Where AI voices fit
- Volume: hundreds of barks, a thousand lines of tutorial narration, a long backlist of audiobooks.
- Iteration: scripts that change weekly, where each rewrite is a new take in seconds.
- Early stages: temp dialogue, animatics, table reads and prototypes, before anything is final.
- Pickups: a changed line in a voice the performer has agreed to lend.
- Localization: the same voice in other languages.
- Projects that otherwise wouldn't be voiced: small games, indie podcasts, school materials.
Where actors are hard to replace
- Interpretation. An actor brings choices to a line that weren't in the script, and a director can shape them in the moment.
- Emotional range in one performance: a scene that moves from calm to furious to broken.
- Sounds that aren't words: laughter, crying, effort, breath.
- Comedy timing, improvisation and chemistry between performers in a room.
- Lead roles that define a work, and the audience's connection to a known performer.
A quick way to decide
| If the line is... | Lean towards |
|---|---|
| One of hundreds, functional, likely to change | AI voice |
| A draft, a placeholder or a test | AI voice |
| Narration that must stay consistent for years | Either; a cloned narrator, with consent, is common |
| The emotional heart of the story | Actor |
| Comedy, improv or ensemble scenes | Actor |
| In a language with no performer available | AI voice, reviewed by a native speaker |
How teams combine both
- AI first, actors last: temp voices through development, final recording with actors once the script is locked. The session is shorter because the script is finished.
- Actors for leads, AI for the crowd: principal roles recorded, minor characters and barks generated.
- An actor's licensed clone: the performer records a session and agrees, on paid terms, that their voice may be cloned for pickups, updates or other languages.
Working fairly with performers
- Consent first: never clone a performer's voice from their past work without their explicit agreement.
- Pay for the voice, not just the session: agree a fee for cloned use, its scope and its duration.
- Credit them, and say the voice is AI-generated.
- Respect union and agency terms. Many performers' agreements now include rules on AI use and payment.
- Let them hear it: share the project with them so they can listen to how their voice is used.
A written agreement covers all of this. See the consent template.
The honest bottom line
AI voices are a tool for getting more scripts voiced, faster, and for voicing work that would otherwise stay silent. They don't make great performances obsolete, and the best results usually come from treating the people whose voices are cloned as collaborators, not raw material.