Digital twin vs AI replica vs AI avatar: which word is right?

Call a Sume Avatar 1.0 handle an AI avatar. It is a reusable identity that you create once and then use to render scripted talking videos. It is not a digital twin, and Sume does not market it as one. A digital twin, in the sense Tavus borrows from manufacturing, stays synchronised with a live counterpart. An avatar handle never syncs with anything.

Tavus's October 2 article Digital twin vs. AI replica, read 2026-10-06, makes the same point about people. It quotes a definition of a digital twin as a representation 'with synchronization between the element and its digital representation', and argues that an AI copy of a person is trained once from video and never resyncs with the original.

The four systems Tavus separates

Why the word matters on a landing page

If your page says digital twin, a reader expects something that tracks the person. A Sume avatar does not. It holds a face and a voice, and each video is a separate job built from the script you send. Describing it as a reusable presenter avoids a claim you cannot back.

The practical test is a single question: will the thing change when the real world changes? A twin does, by design. A replica or an avatar changes only when someone retrains it or creates a new one. Use the word that matches the answer, and your page will be easier to trust.

What an avatar handle actually is

The create call takes a top-level avatar_handle and one input: a text prompt, structured traits (props: ethnicity, sex, age) or a public HTTPS image_url. Creation is job-backed and costs $0.95 once. After the job completes, the handle goes into POST /v1/avatar-1.0/talking-video together with a script, or with scene-by-scene video_inputs.

Each render is a 4-60 second job at standard, plus or max quality, and a final video uses one avatar. Nothing about the person updates between renders unless you create a new avatar. That is the plain description to put on the page.