Which permissions does a brand need before feeding creator clips into AI tools that generate new ad variations?

A Usage License Is Not a Training License

Most creator agreements were written to answer a distribution question. They say where the clip can run, for how long, in which markets, and whether it can be pushed with paid spend. None of that language contemplates a system reading the footage frame by frame and learning from it. When your team drops a folder of creator clips into an AI editing or generation tool, you are doing something the contract never described, and silence in a contract is not permission. Related: How do you handle usage rights when a UGC clip moves from organic posts into paid ads?

We hear the same objection often: the brand paid for the clip, so surely it can do what it likes with it. Usage rights do not work that way. A license grants a specific list of uses and reserves everything else to the creator, who still owns the copyright in the footage. If training, fine-tuning, or synthetic generation is not on that list, the safe assumption is that you do not have it. That is not a scare tactic, it is simply how a grant clause is built. Related: Why Usage Rights Matter for Brands

Keep reading: Why Usage Rights Matter for Brands, Getting Clear Rights From Creators, Tracking Content Rights Without Chaos. See how RightsRoster helps you usage-rights and license tracking for creator ugc clips.

The Three Permissions AI Work Actually Needs

Split the ask into three parts, because tools and vendors treat them very differently. First, ingestion and training: can the footage be uploaded to a system that learns from it, and does the vendor keep it after the job finishes. Second, derivative generation: can the tool produce new frames, new cuts, or entirely synthetic scenes built from the clip. Third, output usage: can whatever comes out the other side run in your ads under the same terms as the original file.

Teams usually secure the third one by accident and miss the first two entirely. A creator who agreed to twelve months of paid social is picturing their clip running as filmed. They are not picturing their kitchen, their lighting, and their delivery becoming raw material for a system that produces ten variants they never shot. Asking for all three plainly, in one short readable paragraph, tends to go far better than asking for a vague right to modify and hoping it stretches.

Likeness and Voice Sit Outside the Clip Itself

Copyright in the video and rights to the person in the video are two different things. When a tool regenerates a creator's face, clones their voice for a new script, or places their likeness in a scene they never filmed, you are using personality rights, not just footage. Several states have specific statutes covering name, image, voice, and likeness, and this area has been getting more legislative attention rather than less. A footage license on its own will not carry you there. Related: Getting Clear Rights From Creators

The practical version is a separate consent line that names the behavior directly: whether the creator agrees to voice cloning, face replacement or reanimation, and synthetic performances of scripts they never recorded. Some creators will say yes for an added fee. Many will say no. A few will say yes for the launch campaign but not for anything that outlives it. All three answers are workable, as long as you have the answer in writing before the tool runs.

How to Write It Into the Agreement and Track It

Add a short AI clause rather than burying the point inside the modification language. Keep it readable: name the permitted activities, name the vendors or vendor categories where you can, state whether the footage may be retained by the tool after processing, and state whether outputs expire when the underlying license expires. That last point catches a lot of teams, because a generated variant feels like brand new content and quietly keeps running long after the source clip was pulled. Related: What Happens When Rights Expire

Then track the generated files the way you track the originals. In practice that means each output carries the source clip's identity, the creator, the expiration date, and the AI permission flag, so anyone pulling assets can see the constraint without opening a contract. We built RightsRoster around that idea because the failure we saw most often was never a badly written agreement. It was a perfectly good agreement that nobody could connect to the file sitting in the shared drive.