A Model, a Contract, and a Line Nobody Has Defined Properly
In May, New York model Francheska Pujols sued budget retailer Rainbow Shops. According to reporting on the complaint, she had been photographed in the brand's clothing against a plain white background. She alleged that Rainbow then used AI to create entirely new images of her, including poses and scenes she had not approved.
The reported contract allowed minor edits. Her argument was that generating new scenarios from her original photographs was not a minor edit. Rainbow denied wrongdoing. Pujols later withdrew the lawsuit while the parties sought to resolve the dispute privately, so there is no courtroom ruling telling the industry exactly where that line sits.
That missing ruling matters. The case put a very practical question in front of every brand using AI-assisted imagery: when does editing become a new performance, and did the person in the original image actually consent to it?
The First Legal Line Is Narrower Than the Headlines
On June 9, 2026, New York's synthetic performer disclosure law took effect. Advertisements distributed to New York audiences must conspicuously disclose the use of a qualifying synthetic performer. The statute defines that performer as an AI-created or AI-modified digital asset that appears to show a human performance but is not recognizable as an identifiable real person.
That last part is important. This is primarily a disclosure law for artificial performers. It does not settle every consent, contract, right-of-publicity, or false-endorsement issue involving a recognizable model whose real photographs were manipulated. Brands still need to treat those as separate legal and contractual questions.
The law carries civil penalties of $1,000 for a first violation and $5,000 for subsequent violations. It reaches ads distributed in New York even when the advertiser is located elsewhere.
The European Union's AI Act transparency rules are also scheduled to apply from August 2, 2026. They include requirements around making certain AI-generated content identifiable and visibly disclosing deepfakes. The exact obligation depends on the content, the actor in the AI chain, and the final implementation guidance. Translation: “we used AI somewhere” is not yet a compliance strategy.
Generated From Scratch Does Not Mean Ethically Blank
The Pujols dispute concerned a recognizable person whose real photographs were allegedly transformed. A second issue is harder to see: the supposedly fictional model generated from scratch.
Generative image systems learn patterns from very large datasets. Depending on the system and its data practices, those patterns may reflect the faces, bodies, poses, styling, and visual work of real people. An output can therefore resemble a working model without the brand intentionally naming or selecting that person. The closer the resemblance, the less useful “the computer made it” becomes as a defense in the court of public opinion, and potentially in an actual court.
There is no clean industry consensus on how close is too close, how similarity should be checked, or who carries responsibility across the platform, agency, and brand. Tool vendors differ sharply in what they disclose about training data, indemnification, model releases, output ownership, and commercial use.
This is why the hesitation inside fashion is real. The question is no longer whether brands can make AI imagery. It is whether they can explain how it was made, what rights support it, and who reviewed it before it went live.
The Jobs Were Always Part of the Product
Every fashion shoot runs on people. Photographers who understand light, product, and the way fabric moves. Hair and makeup artists whose precision is often most visible when nobody notices it. Stylists who know the difference between a garment that photographs and a garment that sells.
Art directors hold the visual language of the brand. Producers make the logistics work and keep a shoot from quietly setting the budget on fire. Location scouts find the exact wall, beach, apartment, or warehouse that makes the story believable. Retouchers turn a shoot into usable assets. Models understand tension, posture, fit, movement, and how to give a garment a point of view.
These are not interchangeable line items. They are a production ecosystem. The honest concern is not simply that AI will change the work. It is that some brands will delete the ecosystem, replace it with a software subscription, and call the quality loss efficiency.
The cost gap is undeniably tempting. Traditional on-model production can cost tens or hundreds of dollars per finished image once talent, studio, crew, usage, logistics, and retouching are included. AI services can bring the marginal image cost down to single digits. The exact savings vary wildly, but the pressure on catalog and performance-content budgets is not theoretical.
AI Does Not Remove Creative Work. It Redistributes It.
The next production model is not “people or AI.” It is a smaller, differently skilled human team directing, checking, fixing, documenting, and distributing a much larger volume of synthetic or AI-assisted work.
This person writes image prompts the way a strong art director writes a brief. They understand lighting, casting, styling, composition, lens language, product truth, and brand standards. The skill is not typing adjectives into a box. It is directing an image system toward an intentional result.
Knowing what to ask for, what to reject, and when the image has drifted away from the brand.
Someone has to manage inputs, model releases, approved reference assets, generation history, version control, output labeling, review rounds, and final channel specifications. At scale, this becomes production operations, not casual prompting.
Keeping high-volume AI content from becoming an untraceable folder called FINAL_FINAL_7.
The creative director reviews AI outputs with the same critical eye used on set, while adding new questions: Is the product accurate? Does the talent look credible? Is the body distorted? Does the image resemble a known person? Is disclosure required? Can the team prove where the inputs came from?
The output volume may be ten times higher. Human judgment becomes more important, not less.
This role focuses on proportions, expression, skin, movement, garment interaction, identity consistency, casting range, and the visual signals that make generated talent look strange. It also needs a strong ethical standard around body manipulation and resemblance.
Making the image feel intentional without quietly turning every human body into the same impossible mannequin.
Retouching does not disappear. It shifts toward correcting anatomy, hands, fabric behavior, seams, logos, hardware, color, shadows, fit, and hallucinated garment details. Ecommerce imagery cannot merely look convincing. It has to represent the product a customer will actually receive.
Turning an impressive generation into a commercially usable asset.
This title is not standardized yet, but the work is arriving. Someone needs to check whether real-person permissions cover AI transformation, whether synthetic talent resembles identifiable people, whether the tool's commercial terms are acceptable, and whether the final ad needs disclosure.
Stopping the brand from learning its AI policy through a cease-and-desist letter.
Once a brand can create ten times more imagery, someone has to decide what deserves to exist. Which channels need variants? Which products need real shoots? What should remain documentary, editorial, or human-led? More content without a distribution thesis is just a larger digital junk drawer.
Matching production method to business need instead of generating because the button is available.
What Needs to Exist Before the Next AI Campaign
- Contract language that names AI. “Editing” is too vague. Define whether photographs, video, voice, body, face, pose, wardrobe, and setting can be generated or materially transformed.
- A documented consent and usage trail. Record the source asset, release, territory, term, channels, transformations allowed, and final outputs.
- A tool-vetting standard. Review training-data disclosures, enterprise privacy, commercial-use terms, indemnity, data retention, and whether uploaded assets can train future models.
- A likeness review. Check outputs for resemblance to known or discoverable people, especially before paid media or a major campaign.
- A disclosure rulebook. Decide who determines whether New York, EU, platform, union, client, or internal labeling rules apply.
- A product-accuracy gate. Confirm fabric, fit, seams, trims, color, logos, closures, and construction against the real sample.
- A human escalation path. Give the team a clear person who can stop an image from publishing when consent, likeness, or accuracy feels wrong.
Will Brands Build the Infrastructure or Just Cut the Budget?
AI imagery still needs taste, judgment, legal awareness, product knowledge, and visual intelligence. The work does not vanish. It moves into direction, review, correction, governance, and content strategy.
The question for the next two years is whether brands invest in those capabilities or use AI as permission to remove people and process from the budget. The second option is cheaper right up until the image is inaccurate, the likeness is contested, the disclosure is missing, or the brand has to explain why its new campaign looks like everybody else's hallucination.
The legal environment is starting to create deadlines. The creative standard still belongs to the brand.
What This Field Note Is Based On
- New York State Senate, S.8420-A
- New York Governor's Office, synthetic performer disclosure law announcement
- European Commission, AI Act overview and timeline
- PetaPixel reporting on Francheska Pujols v. Rainbow Shops
- Reuters legal analysis of the New York law
This article is an industry briefing, not legal advice. The Pujols allegations were not adjudicated. The lawsuit was withdrawn while the parties sought a private resolution. Cost comparisons vary by production type, volume, quality, usage, and vendor.