Talking head vs B-roll: what's the difference?

A talking head is the shot of a person speaking to the camera; B-roll is the supporting footage you cut to while that person's voice keeps playing. The talking head, also called A-roll, carries the words. B-roll shows what the words are about: the product, the place, the steps.

An AI talking-head video follows the same rule: the avatar's voice runs as one continuous track, and the picture cuts between the avatar and the B-roll over it, so a cutaway never cuts the words. The Sume facts come from the Generate avatar video, Timeline 1.0, Audio detach and Models overview docs, read on 2026-09-28; the definitions are general. Anything called current behavior is read from Sume's code.

What is the difference between a talking head and B-roll?

One is the person saying it, the other is the thing being said. In the edit, the voice stays constant and the picture switches between them.

Is a talking head the same as A-roll?

In a presenter or interview video, yes. A-roll is the primary footage, the shots the story is told through, and there that is the person talking. B-roll is everything you cut to. In a video with no speaker on screen, such as narration over footage, there is no talking head at all; a faceless video is built from voiceover and B-roll alone.

How do a talking head and B-roll work together?

The voice runs without a break, and the picture cuts away and comes back. When it comes back to the speaker, the lips have to match the words at that moment, so the speaker's clip resumes at the same point in time as the voice. If the presenter should stay visible during a cutaway, put both in one frame, picture in picture or split screen.

• Stay on the talking head for the opening, a direct question to the viewer, and the call to action.
• Cut to B-roll when the words describe something you can show: a feature, a step, a place, a number.
• Use B-roll to cover a join, for example where two separately rendered clips meet in a longer video.

How do I add B-roll to an AI talking-head video?

A Sume avatar video resolves one avatar and one shared scene, so anything else on screen, B-roll included, is cut in afterwards. The edit is one Timeline 1.0 render with the avatar's voice, detached from its clip with audio detach, as the audio spine, and every clip in it must already be one of your workspace's media.sume.com files, such as B-roll generated on Sume. Keep the voice on the spine: in current code a render plays only the spine and an optional soundtrack, and each clip's own audio is dropped. Add B-roll to a talking-head avatar video has the slot layout and the cut-back timing.