Most brands ask the wrong first question about AI images. They ask whether the image looks good enough. The better question is whether the brand can actually use it, own it and defend it once it is running in paid media.
Commercial use of AI generated images comes down to five checks: the tool's terms, who owns what, whether real people or brands show up in the picture, whether the ad needs a disclosure, and who signs off. None of them is hard. All of them are cheaper to settle before the campaign than after.
This is a working checklist, not legal advice; your lawyer has the final word. For the wider questions buyers ask about AI imagery, see our AI-Based Images FAQ.
Do the tool's terms allow commercial use?
Start with the contract, not the picture. Every image tool publishes terms that say what you may do with the output. Some plans allow commercial use; some free tiers limit it or require credit. The rules can differ between plans of the same product.
Read three things: whether your plan covers commercial use, whether the vendor offers any protection if someone claims the output infringes their work, and whether your uploaded reference images may be used for training. Save a copy of the terms with the date. Terms change, and you want to know what applied on the day you made the image.
A vendor's permission is a license to use the image. It is not the same as owning the copyright, which is the next question.
Who owns an AI-generated image?
In the United States, ownership depends on human authorship. The U.S. Copyright Office published its report on copyrightability of AI outputs on January 29, 2025. Its conclusions are plain: copyright protects original expression by a human author, even when the work contains AI-generated material, but it does not extend to purely AI-generated material or to material where human control over the expression is insufficient. The report adds that, based on current generally available technology, prompts alone do not provide sufficient control. Each case is analyzed on its own facts (U.S. Copyright Office, Part 2: Copyrightability).
The same report says human authors keep rights in what they contribute that is perceptible in the output, in the creative selection and arrangement of material, and in creative modifications of the output.
What this means in practice: a raw image straight from a prompt may carry little or no copyright protection, so a competitor could copy it. An image that a person has composited, retouched, rebuilt and combined with original elements stands on firmer ground. Keep your working files and layered documents. They are the record of human contribution.
What is actually in the picture?
AI tools mix up things a human would never mix up. Before an image runs, look for:
- Real people. A face that resembles a known person, or a model you did not cast, can create likeness problems.
- Other brands. Logos, distinctive packaging, a recognizable product shape or a famous building sitting in the background.
- Your own product. If the ad shows what you sell, the product must look like what the customer will receive: correct label, correct color, correct proportions. An invented detail on a package is a false claim, not just a flaw.
- Text. Stray or garbled lettering, especially anything that looks like a claim.
This is where a retouch and compositing pass earns its fee. We often keep the real product photo or 3D render and let AI build the setting around it, so the product stays exact and the rest is free to change. Our post on how to art-direct AI imagery so it matches your brand guidelines covers the briefing side of that.
Does the ad need to say it is AI?
It depends on where the ad runs and what is in it. There is no single global rule, so check each market.
One concrete example: a New York law that took effect on June 9, 2026 requires a conspicuous disclosure in certain commercial ads that use an AI-generated human performer, called a "synthetic performer." It exempts expressive works such as films and television, audio-only ads, and ads that use AI only for translation, and it appears to exclude ads that merely enhance or alter images of real people, although regulators have not yet clarified how it will be read (Gala Law summary).
A related point from the Federal Trade Commission: its final rule announced August 14, 2024 bans fake reviews and testimonials, including AI-generated ones (FTC press release). If an AI image or video presents an invented person as a real customer, you are in that territory.
Ad platforms also publish their own labeling rules, and they change. Check the current policy for each platform you buy on.
Who signs off, and what do you keep?
Decide the owner before production, not after. Someone on the brand side should approve the final image against the checklist, and the approval should be written down.
Keep a short file for each campaign: the tool and plan used, the date, the terms as they stood, the prompts and reference images, the layered working files, the list of checks and who approved them. If a platform rejects an ad or a third party objects, this file saves days.
Practical takeaway
Before an AI image goes into a paid ad, run it through six questions:
- Does our plan's license cover commercial use, and did we save the terms?
- Have people added real work to the image, and do we have the layered files?
- Is any real person, brand or foreign product visible?
- Does the product look exactly as sold?
- Does any market or platform we buy in require a disclosure?
- Who approved it, and where is that recorded?
If you answer yes to all six, you are in good shape. If not, the fix is usually small and cheap at this stage.
If you want a second pair of eyes on AI imagery before it runs, you can see how we work at 35milimetre.com.