useful.
- Write it down. Add the set to your style guide. Our article on a consistent AI model for your brand explains how.
Describing build and skin tone for a custom model
With Wizia, the built in library is available on every plan. On the Pro+, Business and Enterprise plans you can also create custom models: you describe the model in English, including looks, hair, skin tone and build, generate a preview, adjust and save.
A few tips for descriptions of different bodies:
- Be concrete and neutral. "Woman in her forties, curvy build, wide hips, medium brown skin, short grey hair" works better than a vague or flattering description.
- Describe the body, not a size label. Size labels differ between brands and countries, so describe build and proportions instead.
- Avoid idealised extremes. Very tall, very slim or very sculpted figures look like advertising, not like customers.
- Adults only. Wizia's AI models are adults; describe adult models only.
- Test with your hardest product. Generate a fitted dress or tailored trousers on each new model before you use it across the catalog.
Our article on how to create AI fashion models covers the custom model workflow step by step.
Keep fit information honest
This is the most important point in the article. An AI image is generated from one photo of the garment. It shows a believable picture of the garment on a body, but nobody measured the garment on that body. So:
- Do not write "model wears size M" under an AI image. There was no size M on a real person. Give garment measurements instead, such as length, chest width and waist.
- Keep a clear size chart on every product page, in centimetres.
- Do not let the image promise more than the garment does. If a dress is not cut for a fuller bust, an image that shows it fitting perfectly misleads the buyer. Check every image against the real garment and remove ones that look wrong.
- Add fit notes from real people. A line such as "runs small, tried on by our team" is worth more than any image.
Our comparison of size and fit tools explains the options for fit advice.
Let shoppers see the garment on themselves
However many models you show, the shopper's own body is the one that matters to them. A virtual try-on widget lets shoppers see the garment on themselves before they buy, from their own photo, on your product page.
Wizia Virtual Try-On installs with one script tag pasted into your store theme, and works on Shopify, WooCommerce and CloudCart themes and other site builders that accept a script. Every plan, Free included, can run the try-on on your store while you have credits. Shoppers' photos are used only for the try-on and are not kept. Try-on and a diverse model set work well together: models show the range, try-on answers the personal question.
What it costs
Showing a garment on more models means more images, so plan the budget. As a made up example, a store that adds 40 products a month and shows each on two models, with two images per model, needs 160 images: 160 credits at 2K, where each 2K image costs 1 credit.
The Free plan includes 20 one-time credits to test, with a watermark. Starter costs €20 a month with 150 credits, and Pro costs €98 a month with 750 credits. Custom models start with Pro+, at €298 a month with 2300 credits. Every plan is on the pricing page. For the bigger picture, go back to the generative AI in fashion guide.