Most of what we automate has nothing to do with generative AI. It's folders, versions, captions and chasers. But localisation and repurposing work regularly raises the question: can we just clone the voiceover artist for the other eleven languages?
Sometimes, yes. Only with the right consent, and never by default. Here's how we handle it.
Rule 1: no cloning without signed, scoped consent
A voice or likeness is only cloned when the artist has signed consent that names the project, the languages, the media, the territories and the time period. "Can be used for anything, forever" is not consent. It's a red flag, and we won't build on it.
Rule 2: artists are paid for synthetic use
If a cloned voice replaces a session the artist would otherwise have recorded, the artist gets paid for it. We help clients set up usage-based fees with talent and their agents. Fair pay is the only way this works long term.
Rule 3: licensed models only
For anything that ships to a client, we only use models whose providers offer commercial licences and clear terms on training data. Experimental tools stay in the lab.
Rule 4: label everything synthetic
Every AI-voiced or AI-generated asset is marked as such in the delivery sheet, with the model, the consent reference and the date. If a client or platform asks later, the answer takes one search.
Rule 5: a person approves before anything ships
A native speaker reviews every synthetic voice line, and a creative lead approves every generated visual. No synthetic asset reaches a client automatically.
Our defaultWhen in doubt, we book the human session. AI voice is best for drafts, animatics, internal cuts and markets the artist has explicitly agreed to. It isn't a way to skip paying talent.
Why studios should care
Beyond the ethics, there's a practical risk. Laws and platform policies on synthetic media are changing quickly in many markets, and talent unions and agencies are actively negotiating AI terms. A studio that can show signed consent, licensed tools and labelled outputs for every asset is in a far stronger position with clients, talent and regulators than one that can't.
Consent isn't a box to tick at the end of a project. It's part of the brief.
What this means in practice
- Our localisation pipelines include a consent-check step that blocks synthetic VO when no valid consent record exists.
- Consent records live alongside the assets in your DAM, with expiry alerts.
- Delivery sheets carry a "synthetic" column by default.
This article describes our own policy and isn't legal advice. Please check talent contracts and local rules with your legal team.