A voice written in words instead of recorded
A voice description is text that stands in for a recording: age, tone, pace, accent, personality. Three vendors document one, and they use it differently: to design a voice that is then stored, to steer the delivery of a single generation, or as a line attached to a character that is performed every time it speaks. As of 2026-10-13.
| Vendor | Where the description goes | What it is used for |
|---|---|---|
| Hedra | Descriptors of the character's demeanour in a text-to-video prompt | Steering the delivery of one generation |
| Runway | A description of at least twenty characters naming tone, accent, pacing and personality | Designing a stored voice with its own id |
| SceneMixer | A line on each character covering age, weight, pace and where the voice sits | Performed on every line, unless a preset is matched |
Inclusion rule. Vendors whose documentation describes specifying a voice in text instead of supplying audio. A voice picked from a list by name does not count as a description. Order. Alphabetical by vendor name.
1Free, instant and slightly variable
A description costs nothing to write and needs no recording session, which is why it is often the default. The cost is precision. Words leave room, and a model filling that room can land differently from one take to the next, which one vendor states directly about its own default.
2Once, or every time
Used to design a stored voice, the description is read once and the result is kept; consistency then comes from the stored object. Used on every line, the description is read every time; consistency comes from the text never changing. Used in a single prompt, it is only as consistent as the person retyping it.
3What a description is good at
Accent, age and pace survive being written down better than timbre does. A request for a slight regional accent or an older, slower voice is the kind of thing a description carries well. A request to sound like a particular person is not, and is the territory of recordings and the permissions that come with them.
4When to stop describing
Once a character has a voice the production likes, the case for a recording or a matched preset is strong: it removes the variation the description allows. Where a vendor lets an attached voice take over from the description, anything written in the description, including an accent, stops being performed.
5Sources read for this entry
This page defines the term and logs what each vendor documents about it, each figure read on 2026-10-13. Related: Ambience, Native audio.