An AI fashion video or lookbook can carry a full season today, with one condition: the clothes come from a camera and almost everything else can come from a model. AI is good at mood, location, light, movement and styling variations. It is still unreliable on exact garment construction, fabric behaviour, fit, prints and logos, which are the things a customer is actually buying.
This guide sets out what is realistic, the hybrid workflow most brands settle on, the rules on model likeness and disclosure as they stand on the day of writing, and when a real shoot is still the right call.
What does AI do well in fashion video?
Mood and world. A lookbook sells a feeling before it sells a jacket. A salt flat at dawn, a brutalist stairwell, a 1970s Riviera hotel: generated settings reach places a small brand could never afford to fly to, and the light is whatever the art direction needs.
Location variety. A shoot gets one or two locations a day. A generated season can put each drop in its own world and still look like one brand, provided the look is designed up front as a set of still images.
Motion. Slow walks, turns, hair and atmosphere, camera moves through a space. Short clips of this kind make strong Reels and TikToks, and they can be animated from approved stills.
Styling variations. The same look on different models, in different settings, in different crops for different markets. This is where the saving is largest, because every variation on a shoot costs time on set and every generated variation is a small job.
What does AI fashion video still get wrong?
| Area | What goes wrong | Why it matters |
|---|---|---|
| Garment construction | Seams shift, pockets appear and vanish, button counts change, collars are redrawn between frames | The customer receives a different garment from the one shown |
| Fabric behaviour | Heavy wool moves like jersey, silk loses its sheen, knit texture turns to a smooth surface | Fabric is half the reason to buy |
| Fit | Generated bodies make everything fit perfectly | Returns, and a possible misleading-advertising problem |
| Prints and patterns | Repeats warp, stripes break at seams, florals mutate | A print is a design asset and must be exact |
| Logos and text | Letters deform, labels change | Brand damage, and the detail viewers check first |
| Continuity | The same model and outfit drift across shots | A lookbook needs one cast, not thirty near-lookalikes |
None of this is a reason to avoid AI. It is a reason not to let the model invent the clothes.
How does a hybrid AI lookbook workflow work?
The method that holds up is simple to state: real garments, generated everything else.
- Photograph the garments properly. On a model, a mannequin or flat, in even light, from several angles, with close-ups of fabric, fastenings and prints. This is a studio day, not a campaign shoot. For accessories, packshots are enough, as we describe under product videos.
- Design the season’s look as stills. Casting, locations, light, colour and framing are agreed as still images before anything moves. Changes at this stage are quick. This is also where the cast is fixed, using the reference-sheet method from our guide to keeping an AI character consistent across 20 shots.
- Build each look around the real garment. The photographed garment is the reference for the generated image, or is composited into it, so that construction, print and colour stay exact. The model and setting are generated.
- Animate the approved stills. Short shots, gentle movement. The more a garment has to move, the more likely it is to drift, so ambitious fabric movement is tested early or filmed for real.
- Check garments frame by frame. Every look is compared with the garment photographs: seams, buttons, print placement, colour, logo. Shots that fail are fixed or remade.
- Cut for each placement. Lookbook film, vertical Reels and short-form clips, product-page loops and stills pulled from the same material.
At Zero Shot, an AI video studio in London, this is how we approach fashion work: we start from the client’s own garment photography and check the product before anything is delivered. If you are comparing suppliers, how they keep garments accurate is the first thing to ask, and it is on our list of questions for an AI video production company.
Whose face is it? Model likeness and consent
There are two kinds of AI model, and they carry different obligations.
An invented model is designed from scratch and resembles nobody. The risks are accidental resemblance to a real person and ownership: agree in writing who owns the character and whether you can keep using it if you change supplier. Our explainer on who owns AI-generated video lists what that agreement should cover.
A digital replica is based on a real model. That needs her or his consent, in writing, for a defined use. In New York this is law. Under the Fashion Workers Act, in effect since 19 June 2025, clients must obtain clear written approval before creating or using a model’s digital replica, and the approval must state the scope, purpose, rate of pay and length of time. H&M’s 2025 plan to create digital twins of 30 models was presented on similar terms: the company said the models would keep the rights to their twins and that the images would be marked as AI.
Never build a model “in the style of” a recognisable person, and never train on an agency’s portfolio without permission. Outside New York the legal routes differ, but the contract standard should be the same.
Do you have to disclose AI models?
The public debate is real and recent. In 2025 a Guess advertisement in the August issue of American Vogue used an AI-generated model with a small disclosure, and the reaction was loud: critics raised lost work for models, photographers and crews, unrealistic beauty standards, and the size of the label. Whatever you think of the arguments, the lesson for a brand is practical. Audiences notice, and a hidden AI model is a bigger story than a disclosed one.
The rules differ by place, and they are moving.
- United Kingdom. The ASA says in its note on disclosing AI in advertising that there is no blanket requirement to disclose AI in ads. The CAP Code’s rule that ads must not materially mislead still applies, and the ASA adds that a disclosure is very unlikely to rescue an ad that is fundamentally misleading. For fashion, the practical test is whether the garment shown is the garment sold, and whether the fit is honest.
- New York. A state law in effect since 9 June 2026 requires a conspicuous disclosure when an advertisement includes an AI-generated synthetic performer. If your ads reach consumers in New York, a generated model is likely to be covered.
- European Union. Article 50 of the AI Act has applied since 2 August 2026. According to the European Commission’s Article 50 FAQ, those who deploy deepfakes must disclose them clearly at first exposure at the latest. The Commission describes a deepfake as content resembling someone or something that exists or can plausibly exist and that would falsely appear authentic. Whether a given campaign falls inside that definition needs proper advice.
- Platforms. Rules vary and change. TikTok’s advertising policy, for example, requires an AI-generated content label or a clear disclaimer on ads that are fully generated or significantly edited with AI. We compare the AI labelling rules on Meta, TikTok and YouTube in a separate guide.
Our recommendation is to label. A short line such as “Models and settings created with AI. Garments photographed.” costs nothing, meets most of the rules above and makes the accuracy point for you. This section is general information and not legal advice.
When is a real shoot still the right call?
- The hero campaign of the season, where the brand’s relationship with real talent, photographers and press is part of the point.
- Fit-critical product pages: tailoring, denim, swimwear, lingerie, anything with a size guide people rely on.
- Fabric in motion as the selling point: a bias-cut dress, a pleated skirt, technical outerwear in the rain.
- Anything built around a named person: an ambassador, a collaborator, a founder.
- Communities that expect real bodies and would read generated models as a betrayal.
Often the answer is both. A one-day garment shoot feeds a season of generated content, and one real campaign gives it a human anchor. We compare the two routes in AI video or a real shoot.
If you have a collection coming and garment photography in hand or planned, send us a short brief. We will tell you which looks can be generated, which should be filmed, and give you a fixed price.
Questions people ask
Can AI replace a fashion photoshoot completely?
For mood films, social content and early concept lookbooks, often yes. For imagery that customers use to decide whether a garment will fit and how the fabric falls, no. Most brands end up with a hybrid: real garment photography with generated models and locations.
Do I have to disclose that the models in my lookbook are AI-generated?
It depends on where the content runs. The UK has no blanket disclosure rule, but ads must not mislead. New York requires a conspicuous disclosure in ads featuring synthetic performers, and the EU AI Act has transparency duties for realistic generated content. This is general information, not legal advice.
Will the clothes in an AI lookbook look exactly like the real ones?
Only if the process starts from photographs of the real garments and each frame is checked against them. Clothing generated from a description will drift: seams move, buttons multiply, prints change. For e-commerce, the garment must match what the customer receives.
Can I create an AI model that looks like a real model I have worked with?
Only with her or his clear written consent covering scope, purpose, pay and duration. That is the legal standard in New York under the Fashion Workers Act and a sensible contract standard everywhere else.