Runway Characters voice clone: 10 s to 5 min, 10 MB; the Sume route

Runway's custom voices page says a clone sample must be 10 seconds to 5 minutes long and at most 10 MB, or you can design a voice from a text description of at least 20 characters. Sume's avatar video request documents voice as a text script or a silence beat, with no sample upload; if you need your own recorded voice, the Fabric route takes an audio_url.

What Runway documents

From the Runway custom voices page, read 2026-10-02:

• Design models named on the page are eleven_ttv_v3 and eleven_multilingual_ttv_v2.
• Voices are assigned to an avatar with type: custom and the voice id.

What the Sume avatar request documents

In the avatar video guide, each video_inputs scene has a voice object: type: "text" with a script and duration, or type: "silence". The documented route has no field for uploading a voice sample, and these pages do not describe voice cloning for avatar video, so this article does not claim it.

If you already have the voice recorded

The Models page lists VEED Fabric 1.0 at POST /v1/veed/fabric-1.0: a talking still plus your audio. The body needs audio_url, a measured duration_seconds and one visual source. That moves the voice question to you: record or generate the audio first, host it at a public HTTPS URL, and Sume animates the face to it.

Mind the file size before you send it. Fabric audio has its own limit, so check the live reference rather than reusing Runway's 10 MB figure.

• Voice must match a person? Record them, then use Fabric.
• Voice just needs to sound professional? Use the avatar script route.
• Do not clone a voice you have no rights to, on any platform.

Summary

Runway makes the voice a managed object attached to a live Character. Sume keeps it simpler: speech comes from the script, or you bring finished audio to a still. Choose by whether you want a reusable cloned voice object, which Runway documents, or a render per clip.