What Runway documents about speech

Runway (docs.dev.runwayml.com) documents a custom voice built either from an audio sample of ten seconds to five minutes or from a written description of at least twenty characters, then referred to afterwards by its own id. As of 2026-09-22.

Inclusion rule. Fields are the same five for every model here; a field the vendor does not document is recorded as not documented rather than inferred from generated output. Order. Fixed field order, identical on every model page.

1A voice can be asked for in words, which nothing else here offers

Two routes sit side by side. One is the familiar sample: ten seconds to five minutes of audio, ten megabytes at most, handed over as a link, an upload or inline data. The other is a text description, and the constraint published on it is a minimum length rather than a maximum, which reads as the platform asking for enough adjectives to work with.

The described route is the one that solves casting a minor character nobody has recorded. It also steps around the consent question a sample carries, because there is no identifiable person behind a sentence about a voice.

2A voice that exists before any shot does

Creation is asynchronous. A voice moves through processing into a ready state, arrives with a preview to listen to, and is then addressed by id. That sequence is what separates a voice from a parameter: it can be auditioned, approved and reused, and the approval happens once instead of inside every generation.

This register records that as a binding with a caveat. An id is a durable handle, but nothing on the page ties one to a character object, so keeping a cast straight remains a production's own bookkeeping.

3What the page settles, and what it does not

Where the sound in a generated shot originates, which languages are spoken, and what drives a mouth are all absent, because this page defines voices rather than rendering them. A model option with multilingual in its name appears; a model name is not a language list, and nothing here promotes one into the other.

What it does settle is the part most vendors leave vague: the exact shape of what a production hands over, written in seconds and megabytes.

4This entry, one column at a time

Each of these stays inside a single field: what Runway puts there, what the wording settles, and what it leaves for a take to answer.

• Voice source — a 10-second to 5-minute sample, or a sentence.
• Per-character binding — on a stored voice.
• Languages — not documented by the vendor.
• Lip-sync — not documented by the vendor.

5Read against another entry

Each of these puts Runway beside one other entry on a column where the two land at opposite grades of answer.

• Runway and MiniMax — on voice source.
• Runway and Luma Ray — on voice source.

6Where this model sits against the rest

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

7Sources

Taken from the custom voices reference at docs.dev.runwayml.com on 2026-09-22. Every entry against the same five fields sits on the speech controls page; the rule behind each field is on how read. The entries either side of this one: SceneMixer, Sora 2.

• Voice source
  A custom voice can be built from an audio sample between 10 seconds and 5 minutes long and at most 10 MBa published sample windowRunway, custom voices reference / recorded 2026-09-22
• Voice source
  A voice can instead be asked for in words, with the description required to run at least 20 charactersa route that needs no recordingRunway, custom voices reference / recorded 2026-09-22
• Per-character binding
  A voice is created asynchronously, reaches a ready state with a preview, and is then referred to by its own ida stored voice objectRunway, custom voices reference / recorded 2026-09-22
• Lip-sync
  Not documented by the vendor (as of 2026-09-22)not addressed where voices are definedRunway, custom voices reference / recorded 2026-09-22