Should an AI model match a real person’s measurements?

For fit photos, yes. Set the synthetic model’s height and size to match a fit model you have measured, so the size printed under the photo is true. For mood and campaign images, match the range of your customers instead of one person.

How many models should a small brand cast?

Enough to cover the size groups and ages you sell to. A brand that sells straight and plus sizes needs at least one model for each, and one approved character sheet per model (one image of that person from several angles) keeps each person the same across sets.

Can I make a digital twin of myself or my staff for our own brand?

Yes, with the same care you would give a hired model. Put the terms in writing: what the twin is used for, for how long, and what happens to it if the person leaves the company.

How to cast AI models responsibly

Checked September 24, 20266 min read

Short answer

Cast AI models the way you would cast people: match the sizes, ages, skin tones and abilities of the customers you sell to, write the casting down, and check every image for stereotypes. Never build a model from a real person’s face or body without their written, paid consent. Disclose synthetic people where platforms or laws require it, and do not present AI casting as a diversity commitment, the framing Levi’s withdrew in 2023.

What went wrong in the public cases

In March 2023 Levi’s announced a pilot with Lalaland.ai, a studio that generates models, to show products on more body types, ages, sizes and skin tones. Critics said the company should hire diverse models instead, and others pointed out that shoppers look at product photos to see clothes on a real body. On March 28, 2023, Levi’s added a note saying it did not see the pilot as a way to advance diversity or as a substitute for its diversity goals, and that it was not scaling back live shoots or its work with diverse models.

In Vogue’s August 2025 US issue, a Guess ad showed two AI-generated women made by the agency Seraphinne Vallora. Readers objected to how real they looked and to what it meant for models’ and photographers’ jobs, and the AI credit ran in small print. Vogue said the ad came from its advertising side, not its editors.

Both drew criticism for how they presented synthetic people: as progress on diversity in one case, and as an ordinary fashion ad in the other.

Synthetic models should not look like anyone

A fully synthetic model should not resemble a real person. Do not prompt with a celebrity’s or model’s name, and do not use someone’s photos as face references without their consent. If a generated face turns out to look like someone well known, recast. The legal side is covered in AI images of real people.

Cast for the customers you have

Start from your own data: sales by size, customer ages, the countries you sell in. Baymard Institute’s 2026 survey of 1,922 US online shoppers found that 24 percent of women and 14 percent of men who buy apparel call themselves plus-size, and 16 percent of women call themselves petite. If your size range covers those shoppers, your model photos should too; the fit risks are in how to show clothes on different body sizes.

Amazon’s main-image rule for adult clothing asks for a standing model, and in the same line encourages showing different physical mobilities, including models who use wheelchairs or prosthetics.

Image models carry their own defaults, so check them. A 2023 study of DALL-E 2 and two versions of Stable Diffusion found that their outputs underrepresented marginalized identities. A 2024 study that ran eight focus groups with 25 disabled people found that text-to-image models repeated reductive stereotypes of disability, such as wheelchair users shown sad, lonely or in pain. Newer models may do better, so review every image rather than assume.

Render bodies and disabilities with care

Describe the person doing something ordinary in your clothes, not their condition.
Specify the body the way a casting call would: height, build, size worn, posture. Leave out words that frame a body type as a joke or a problem.
If you show assistive technology, get it right: a real wheelchair’s wheels and frame, a prosthetic that fits the limb. Ask someone who uses one to review the image before it runs.
Check skin tone in the final color grade. A warm or cool grade shifts skin as much as it shifts fabric.
Keep each cast member consistent across the set, so a plus-size or disabled model appears throughout the collection. See how to keep the same model in every AI photo.
Hire real people where the body is the information: fit photos for extended sizes, adaptive clothing, lingerie and swimwear.

Disclose synthetic people where it is required

Amazon (US): tag images of photorealistic, fully AI-made people with contains-synthetic-performer in the XMP metadata, the information stored inside the file. Amazon adds a disclosure where it applies.
New York: an ad with a synthetic performer must disclose it conspicuously when the advertiser knows. The law was signed December 11, 2025, takes effect 180 days later, and sets penalties of $1,000 for a first violation and $5,000 after that.
Google Merchant Center: keep AI metadata, such as the IPTC value TrainedAlgorithmicMedia, in the image file.
Ad platforms have their own AI labels; see whether you need to label AI images in ads.

Put a plain line where shoppers will read it, next to the size information: "Model image generated with AI. Model wears size M." It answers the fit question at the same time.

A casting brief you can copy

Casting brief for AI models

Casting brief: [collection or campaign]
Customer: [who buys this: age range, countries, the sizes that sell most]
Cast: [number of models] for [number of looks]

For each model
  Name in the files: [Model A]
  Real or synthetic: [fully synthetic / digital twin of (name), written consent signed (date)]
  Age range: [e.g. 30 to 40]
  Height and build: [e.g. 5 ft 9 in, athletic]
  Size worn in photos: [e.g. M in our size chart]
  Skin and hair: [describe what you want to see, e.g. deep brown skin, short black coily hair]
  Mobility or assistive technology: [none / wheelchair user / prosthetic leg, reviewed by (name)]
  Face references: [none for synthetic / consented photos only]

Never: [celebrity or model names in prompts, real people’s faces, stereotyped props or poses]
Disclosure line: ["Model image generated with AI. Model wears size M."]
Platform tags: [Amazon contains-synthetic-performer: yes / no]
Approved by: [name, date]

Questions people also ask

Should an AI model match a real person’s measurements?: For fit photos, yes. Set the synthetic model’s height and size to match a fit model you have measured, so the size printed under the photo is true. For mood and campaign images, match the range of your customers instead of one person.
How many models should a small brand cast?: Enough to cover the size groups and ages you sell to. A brand that sells straight and plus sizes needs at least one model for each, and one approved character sheet per model (one image of that person from several angles) keeps each person the same across sets.
Can I make a digital twin of myself or my staff for our own brand?: Yes, with the same care you would give a hired model. Put the terms in writing: what the twin is used for, for how long, and what happens to it if the person leaves the company.

Where Overs fits

Overs casts a model for worn products as a character sheet, one person from several angles wearing your product, which you approve before any other photo is made. Earlier sheets can be reused as a talent library, so a cast you approved stays the same across sets.

Free for 40 photos a month. The AI that makes the photos is billed separately, on your own key, with no markup from Overs.

Start free How it works

Sources

Checked on September 24, 2026. Prices, specs and rules change; follow the links for the current versions.