Tennessee's ELVIS Act reaches the tool, not just the output

The process wall

Before generating a real person's voice or face, three different documents already require a consent step, aimed at three different points in the process. Tennessee's 2024 amendment to its personal rights statute, retitled the Ensuring Likeness, Voice, and Image Security Act, adds a definition of “voice” covering any sound “readily identifiable and attributable to a particular individual, regardless of whether the sound contains the actual voice or a simulation.” It gives every individual “a property right” in their name, photograph, voice or likeness, and creates civil liability not only for using an unauthorized likeness but for distributing “an algorithm, software, tool, or other technology” whose primary purpose is producing a particular identifiable person's photograph, voice, or likeness without authorization.

What the documents show

That statute reaches past the finished output to the generating tool itself, which is a different target from the SAG-AFTRA 2023 TV/Theatrical summary agreement, which regulates a specific employment relationship. The agreement defines an “Employment-Based Digital Replica” and requires consent for its creation and, separately, for each use, stating consent “must be clear and conspicuous” and specifying that consent obtained during a performer's lifetime continues after death unless it was explicitly limited. Civitai's own real-people policy regulates a third point, the model-sharing platform itself, prohibiting content depicting “real people — living or deceased,” and stating plainly that “consent to appear in one context (like film) doesn't equal blanket permission for AI generation.”

Authorship and limitation

The mechanism each document names is different: the statute creates a property right and reaches the distributor of a purpose-built tool; the union agreement creates a bargained consent-and-compensation obligation tied to employment; the platform policy is a content rule enforced by moderation, independent of any statute or contract. None of these documents describes consent as transferable between contexts by default — the statute's fair-use carve-outs are narrow (news, commentary, parody, self-representation without creating a false impression of an authentic recording), the agreement requires fresh consent for new uses, and Civitai explicitly rejects the idea that consent for one context covers another.

What survives

This is an editorial checklist, not legal advice: before generating a real person's likeness, check which of these three regimes actually applies — a statute where the subject or the tool's distributor is located, a union agreement if the work is covered employment, and the hosting platform's own content rules regardless of either. Consent obtained for one purpose, one document, or one contract does not travel automatically into a generation context on these documents' own terms.

• Check whether the jurisdiction where the work will be used or the person resides has a likeness or voice statute reaching AI tools, not just outputs.
• If the work is covered union employment, confirm whether a digital replica clause requires separate consent for creation and for each new use.
• Read the hosting or generation platform's own real-people policy before assuming a public appearance implies consent to generate.

Consent for a generated likeness is not one checkbox but three, set by a statute, a contract, and a platform policy that do not defer to each other.