Who Owns the Copyright on AI-Generated Ad Content? A Usage Rights and Liability Guide for Agencies

Three days before the campaign goes live, the brand's legal department asks one question: is that woman in the hero image a real model, or something AI generated — and either way, who owns the copyright, and whose usage rights did we actually get? The production team loved the image, the brand manager signed off, the production calendar is locked. But nobody asked this question at the brief stage. Launch slips by a week, the agency tells the client 'we're looking into it,' and nobody actually has a clear answer.

##Where does the copyright and usage-rights question actually sit in an agency's production line?

There are really two separate questions here, and most teams blur them together. The first: does the AI tool itself grant you commercial usage rights over what it generates? Almost every serious platform agencies actually use is explicit about this — the account holder can use generated images commercially in ads, print, and social media; it's a clause written plainly into the platform's terms of service, and it's the first document worth reading before a brief even starts. The second question, and the one that actually gets missed, is this: if the generated image resembles an existing real person, a trademarked brand asset, or someone else's copyrighted work closely enough, the commercial usage right the platform gave you does not make that risk disappear. The platform grants you rights over its own output; it does not guarantee you haven't infringed on someone else's rights — those are two different things, and an agency needs to tell them apart at the brief stage.

This distinction gets critical specifically around faces, logos, a recognizable public figure's silhouette, or a known location. In a generic product-catalog image — a 'model' holding a product, resembling no one in particular — this risk is close to zero. But in a hero shot, where the character being generated is meant to become the face of the campaign, and that face isn't random but a distinct identity that repeats across the whole campaign, the situation changes; how that face was derived and how close it sits to a real person becomes a question worth asking. Catching this at the brief stage costs far less than having legal halt the campaign after the fact.

##A real scenario: a food brand's social media campaign

Take a hypothetical but realistic example: a food brand asks for 15 social media frames, each showing a 'consumer' holding the product. Fourteen frames use AI generation — generic, stylized characters of different ages and backgrounds, none referencing a specific real person. But legal flags exactly one frame: the hero image, destined for the main ad and outdoor billboards, where the generated face resembles a well-known social media personality a little too closely. The fix has three parts: the fourteen frames stay as they are, the hero frame gets reshot on the TVN side with a real actor under a signed release, and the asset library on the Orni side now tags every image by source — AI-generated or real shoot — and by licensing status.

What changes in the calendar: the hero frame adds three days, but the other fourteen frames are already delivered, so the campaign launches on schedule overall. More importantly, the next brief treats this check as a question asked up front, not a risk left to be discovered at the end.

##How it's set up (briefly)

• ▸At the brief stage, every frame gets a 'usage rights tier': internal use, social media, paid media, outdoor / long-term brand asset
• ▸Every AI output is tagged with production metadata: which model, which date, which license terms
• ▸Brand-critical, high-visibility frames (hero image, outdoor, paid media) require a real shoot and a signed talent/location release
• ▸Legal's pre-launch review happens against a standard checklist, not inside an email thread
• ▸The asset library stores every image's licensing status centrally, and flags releases that are approaching expiry

##The honest limit: where AI can't guarantee copyright clearance

Worth being blunt here: the commercial usage right an AI platform grants you is real and valuable, but it doesn't cover everything. Third-party rights — a person's right of publicity, a brand's registered imagery, a real location that requires permission to use — sit outside the scope of the platform's terms of service. That's exactly why an agency keeps a production house (TVN) alongside the AI line: for work that genuinely needs a real actor with a signed release, or a real location with shoot permission. And it keeps a software house (Orni): to hand the job of tracking which image was produced under which license, shared with whom, and when it needs renewing, to a system instead of to someone's memory. That's not a gap in AI production — it's an honest division of labor.

It's also worth saying plainly: the regulatory ground under AI-generated content is still shifting in most jurisdictions, and nobody can promise you a fully settled legal answer today. Given that, the cost of being careful — tagging sources, keeping releases on file, routing brand-critical frames through a real shoot — is low, while the cost of guessing wrong on a campaign running in paid media or outdoor is not.

##Frequently Asked Questions

>Who owns the copyright on an AI-generated ad image?

Most major commercial platforms grant the account holder commercial usage rights over what they generate — that's written plainly into the platform's terms of service. But that doesn't mean every use is risk-free; third-party rights (a person's face, a trademarked brand asset) sit outside the scope of that license entirely. For a high-budget or long-running campaign, actually reading your platform's commercial terms — and asking legal counsel when it matters — beats assuming.

>What happens if an AI-generated face resembles a real person?

Being AI-generated doesn't mean the image belongs to no one — if the generated face is recognizably close to a real person, that person's right of publicity can come into play. The risk is low for generic, stylized output and high for hyper-realistic faces that become a repeating 'character' in hero assets. The practical fix: keep brand-critical frames clearly stylized, or shoot them with a real actor under a signed release.

>How does an agency actually guarantee a client usage rights on AI content?

A verbal assurance proves nothing; what protects both the client and the agency is an asset library that records every image's source, production date, and license terms. If a dispute ever comes up, you need a dated record, not someone's memory. That's why rights-tracking isn't a nice-to-have — it's part of how any agency serious about AI production actually operates.

>Does mixing real shoots with AI generation in the same campaign create extra copyright risk?

No — if anything, it reduces it, because as long as each asset type is tracked under its own licensing lineage, the risk sources don't get mixed up. The real risk comes not from hybrid use but from nobody recording how an image was actually produced. A properly set-up hybrid line actually carries less uncertainty than pure AI production, because the brand-critical frames are already produced at the highest safety tier — real shoot, signed release.

If you want to talk through which frames in your own campaign are usage-rights sensitive, or which images are currently running with no licensing record at all, bring the brief: reach out through https://luvi.agency. The hybrid line — AI production, real shoot, and rights-tracking — sits on the same team.