AI model casting is choosing the faces that will represent your brand in generated images. It's the same job as a traditional casting, with one difference: instead of browsing an agency's books, you design the face. That brings a lot of freedom and just as much responsibility. This guide covers how to define a cast that genuinely looks like your brand and your customers, how to handle diversity across markets, and which consent and rights rules you can't ignore.
What AI model casting is
In a traditional casting the brand chooses between people who exist. In AI casting you start from a brief and generate proposals for faces, builds and styles until the cast matches the idea. The result is a group of virtual models with a defined identity, ready for campaigns, lookbooks and e-commerce.
The main risk isn't technical. It's ending up with generic faces, all alike, that look like they came out of the same filter: the so-called AI look. Good casting exists precisely to avoid that.
Start from the brand, not from the generative model
Before generating a single face, we answer a few questions with the brand:
- Who actually buys? Age, lifestyle, markets. The cast has to speak to those people.
- Aspirational or relatable? A luxury brand can choose an editorial face; an everyday brand works better with faces customers recognise themselves in.
- What's the brand's character? Bold, soft, ironic, minimal. It translates into expressions, posture, styling.
- How many lines and channels? A basics line and a capsule can have different faces; an online store needs continuity.
- Who don't we want to be? Often the most useful reference is a competitor to stand apart from.
How to build the cast: a 6-step process
1. Casting brief
One page with target, markets, brand character, number of faces and roles. Who is the lead face, who are the secondary ones, for which lines.
2. Moodboard
References for faces, styling, light and attitude. Not to copy real people, but to align on taste.
3. Proposals
For each role we generate a shortlist of candidates, different from each other and consistent with the brief.
4. Test on the real garments
The finalists wear two or three real pieces from the collection. A face that looks great on a neutral background may not work with your garments, your colours, your light.
5. Selection and reference set
The brand approves the cast, and for each face we build a reference set: the base for keeping it the same over time. We explain it in our guide to a consistent AI model.
6. Usage sheet
For each face: where it's used, in which markets, with what styling, what it must never do. It serves the team and whoever works on the brand in the future.
Diversity by market, without stereotypes
A brand selling in Italy, Germany, the Middle East and the United States doesn't speak to one audience. AI casting lets you adapt the cast to each market without multiplying shoots, but it has to be done with care:
- Represent your real customers. Skin tones, ages, builds and sizes that match who buys, not an abstract idea of "diversity".
- No tokens. A "different" face added to tick a box shows, and it doesn't help the brand.
- No stereotypes. Styling, poses and settings must not reduce a culture to a cliché.
- Real sizes. If you show a garment on different bodies, the fit must be believable on each: it's a matter of trust, not just image.
- Local sensitivities. Clothing, gestures and settings accepted in one market may not be in another.
"AI casting doesn't take choices away from the brand. It adds them. That's why it has to be directed, not left to chance."
Invented face or digital twin?
| Aspect | Invented face | Digital twin |
|---|---|---|
| What it is | A face that belongs to no real person | The digital version of a real person |
| Consent | No individual consent needed; check it doesn't resemble anyone recognisable | Needs the person's explicit consent and a contract |
| Authenticity | Low if the audience wants real people | High: the person exists and can take part in communication |
| Control | Full control for the brand | Shared, within the agreement's limits |
| When to use it | E-commerce, catalogues, campaigns with a brand face | Ambassadors, testimonials, ongoing collaborations |
We only work with invented faces or consensual digital twins. We cover them on our digital avatars for brands page too.
Consent, rights and transparency
An identifiable person's face is personal data under the GDPR, and many countries, Italy included, add specific image-rights protection on top. A digital twin therefore needs explicit consent and a contract that sets out uses, channels, territories, duration, fees and what happens if the person asks to end the collaboration.
With invented faces the risk is different: too strong a resemblance to a well-known person. It's checked at casting stage, and any face that "reminds you of someone" is dropped.
Finally, transparency: the AI Act sets disclosure obligations for generated or manipulated content that realistically depicts people. We label AI content where required. To go deeper, read AI and image rights.
When AI casting isn't the right choice
If the campaign revolves around a real person and their story, if your audience asks for authenticity (think of a brand born from its community) or if you need runway shows and live events, real models are still the better choice. Often the answer is mixed: real models for the main campaign, a consistent AI cast for e-commerce, variants and markets. That's how we use virtual models in AI on-model photography.
Frequently asked questions
How does AI model casting work?
You start from a brief with target and markets, generate face proposals, test the finalists on real garments and approve the cast. Each face gets a reference set to keep it consistent.
Are AI fashion models real people?
It depends: an invented face belongs to no one, while a digital twin is the digital version of a real person and needs their explicit consent.
Do I need a contract to use an AI model?
For a digital twin, yes: consent and an agreement on uses, territories, duration and fees. For an invented face, check it doesn't resemble recognisable people.
How do you handle diversity in AI casting?
By representing each market's real customers, with believable ages, skin tones and sizes, and by avoiding tokens and stereotypes in styling and settings.
Do I have to disclose that the model is AI-generated?
The AI Act sets transparency obligations for generated content that realistically depicts people. We label AI content where required.