Captions Avatar Looks in Prompt to Video: the Sume handle workflow

To reuse several looks across Sume videos, create one avatar per look, name the handles so you can tell them apart, and send the right avatar_handle with each video request. Captions, per its release notes, instead saves looks under an avatar for use across Prompt to Video and other avatar projects.

Captions facts are from its release notes; Sume facts from Create new avatar and Generate avatar video, read 2026-10-01. For the plain comparison see captions avatar looks vs a new avatar handle.

How do I organise several looks in Sume?

Pick a handle per look, for example host_studio and host_street. The handle may include a leading @; Sume stores it normalized without @, so do not rely on the @ to distinguish two handles. Create each with POST /v1/avatar-1.0/generate and a distinct Idempotency-Key, poll the job, then keep a small table of handle to purpose in your own app.

Which input kind fits a look?

Prompt describes the avatar from text, Profile gives structured traits, Image uses a reference image. The image_url must be a fetchable public HTTPS image URL, or the request is rejected before generation.

How do I pick a look per video?

Set avatar_handle on POST /v1/avatar-1.0/talking-video. Current execution supports one resolved avatar per final video, so a video cannot cut between two handles; render separate jobs and join them downstream.