Who Owns the Copyright to Your AI Virtual Try-On Images?

This question sits next to two others fashion retailers have had to think about this year: whether an AI generated image needs a visible disclosure label, covered in an earlier post on the EU AI Act's rule and its California counterpart. Disclosure and ownership are different problems. A label tells a shopper what they are looking at. Copyright determines whether you can stop someone else from using the same image.

Can you copyright a purely AI generated image?

No, not by itself. US copyright law has always required a human author, and the Copyright Office has applied that rule consistently to AI output since generative image tools went mainstream. Typing a detailed prompt, even a long, carefully worded one, does not count as authorship in the legal sense, because the tool is the one making the actual creative decisions about composition, lighting, and pose.

This got tested directly in Thaler v. Perlmutter, a case over an image an AI system generated with no human listed as author at all. The Supreme Court declined to review the case in March 2025, which left the lower court's ruling in place: a work needs a human author to be copyrighted, full stop. That closed the door on the argument that AI output should qualify simply because a person set the process in motion.

What is the human authorship rule, exactly?

The Copyright Office published a report on AI and copyrightability in January 2025, the most detailed guidance it has issued on this exact question. Its conclusion is narrower than a flat yes or no: a work is protectable to the extent a human exercised creative control over its expressive elements. Prompting alone usually does not supply that control, no matter how specific the prompt.

• A prompt-only image gets no protection. Typing a description and accepting whatever the model produces does not create an author under the Office's rule.
• Selection and arrangement can count. Choosing one generated image over many, then arranging it alongside other elements on a page, can support a thin layer of protection over that specific arrangement.
• Edits and additions can count too. Cropping, retouching, compositing, adding text or a logo overlay, these are the kinds of creative choices the Office recognizes as human authorship, applied to the parts you actually changed.
• Registration requires disclosure. If you apply for copyright registration on a work that includes AI generated material, you have to identify and exclude the AI generated portion from your claim. You cannot register the whole image as if a person made every part of it.

What does this mean for a store's try-on images specifically?

A retailer generating hundreds of try-on images across a catalog is, in the strict legal sense, mostly generating uncopyrightable output. If a competitor fed the same garment photo into a similar tool with a similar prompt, the resulting image would likely be just as uncopyrightable on their end, and there is no infringement claim to make against them over the picture itself.

That sounds worse than it is in practice. Two things limit how much this actually matters day to day. First, nobody is going to reconstruct your exact garment photo and reproduce your exact styling choices by accident, so the realistic risk is narrower than the legal exposure suggests. Second, the image is rarely the only thing worth protecting on a product page, which the next section covers.

What can a merchant do to strengthen its claim?

1. Shoot your own garment reference photos rather than sourcing them from a supplier's catalog. A photo you took is copyrighted the moment you take it, independent of anything that happens to it afterward.
2. Curate results instead of publishing the first generation. Reviewing a batch of outputs and selecting the ones that best represent the garment is a creative judgment the Copyright Office recognizes.
3. Crop, color correct, or composite the final image before it goes live. Even light editing is the kind of human touch that can support a claim over that specific file.
4. Keep a record of what you changed and why. If you ever need to register or defend a claim, a simple note of your prompts, rejected outputs, and final edits is the documentation the Office expects to see.
5. Do not assume a vendor's terms of service settle the question for you. Some AI tools grant broad usage rights in their terms regardless of copyright status. Read what you actually signed up for rather than assuming ownership.

What stays protected even without image copyright?

Copyright is one legal category among several, and it is not the one doing most of the real protective work for a fashion brand. Trademark protects your brand name and logo regardless of how a product photo was made. Trade dress can protect distinctive packaging or a signature visual style. Design patents, where they apply, protect the ornamental design of the garment itself, separate from any photograph of it.

None of those depend on whether the try-on image sitting on your product page passes the human authorship test. A competitor copying your generated image is a narrow, mostly theoretical risk. A competitor copying your garment design or trading on your brand name is a different problem, and one the law already handles without needing the photo itself to be copyrighted.

How does this apply to images generated with Corlen?

Corlen's virtual try-on starts from the merchant's own garment photo and a person's photo, not from a text description alone. That garment photo is your original photography, protected under ordinary copyright rules from the moment you shot it, no different from any product photo you have ever taken. Nothing about running it through Corlen changes that.

The generated result follows the same human authorship rule described above. A store that picks which results to publish, crops them to fit a product page layout, or pairs them with its own styling and copy is adding exactly the kind of creative control the Copyright Office looks for. A store that auto-publishes every raw output with no review is leaning entirely on a category of image that carries the least legal protection.

None of this is legal advice, and the practical stakes for most stores are lower than the legal theory suggests: the realistic risk is a competitor's near-identical picture, not a wave of lawsuits over indistinguishable try-on photos. A store with real exposure, like a licensing dispute or a large catalog it plans to register, should talk to a lawyer rather than rely on a blog post to settle the question.

On a Shopify store, virtual try-on runs directly on the product page through Corlen's Shopify app, generated from your own garment photos the same way described above. A team building on a custom stack gets the same generation behavior through the developer API, so the ownership question comes down to your own editorial process either way, not which platform you run on.

For the separate question of what disclosure a shopper needs to see, read the earlier posts on the EU AI Act's disclosure rule and its California counterpart. For how to shoot the source photos this whole process starts from, see how to photograph clothing for virtual try-on.

See the whole process on your own photo: try Corlen for free, no account needed for a first result.