So, Who Owns Your AI-Generated Face and Voice?

The most accurate answer is not “the AI,” and it is not yet “the person owns everything.” Under the law described here, a person’s face or voice is not automatically controlled by whoever owns the photograph, recording, or model output that depicts it. Control depends on the identity being reproduced, the authorization given, the place and purpose of the use, the route by which the replica reached the public, and the remedy available in the relevant jurisdiction. The Copyright Office’s recommendation and the introduced SOUL Act would make that control more direct, but they are proposals with different boundaries—not current universal law.

That answer may feel less satisfying than a single ownership label. It is also more useful. An AI-generated face or voice is not one legal object. It is a relationship among a person, a source work, a synthetic representation, an audience, a contract, a platform, and a government rule. The question “who owns it?” becomes answerable only after those relationships are separated.

The Present Answer Is Layered, Not Singular

Start with the source media. A photograph, sound recording, or video may be an original work protected by copyright. The person shown or heard may own that work, may have licensed it, or may have no ownership interest in it at all. Copyright can therefore control copying of the work or protectable expression from it without automatically giving the copyright holder the right to make the person appear to speak, sing, endorse, or participate in a new event.

The voice example makes the distinction clear. An independently generated recording can imitate the sound of a performer without copying the actual sounds fixed in an earlier recording. The Copyright Office explains that copyright treats the authored sound recording and the human voice differently. Section 114(b) permits certain independent sound-alike recordings, while a person’s voice is a personal characteristic rather than a copyrighted work merely because it has been recorded. That does not settle whether the imitation is authorized. It means that a copyright answer and an identity answer are asking different questions.

The current identity answer may come from state privacy or publicity law. Privacy doctrines can address personal dignity, false portrayal, or related injuries. Publicity doctrines may address the unauthorized use of an identity for another party’s advantage. But the coverage varies. Some jurisdictions protect a voice or broad identifying attributes; others emphasize commercial use, fame, professional performance, or commercial value. Speech defenses, postmortem terms, intermediary rules, jurisdiction, and remedies vary as well. A person can have a stronger or weaker path depending on where the claim is brought and what kind of use occurred.

Federal law supplies other partial routes. A synthetic advertisement may raise Federal Trade Commission or Lanham Act issues if it deceives consumers, creates false endorsement, or causes confusion about sponsorship. An AI voice used in a robocall may fall within communications regulation. Those tools can be important, but they are tied to commerce, consumer confusion, telephony, or another specific jurisdictional trigger. A noncommercial replica used to humiliate a private person may not fit those federal routes even if the harm is substantial.

Contracts add a separate layer. A performer may authorize a voice for one recording, game, advertisement, or production without authorizing every later use. The meaning of permission depends on scope, duration, context, compensation, and the bargaining process that produced it. A contract can give a company a lawful use while leaving the person in control of other uses. It can also be written so broadly that the practical question becomes whether the person understood what they were signing and whether another legal limit applies.

This is why the present answer cannot be summarized as either “copyright protects your face” or “there is no protection.” Copyright may protect the file. State law may protect the identity. A contract may authorize a bounded use. Federal consumer or communications law may address a deceptive channel. The same synthetic output may implicate several of these at once, or none of them clearly. The evidence supports a conclusion about fragmented coverage, not a universal conclusion about the result of every dispute.