Can you legally use AI images in your ads?
Commercial use, real people's likenesses, and somebody else's logo in your frame — the three questions your legal team will ask, answered plainly.
Can AI-generated images be used commercially?
Generally yes, and the risk almost never sits in the words 'AI generated'. It sits in what is inside the frame.
Three things get campaigns pulled: a real person who did not agree to be there, a trademark that is not yours, and source material you were never licensed to use. Those are the three worth checking before you spend, and they were risks long before any of this was generated. The technology changed the speed, not the law.
What about a real person's likeness?
A likeness is not a licence you can buy at checkout. It belongs to a person, and that person can change their mind — which is the part that makes it different from every other right you deal with.
So it should be recorded per person, carry an expiry, and be visible inside the product rather than living in somebody's inbox as a forwarded PDF. If you cannot see the consent from where the work is being made, you do not functionally have it.
Does that include people who no longer work with you?
Yes, and this is the case that catches people. A founder who has left, a model whose agreement covered one campaign, an employee who appeared in a video two years ago — none of those are open-ended permissions.
An expiry date that is visible is the difference between knowing that and finding out.
Can you put another brand's logo in the shot?
Not with us. A third-party mark on a product is refused outright — no 'for scale', no 'it's only in the background', no 'everyone does it'.
It is the cheapest refusal we make and it prevents the most expensive conversation you can have. The comparison shot that seems harmless in a studio looks different in a letter from someone else's counsel.
What about a location, a building, or someone else's artwork?
Same category, less obvious. Some buildings are protected. Some artworks are. A recognisable interior can carry rights that a photograph of it carries too.
The practical rule: if a real thing in the frame is identifiable and belongs to somebody, treat it as a right to check rather than a detail to add.
What does an attestation actually do?
It records the scope of a right you already hold. It does not create one and it does not transfer one.
If the underlying right is not real, the attestation is worth nothing to you and nothing to us. Saying that plainly is more useful than a checkbox that quietly implies otherwise, and a checkbox that implies otherwise is worse than no checkbox at all.
Who owns the output?
You do, and you should read that sentence carefully wherever you see it. The question worth asking any vendor is not only who owns the output — it is what happens to your inputs.
Your briefs, your uploads and your deliverables are not used to train or fine-tune a model. That is a contractual position, not a preference, and it is the one most worth checking in anybody's terms.
Why refuse before the work, and not at export?
A refusal at export is an invoice for something you cannot use. It is also the moment you have least time to fix it.
The check runs first, states the reason in one sentence, and leaves your brief intact with the routes still open: attach a mark you own, describe the style instead, record the consent. You lose a minute, not a budget and not a deadline.
What should your legal team actually ask for?
Four things, and they are all answerable. Where consent is recorded and how it expires. What happens to inputs. What the product refuses outright. And whether there is a record of who approved what, after the fact.
That last one is the one people forget and the one that matters when somebody asks a question about a campaign that shipped eight months ago.
What questions should you ask before a campaign ships?
Five, and they take a minute between them. Is every identifiable person in these frames covered by a consent that has not expired? Is every visible mark ours? Is every reference we used ours to use? Is there a record of who approved this? And would we be comfortable explaining any frame of it in public?
That last one catches most of what the first four miss. It is not a legal test and it is a good one.
How is this different from stock photography?
Stock arrives with a licence attached that answers most of these questions for you, which is what you are paying the premium for.
Generated work does not come with that document, so the responsibility sits with whoever briefs it. That is a real difference and it is not an argument against generated work — it is an argument for the checks happening inside the tool rather than in somebody's memory.
What changes if you are an agency?
The rights are usually somebody else's, which makes the record the important part.
You need to show a client not only what was made but what it was made from and who signed off. An approval that names a person, and a consent with a visible expiry, are worth more in that conversation than any assurance about the technology.
Where is the law still moving?
Disclosure rules, training-data claims, and how likeness protections apply to synthesised performances. These differ by country and are actively changing.
The practical response is not to guess the outcome but to keep records that would satisfy the stricter reading — what was used, who agreed, when it expires. Those records cost nothing if the rules stay loose and everything if they tighten.