Entries where a voice exists before any shot does
Four entries let a voice outlast the call that made it: bound to an element, attached to a character, or created as an object with an id. Whatever is wrong about it is wrong once rather than forty times. As of 2026-09-22.
| Model | Where the voice lives | And what it was built from |
|---|---|---|
| HeyGen | On a stored clone | A recording to clone from |
| Kling AI | On the element | Nothing documented as an input |
| LTX Studio | On the character Element | Nothing documented as an input |
| Runway | On a stored voice | A 10-second to 5-minute sample, or a sentence |
Inclusion rule. Entries whose documentation describes a voice that persists between generations, whether on a character object or as a voice object with an identifier. Entries that re-supply a voice on each call are on a separate route page. Order. Alphabetical by model name.
1Wrong once is a different kind of risk from wrong repeatedly
A stored voice concentrates the decision: the source recording, the description or the first generation settles the character, and nothing afterwards can drift. A voice rebuilt per call spreads the same decision across every shot, and the drift is silent because nothing in a returned file records what produced it.
Over a season that difference is the one that reaches an audience. Nobody notices a single shot whose voice is slightly off; everybody notices that a character sounds different by episode nine.
2Two of these store a character, and two store a voice
Where the voice belongs to a character object there is nothing to pass and nothing to pass wrongly, which removes the commonest continuity error rather than guarding against it. Where it is a voice object with an id, the pairing to a person in the story lives in a production's own records.
The second arrangement is more flexible and more clerical. A voice can be reused across series, shared deliberately or retired on its own, and somebody has to keep the spreadsheet that says which id is which character.
3Only one of the four lets you hear it first
Persistence and selection are separate problems, and most of this group solves only the first. Two bind a voice to a character without saying where the voice came from, so the first generation of each character is a casting session with a single candidate.
One publishes an asynchronous lifecycle with a preview, which is the only documented audition step in the register. That combination, a stored voice that can be heard before a shot exists, is what a cast actually needs and almost nothing here provides.
4The entries on this route, one page each
Each of these links to that entry on the field this route groups by. The wording behind every cell, and the date it was read, sits on the page it links to.
- HeyGen — on a stored clone.
- Kling AI — on the element.
- LTX Studio — on the character element.
- Runway — on a stored voice.
- Per-character bindingVoices are bound to elements, so a character carries its voice between generationstied to the element system
- Per-character bindingVoices are attached to character Elementstied to the element system
- Per-character bindingA voice is created asynchronously, reaches a ready state with a preview, and is then referred to by its own ida stored voice object
- Voice sourceA voice clone is instant from one recording or professional from twenty minutes or more, and is then passed as a voice idtwo enrolment routes with a threshold
- Voice sourceA custom voice can be built from an audio sample between 10 seconds and 5 minutes long and at most 10 MBa published sample window
- Audio sourceNative audio on VIDEO 3.0 in five languagesgenerated with the picture
5Sources
Membership of this route is decided by the wording in the field it groups by, read from the documentation each vendor publishes. The column itself is on the field note, all five columns are on the speech table, and what counts as documented is on how read. Other routes: A voice per call, Languages named.