Talking head video vs conversational avatar: three things you get back

A talking-head video is a finished file you can review and reuse, while a conversational avatar is a live session that reacts to a person. Searching for either term usually means one of three needs: a recording to publish, a face to talk to, or both. Sume Avatar 1.0 makes the first. The second is what Tavus describes for Griffin-Lite, which is limited to select trusted testers as of 2026-10-08.

1. A file versus a stream

A Sume job ends with a video_url on media.sume.com. You can open it, trim it, caption it and play it for a thousand viewers. A streaming model produces frames for one conversation, and there is no file unless you record the session yourself.

That changes storage and sharing. With a file you can send a link, and a retry is a new job with its own Idempotency-Key.

2. Review happens before or after

With a script you can read every word before spending money, and you can check first-frame stills with the preview endpoints. With a live model you review afterwards or you set rules in advance.

Tavus says Griffin-Lite can deceive a person into thinking it is not AI and says it is working on safe disclosure features. For a rendered clip, you control disclosure yourself: a caption cue, a line in the description, or a spoken sentence.

3. How you learn it finished

For a rendered clip the contract is a job. Sume sends signed job.completed, job.failed and job.canceled webhooks, terminal events only, and the docs say to keep polling in place as a fallback. There are no progress events in webhooks.

For a live session the question is not completion but continuity, which is a different engineering problem.

Picking one

If the viewer does not need to talk back, render a clip. If they do and you have access to a streaming model, use it for that part only and keep rendered clips for answers you already know.

A quick test for your own brief

Read your brief and ask whether the words can be written before the viewer arrives. If yes, you are describing a talking-head video, and Sume Avatar 1.0 can make it today at $0.184, $0.245 or $0.55 per second by tier. If the answer depends on what the viewer says, you are describing a conversational avatar, and the access question for any streaming vendor comes first.

Many briefs are a mix. Split them. The welcome, the product explanation and the sign-off are scripts. The question-and-answer part is a conversation. Treat them as two projects with two budgets.