A synthetic performance needs a permission record that travels with the asset

A generated performance should remain connected to the person, proposed use, reviewed agreement and actual acceptance behind it.

A synthetic performance needs a permission record that travels with the asset

A game producer asks for a short line of dialogue to be changed after the recording session. A synthetic voice tool could produce the new wording quickly. But the existence of a usable recording does not answer the production questions. Which performer agreed to which use? Does the proposed line fit the agreed description? Who can authorise the generation, review the result and account for the work? A faster tool makes those questions more urgent, rather than making them disappear.

This analysis was prepared and source-checked on 17 October 2026. It considers a particular US video-game agreement, a dated US policy recommendation and a UK performers' union position. They are different kinds of authority, with different scope. They are not a worldwide legal rulebook. The game-production workflow below is a fictional proposal; an actual project still needs review of its applicable agreements and permissions.

Keep the institutional claims within their boundaries

SAG-AFTRA's account of its 2025 Interactive Media Agreement describes consent and disclosure provisions for AI digital replicas. Its FAQ also connects permitted use to the reasonably specific description supplied when consent was given and notes that separately negotiated agreements can differ. This is information about a named bargaining framework, not permission to reuse any performer's voice or a summary of every entertainment contract. [Agreement account and FAQ](https://www.sagaftra.org/contracts-industry-resources/interactive/2025-interactive-media-video-game-agreement)

On 31 July 2024, the US Copyright Office announced its digital-replicas report and recommended new federal legislation. That is a dated policy recommendation concerning unauthorised replicas; it is not itself an enacted law or evidence of the legal position in every jurisdiction in October 2026. The report's existence also does not establish that a specific recording is licensed for a proposed production. [Copyright Office announcement](https://copyright.gov/newsnet/2024/1048.html)

Equity's AI Vision Statement sets out the UK union's proposed artist-centred principles, including consent, project-limited use, transparency, consultation and remuneration. These are attributed union positions. They are useful questions for a production discussion, but should not be presented as proof that a particular contract grants those rights or that all countries apply the same rules. [Equity statement](https://www.equity.org.uk/advice-and-support/know-your-rights/ai-toolkit/equitys-ai-vision-statement)

The common production lesson is a need for specificity. A file's technical availability, a platform subscription and a performer's permission concern different things. A team should be able to identify the basis for its intended use before asking a model to generate the asset.

Give the proposed use an identity

In the fictional game, a permission record could identify the performer, source session, project, proposed character and use description, with links to the reviewed agreement. Access should be limited to people who need the information. Do not scatter private contract terms through every editing folder simply to make them convenient to search.

Connect the record to the actual source files and generated versions. A renamed audio file should not lose the information needed to understand its authorised use. Keep training or model-building permission separate from permission to generate a particular performance. Also distinguish an internal prototype from a final release, a new line from an edit to an existing recording, and the original project from a later promotional campaign.

These categories are a proposed operational checklist, not a substitute for interpreting the actual agreement. When the intended use is unclear, the record should show that it is unresolved. A vague “approved” label should not let a new use inherit authority from a different one. The person reviewing the request needs the actual proposed consequence and the relevant source material.

Review the performance, not just the sound

After a properly authorised generation, the fictional team could review pronunciation, timing, emotion and the meaning of the new line. A technically natural voice can still change a character's intention or produce a performance the actor would not recognise as appropriate. Preserve the original take and mark the generated version so that editors do not confuse the two.

Record who reviewed the result and what was accepted. A producer approving a draft for discussion is not necessarily approving its public release. When a scene changes, examine whether the change affects the recorded permission and creative review. Reusing the same audio in a different context can change what the audience understands, even if the waveform is identical.

Make a correction route practical. If a performer or production reviewer identifies a problem, the team should be able to locate the affected versions and distributions. Removing a clip from one edit should not be reported as proof that every copy, training input or generated derivative has disappeared. The record needs to distinguish actions actually completed from actions still requested or outside the team's control.

Account for the human work around generation

Synthetic tools can change when and where performance work occurs. Preparation, direction, permission review, listening, correction and record maintenance still take effort. Include these tasks when comparing a proposed workflow with a conventional pickup session. A shorter generation step alone does not establish lower total cost or a fairer outcome for workers.

The fictional pilot could track the complete time from a requested line change to an accepted asset, including rejected generations and unresolved permissions. It could also record the creative decisions that required the performer or director. This would help a producer understand whether the tool solves a real scheduling problem or simply moves work into less visible review stages.

Discuss adaptation with the people whose work is changing. Performers may need clear ways to review proposed uses and raise concerns; editors and asset managers may need training in version and permission tracking. These are proposed arrangements, not a forecast that new roles will replace every lost session or that synthetic production guarantees employment growth.

The possible benefit is a controlled way to make certain authorised changes without losing the connection to the person and agreement behind the performance. The costs include negotiation, review, compensation and reliable asset management. Begin with one clearly described use and a record that travels with it. The test of a responsible workflow is whether everyone can explain what was permitted and what actually happened, not merely whether the tool can imitate the voice.