Can you copyright an AI-generated product photo?
Partly — and the part matters. The Copyright Office's position, set out in its 2023 registration guidance and expanded in Part Two of its AI report in January 2025, is that human contributions to an AI-assisted work can be enough for authorship, but the analysis is case-by-case. Prompting alone is not enough. The Office has been clear that selecting prompts, even carefully crafted ones, does not yield a copyrightable work.
What does count, per the report's framework, includes using AI to assist an existing creative process, contributing "expressive inputs" (your own photograph being the obvious one), and creatively modifying or arranging AI-generated material. Hundreds of works containing AI-generated material have been registered since the 2023 guidance — the rejections get the headlines, but registration is routinely granted when a human contribution is disclosed and identifiable.
This maps cleanly onto how AI is actually used in e-commerce. Most sellers are not conjuring products from nothing. They are shooting a real product and using AI to change the background, clean up the garment, or place it on a model. Your original photograph is a copyrighted work you own, and it does not stop being one because you edited it.
| Workflow | Human authorship | Registration outlook |
|---|---|---|
| Text prompt only, no source image | None recognized | Not registrable |
| Your photo + AI background swap | Original photograph | Photo protected; disclose the AI parts |
| Your flat lay → AI on-model image | Original photograph | Underlying photo protected |
| AI output + substantial manual retouching | Your modifications | Case-by-case |
| Curated set of AI images, creatively arranged | Selection and arrangement | Compilation may qualify |
If you do register, you have a duty to disclose AI-generated content that is more than de minimis and to explain your own contribution. AI material gets excluded in the "Limitation of the Claim" section of the application. Disclosure is not optional, and a registration obtained by omitting it is a weak registration.
Why registration is the least important question for most sellers
Here is the practical reality: copyright registration matters when you intend to sue someone in U.S. federal court over copying. For a typical e-commerce seller, that scenario is rare, and when it does happen the thing being copied is usually the product, the brand, or the listing — not the artistic expression in the photograph.
The tools that actually protect a product catalog are different, and none of them depend on the photo being copyrightable:
What copyright covers
- Someone copying your exact image file
- Statutory damages, if registered before infringement
- Only the human-authored elements
- Requires registration to sue in the U.S.
What actually protects your catalog
- Trademark and trade dress on brand elements
- Marketplace image-theft takedown processes
- Contracts with photographers, agencies, and freelancers
- Your tool's license terms and indemnification
- Practical obscurity — most copying is of listings, not files
Amazon, Etsy, eBay, and Shopify all run intellectual-property complaint processes that operate on a much faster clock than litigation, and they generally do not require you to produce a registration certificate to report a seller who lifted your images. In day-to-day catalog defense, that is the lever you will actually pull.
There is one exception worth flagging. If you license images to others, sell them, or your imagery is a product — stock, templates, lookbooks, print files — then registrability moves to the top of your list, and pure prompt-to-image output is a genuinely weak asset.
AI-generated people: the risk that is actually growing
This is where the legal exposure has real teeth, and it is moving fast. Right-of-publicity law protects a person's name, image, likeness, and voice from unauthorized commercial use. It is a state-law patchwork in the U.S., and several states have been amending it specifically to cover AI.
- Tennessee's ELVIS Act (in effect since July 2024) covers unauthorized AI replication of voice and likeness and applies to any individual, not only performers.
- Arkansas amended its publicity-rights statute effective February 2025 to explicitly cover AI reproduction for commercial purposes.
- Washington updated its Personality Rights Act effective June 2026 to cover "forged digital likenesses," raising civil penalties and adding noneconomic damages.
- Federally, the NO FAKES Act (S. 4591) was unanimously advanced by the Senate Judiciary Committee on June 18, 2026. It would create liability for producing unauthorized digital replicas and for platforms that knowingly host them, with a DMCA-style notice-and-takedown process. It has not been enacted.
Advertising is squarely commercial speech, so the First Amendment carve-outs in these laws — news, parody, criticism — do not help a product listing. Two scenarios create exposure for e-commerce sellers.
1. Generating a model that resembles a real, identifiable person. Whether accidental or not, if a reasonable viewer would identify a specific individual, you have a publicity-rights problem. 2. Using a real person's photo as an AI input without a release covering AI-derived works. An old model release signed for a 2019 shoot almost certainly does not authorize generating new images of that person in 2026.
The workaround is not complicated: use fully synthetic models that are not derived from and do not resemble an identifiable individual, and when you do use real people, get a release that explicitly covers AI-generated and AI-modified derivative works. If you are commissioning a shoot today, add that clause. It costs nothing at signing and is expensive to retrofit.
Read your tool's license before you scale
Whatever copyright law says in the abstract, your day-to-day rights are governed by a contract you already agreed to. AI image tools vary enormously on terms that directly affect whether you can put an output on a product page at all.
Five clauses to check before you build a catalog on a tool:
| Clause | What to look for |
|---|---|
| Commercial use rights | Explicitly granted, not just implied. Some tiers restrict commercial use. |
| Ownership of outputs | Does the vendor assign output rights to you, or retain a license? |
| Rights after cancellation | Can you keep using images you generated once you stop paying? |
| Training on your uploads | Are your product photos used to train models other customers use? |
| Indemnification | Does the vendor defend you if an output triggers a third-party claim? |
The cancellation clause is the one that catches people. A brand that generates two thousand catalog images over a year and then switches tools needs to know whether those images can stay live. Search the terms for "termination" and "survive" and read what comes after.
Training rights matter more than they look, too. If you are shooting unreleased products, uploads that feed a shared training set are a confidentiality issue before they are a copyright one. Tools built for commercial e-commerce work — Retouchable among them — generally grant full commercial rights to outputs and do not train shared models on customer uploads, but "generally" is not a substitute for reading the agreement you signed.
The licensing risk you probably already have
Before worrying about AI-specific issues, it is worth checking a much more common exposure: images already in your catalog that you never had the right to use.
Dropshippers and resellers routinely publish supplier-provided images with no written license. A supplier catalog is not a license. When the supplier relationship ends — or when the supplier's own photographer comes asking — you have thousands of listings built on images you cannot document rights to. This is a far more frequent source of takedown notices than anything involving generative AI.
The same applies to stock photography with the wrong license tier (editorial-only licenses are common and prohibit commercial use), to influencer and UGC content reposted without a written grant, and to images from a previous agency whose contract transferred nothing.
Audit your existing catalog for images you cannot produce a license or contract for. Ironically, replacing undocumented supplier images with AI-generated ones from your own product photos usually improves your rights position — you own the source photograph, and you have a license agreement covering the output. That is more documentation than a scraped supplier JPEG ever had.
A practical checklist
If you are running an e-commerce catalog with AI in the workflow, this is the short version of everything above:
- Shoot your own source images where you can. Starting from a photograph you took is the single highest-leverage decision — it gives you a copyrighted work at the base of everything downstream and dramatically strengthens any registration claim.
- Keep your originals and your edit history. Human authorship has to be demonstrable. Archived RAW files and version history are the evidence.
- Use synthetic models, or get AI-inclusive releases. Never generate a likeness resembling a real identifiable person for commercial use.
- Read the license, especially the termination clause. Confirm commercial rights, output ownership, post-cancellation use, training rights, and indemnity.
- Disclose AI content if you register. Non-disclosure undermines the registration you were trying to obtain.
- Audit for undocumented images. Supplier and stock images without a paper trail are your most likely real-world problem.
- Check platform and advertising rules separately. Marketplace policies on AI-generated and materially-altered imagery are their own requirement, independent of copyright law, and they change.
The overall picture is less alarming than the headlines suggest. A seller who photographs their real products, uses AI to handle backgrounds and staging, avoids real people's likenesses, and keeps their paperwork is in a stronger legal position than most catalogs were five years ago.