What is an AI digital twin? Twins vs photo and stock avatars

An AI digital twin is a custom AI avatar of one real person that can be made to speak new scripts in their likeness without filming again, with a clone of their own voice when the vendor makes one. HeyGen uses the name for an avatar trained on video footage of the person; Synthesia uses it for its Personal Avatars, which it now recommends building from a single photo.

In engineering, "digital twin" also means a virtual model of a machine or system; this page is about the AI video meaning. Vendor facts come from HeyGen's and Synthesia's own docs, read on 2026-09-28, and Sume facts from Create new avatar and Sume's code.

What kinds of AI avatars are there?

Vendors use different names, but the avatar types differ mainly in what the avatar is made from:

Is a digital twin the same as a photo avatar?

No. A digital twin is trained on video of the person; a photo avatar has only one frame to work from. Synthesia says it plainly: its photo avatars use a speech-driven animation model rather than motion learned from recorded footage, so the avatar's motion will not exactly match how you move in real life.

Voice follows the same split. HeyGen's digital twin clones one voice from the training footage automatically. Synthesia's photo route makes voice cloning optional; if you skip it, the avatar uses a Synthesia voice.

Is consent required for a digital twin?

Yes, from the person being cloned, and a vendor checks it before the twin can be used. The recording rules, consent steps and turnaround for each vendor are in how to create an avatar from a video. Two limits sit outside that flow:

• Age: Synthesia's Studio Avatars page says you must be at least 18 to create one.
• Likeness: a photo avatar skips the consent step in HeyGen's API, but HeyGen's consent page says that is not permission to use someone's likeness.

Does Sume make digital twins?

No. Sume's avatars are generated likenesses, not twins of a real person. POST /v1/avatar-1.0/generate takes one of three inputs: a prompt, a profile of traits, or a reference image at a public HTTPS URL. There is no video input, and even from a reference image the face and the voice are generated rather than copied.

• For ready-made presenters, POST /v1/avatar-catalog/search searches Sume's stock catalog; see stock AI avatars.
• To pair your own face and voice anyway, clone yourself with AI covers the lip sync route.