ChatGPT’s Virtual Try-On Creates a New Copyright Problem for Fashion Brands

For a small fashion label, product photography may be among the clearest pieces of intellectual property it actually controls.

The garment itself is another matter. Under the useful-article doctrine, copyright generally does not protect the utilitarian aspects of clothing, although separable artistic features may qualify for protection. A photograph of that garment is different. The U.S. Copyright Office explainsthat copyright in a photograph can protect the photographer’s original expression, while copyright in a useful article does not extend to the article’s utilitarian aspects.

That distinction has become more important this week.

On October 1, OpenAI launched virtual try-on in ChatGPT, allowing users to see clothing and accessories on themselves using a selfie or other reference photo. OpenAI’s release notesdescribe the new shopping features, including virtual try-on and the ability to save reference photos for future try-ons.

OpenAI’s shopping documentationalso says users can upload a picture of clothing or an accessory directly into a conversation and ask ChatGPT to show how it might look on them.

The result is a deceptively simple new workflow: a shopper sees a garment, supplies an image of it to an AI system, and gets a new image showing the garment on their own body.

For fashion brands, the legal question is not whether that is technologically possible. It is who owns the rights in the image going into the system, what the system is permitted to do with it, and where the boundary lies between ordinary consumer use and copyright infringement.

The image is the issue

The most important distinction is between the clothing and the photograph.

Copyright generally does not give a designer a monopoly over the functional design of a dress, jacket, or pair of trousers. The Copyright Office identifies clothing as a useful articleand explains that copyright in a photograph of a dress protects the artistic expression in the photograph, not the underlying dress design itself.

That makes product photography unusually significant for smaller labels.

A carefully produced campaign image can contain protectable expression in the lighting, composition, styling, model positioning, background, and other photographic choices. The fact that the image happens to depict a garment does not strip that expression of copyright protection. The Copyright Office's guidance for photographerslikewise recognizes copyright protection for original photographs.

ChatGPT's new shopping tools create an obvious new pathway for those images to be reused.

A shopper does not necessarily have to begin with a product that ChatGPT itself surfaced. OpenAI saysthat a user can upload a picture of clothing or an accessory directly in a conversation and ask ChatGPT to show how the item could look on them.

In practical terms, that means an image appearing on a retailer's or designer's website can become the input for an AI-generated try-on.

That does not mean OpenAI has declared every public fashion photograph licensed for AI use. Nor does it establish that every consumer who uploads one is infringing copyright. Those are separate legal questions.

But it does create a use of product photography that many photographers and fashion labels did not have in mind when those images were commissioned.

Who authorized the upload?

The first legal question may actually sit with the user rather than OpenAI.

OpenAI's Terms of Useplace responsibility on users for having the necessary rights, licenses, and permissions for material they provide as input. That matters because the person uploading a product photograph may have no relationship with the photographer or copyright owner.

A consumer who screenshots a campaign image from a designer's website does not necessarily acquire any right to reproduce that photograph merely because the image was publicly accessible.

At the same time, that observation does not resolve whether the particular upload is infringing. Copyright law contains limitations and exceptions, including fair use, and the analysis is highly fact-dependent. 17 U.S.C. §107identifies the familiar four fair-use factors: the purpose and character of the use, the nature of the copyrighted work, the amount used, and the effect on the potential market.

The generated output can raise additional questions about what expressive material was actually reproduced.

That makes the virtual-try-on scenario different from simply asking whether an AI company trained a model on copyrighted images.

Here, the potentially protected photograph may be supplied by an individual user for a specific generation request. The legal analysis could therefore turn on the conduct of the user, the system's processing of the input, the resulting image, and what happens to that output afterward.

There is no simple rule that answers all of those questions.

The output creates a second problem

Copyright is not the only potential concern.

OpenAI itself warns that try-on images may not represent the product or the user's appearance exactly and do not guarantee fit or size. OpenAI's shopping FAQincludes those limitations.

That disclaimer is sensible from a shopping perspective. It is also significant from a brand perspective.

An AI-generated image might alter the proportions of a garment, change a print, distort a logo, invent details, or make a fabric appear to drape differently from the real product.

A shopper looking at the resulting image may still understand the garment to be the brand's product.

That does not automatically create a trademark claim. Trademark law is not a general right to control every depiction of a branded product, and a consumer's private generation of an image is different from a brand advertisement or other commercial use.

But the analysis changes if an inaccurate AI rendering is subsequently used to advertise, promote, sell, or otherwise market the garment. Depending on the facts, trademark, false-advertising, trade-dress, or other claims could become relevant. The USPTO's explanation of trademark infringementemphasizes the role of consumer confusion in infringement analysis.

The important point is that the legal risk is not simply “AI makes a bad copy.” It is the possibility that an AI-generated representation of a branded product enters commerce while looking sufficiently authoritative that consumers assume the brand made or approved it.

Then there is the data question

The try-on feature also creates a separate privacy and data-governance issue.

OpenAI says reference photos used for virtual try-ons are saved for future try-ons. Users can change or delete those images through Settings → Personalization → Reference photos. OpenAI's release notesdescribe that functionality.

OpenAI's consumer data policyalso says that content submitted to ChatGPT—including images and files—may be used to improve model performance, depending on the user's settings.

That policy concerns the user's submitted content, most obviously the selfie or full-body photograph used for the try-on. But it underscores a broader point about the new workflow: an image supplied for what looks like a single shopping interaction may be retained and processed beyond that immediate moment.

For a fashion label, the questions are therefore not limited to copyright.

Who created the original product photograph? Who owns it? What rights were granted to the label? Can the label control downstream uses? What happens if a consumer uploads the image to an AI system? And what happens if the resulting image is later used publicly?

Those questions become especially important where the label does not actually own the underlying photography. A brand's right to use an image on its website does not necessarily mean it owns every copyright interest in that image.

A gap in the regulatory map

The new technology also exposes an awkward gap between several existing legal regimes.

Copyright law addresses rights in the photograph. Trademark law can address certain confusing or misleading commercial uses of a brand. Privacy law can address the shopper's own image in some circumstances. New synthetic-media laws address particular categories of AI-generated content.

But none of those regimes was designed specifically around this scenario:

A consumer finds a fashion photograph online, supplies it to a general-purpose AI system, and receives a new image showing themselves wearing the depicted garment.

New York's synthetic-performer disclosure law illustrates the problem. The law, which took effect in June 2026, requires disclosure in advertisements containing AI-generated synthetic performers. New York Governor Kathy Hochul's announcementdescribes the law as the first of its kind in the United States.

Its statutory framework is directed at synthetic performers and advertising transparency rather than the broader universe of AI-generated fashion imagery. Whether a particular AI fashion image falls within an existing disclosure requirement therefore depends on facts that may have little to do with copyright ownership of the original garment photograph.

What fashion brands should watch

Three developments are worth watching.

First: platform-level controls. Will OpenAI eventually give photographers, brands, or other rights holders a mechanism to opt out of particular uses of publicly available imagery? At present, there is no publicly documented, fashion-photography-specific system that functions as a universal “do not use this image for AI try-on” signal.

Second: the first test case. The interesting lawsuit would not necessarily be another dispute over AI training data. It could instead involve the downstream use of a particular copyrighted product photograph as an input to a consumer-facing generation system, followed by a disputed output or commercial use.

Third: the line between private experimentation and commercial exploitation. A consumer privately asking what a dress might look like on them is legally and factually different from a competitor using an AI-generated version of another brand's campaign image, or from a seller using an AI rendering to market a product.

That distinction is likely to matter enormously.

The practical takeaway

For fashion brands, the immediate lesson is not that OpenAI has made product photography “free to use.” The law does not support such a sweeping conclusion.

The more defensible conclusion is narrower—and potentially more consequential.

Public product photography can now be fed into a mainstream consumer AI workflow designed specifically to transform fashion imagery into new shopping content.

Whether a particular use is lawful will depend on the rights in the photograph, the user's conduct, the nature of the AI processing, the resulting image, and what happens to that image afterward.

For small labels, that means the first step is not necessarily sending a cease-and-desist. It is understanding what you actually own.

Know who owns the photographs. Review photographer agreements and licenses. Know what rights were granted and for what purposes. Monitor how your brand's imagery is being used. And recognize that the legal significance of a product photograph is changing as AI turns an image that once existed primarily to sell a garment into an input that can generate entirely new representations of it.

The virtual fitting room has arrived. The harder question is who owns the photograph after the fitting room disappears.

Sources

Primary sources

  1. OpenAI — ChatGPT Release Notes— October 1, 2026 launch of virtual try-on, saved reference photos, deletion controls, and availability on web and mobile.
  2. OpenAI — Shopping with ChatGPT Search— Details on the virtual try-on workflow, uploading clothing/accessory images, reference photos, and limitations of generated try-on images.
  3. OpenAI — Data Usage for Consumer Services FAQ— OpenAI's policies concerning the use of consumer-submitted content, including images, for model improvement.
  4. OpenAI — Terms of Use— User representations concerning rights, licenses, and permissions for content submitted as input.
  5. U.S. Copyright Office — Useful Articles— Copyright treatment of useful articles, including clothing, and photographs depicting useful articles.
  6. U.S. Copyright Office — What Photographers Should Know About Copyright— Copyright protection and ownership considerations for photographs.
  7. 17 U.S.C. § 107 — Fair Use— Statutory framework governing the four fair-use factors.
  8. USPTO — Trademark Infringement— Trademark infringement and the role of consumer confusion.
  9. New York Governor's Office — AI-Generated Synthetic Performers Law— New York's 2026 legislation concerning disclosure of AI-generated synthetic performers in advertising.

Reporting and launch coverage

  1. TechCrunch — “ChatGPT can now virtually try on clothes for you”— Independent reporting on the October 1 launch and the ability to use uploaded product imagery for try-on.
  2. PYMNTS — “OpenAI Adds Virtual Try-On Feature to ChatGPT” — Additional contemporaneous reporting on the virtual try-on launch.
  3. OpenAI — Introducing ChatGPT Images 2.5— Background on the image-generation model underlying the new experience.