Who owns your AI UGC video? A rights map for inputs, outputs, and library footage

Not legal advice, and a moving target

This is a framework for thinking about rights in AI-generated ads, with links to primary sources, written in September 2026. It is not legal advice, and the law here is changing quickly in several countries. Use it to ask your counsel better questions, not to replace them.

The reason to care is practical. An ad that performs will be copied, shared, and possibly disputed. Knowing what you can and cannot enforce, and what someone else could enforce against you, is part of scaling it.

Question one: what rights did you have in the inputs?

This is where almost every real problem lives. A reference-driven render processes an actual video clip. A character render processes an actual portrait. If you did not have the right to use those inputs for advertising, no downstream term fixes it.

Likeness rights deserve a specific mention. Many jurisdictions protect a person's right to control commercial use of their face and voice, separately from copyright, and several have extended that protection to digital replicas. A portrait you are allowed to post is not necessarily a portrait you are allowed to animate into a spokesperson. Get advertising use of the likeness in writing, with a duration.

Question two: what do the tool's terms grant?

Every AI video tool sits on top of one or more model providers, and each layer has terms. The questions to answer for any tool, including ours, are the same.

• Does the tool claim any ownership of the output? Most do not, but read the clause.
• Can you use outputs in paid advertising, and on which platforms? Some model providers restrict commercial use on lower tiers.
• Does the provider retain or train on your inputs? This affects confidentiality more than ownership, but it matters for unreleased products.
• Is uniqueness promised? It never is. Similar prompts and references produce similar outputs for different customers, and the tool will not guarantee exclusivity.
• Are library assets licensed to you or only to the tool? A clip you can use inside the tool may not be a clip you can download and use elsewhere.

Our own draft terms say it plainly: you keep whatever rights you hold in what you submit, we and the providers needed for your task get a license to process it, outputs may be similar to other outputs or ineligible for copyright, and library materials remain subject to their own licenses. Read them at /terms. They are marked as drafts pending operator details, which we say on the page rather than hiding.

Question three: is the output copyrightable at all?

In the United States, the Copyright Office published Part 2 of its report on copyright and artificial intelligence on January 29, 2025, and its conclusions are the clearest public statement on this so far. Human authorship remains a requirement. Material generated entirely by AI is not copyrightable. Prompts alone, under the technology as it currently works, do not give a person enough control over the output to count as authorship, because the same prompt can produce different results. Where a human contributes expressive elements, such as selecting and arranging AI outputs, editing them, or combining them with human-authored material, those contributions can be protected, and the analysis is case by case.

For an AI UGC ad, that means the synthetic performance itself may not be protectable, but the ad as a whole often includes plenty of human authorship: the script, the edit, the selection and arrangement of cutaways, the captions, the product footage you shot. Those parts are yours in the ordinary way. The practical consequence is that the raw render is not the asset. The finished, edited ad is.

Other jurisdictions differ. The UK has a specific provision for computer-generated works that is under review; the EU's approach runs through its member states' human-authorship requirements. If you operate in several markets, assume the least protective answer and build value in the human layers.

What this means in practice

• Compete on the trail, not the render. Your reference library, beat sheets, cast, and test data are what a competitor cannot copy, and none of them depend on copyright in a render.
• Register what you can. The edited ad, the script, and your product footage are registrable in the ordinary way in the US. The synthetic performance alone probably is not.
• Do not rely on exclusivity. If a tool's stock avatar appears in a competitor's ad, that is expected. A character built from a portrait you own and control is the only way to have a face nobody else can use.
• Watch content credentials. Provenance metadata in outputs is not an ownership claim; it is a disclosure signal that platforms read. Do not strip it.

The rights file: keep this for every render

This takes five minutes per campaign and turns a possible dispute into a document. It also makes handoffs painless when a media buyer, an agency, or an acquirer asks what they are allowed to do with your creative.