Why Did Changing One HeyGen Ad Scene’s Voice Also Change the Other Scenes?

Describe the intended speaker map

A product advertisement can use one presenter throughout, or an explicitly staged exchange between two roles. Those are different sound designs. Write who should speak each section before editing the voice, especially when the same avatar appears in more than one scene.

An unexpected change in another scene can indicate that the edit was made at a broader configuration level. Diagnose that scope before changing pronunciation, replacing the script or generating the whole advertisement again. A local-looking control is not proof that its effect is local.

Set the scope before selecting the voices

Use the documented per-scene option when the project calls for separate scene voices, and inspect the actual available controls in that project. The help also describes Apply to all scenes. Treat that as a deliberate propagation choice rather than a harmless confirmation button.

Keep the speaker map near the edit request. It should distinguish the words to be read, the intended voice and the scene that owns them. A script revision and a voice reassignment can happen independently; do not assume that dividing text into sections automatically performs both.

After changing scope, inspect all affected scenes. Do not promise that previous assignments or already generated files are automatically repaired. The record should describe what was selected and what was actually heard afterward.

Illustrative exercise: presenter and staged question

This illustrative exercise was not executed in HeyGen. A fictional desk-lamp ad begins with a presenter explaining the switch, followed by a staged question read by a different narrator. Neither role is presented as a real customer testimonial.

The first production request requires two deliberate sound identities. The editor records the assignment for each scene, checks the scope option and listens to both sections after making changes. If both sections still use the same voice, they investigate assignment scope before rewriting the question.

A second request removes the staged exchange and uses one narrator for the whole ad. That is a changed creative decision. Applying one voice consistently can now be appropriate, but the earlier two-role approval should not be silently relabeled as if it always called for one speaker.

Review continuity after the scope is correct

Voice assignment is only the first acceptance question. Listen for product-name pronunciation, sentence endings and unexpected changes in emphasis. Where picture actions or captions depend on a previous recording, check their timing against the actual new audio.

Keep the scope choice, scene map and accepted output together. A voice label alone is insufficient if a later edit uses different text or timing. This workflow describes documented configuration boundaries and an original planning exercise, not an account test, a guarantee of identical performances or a built-in approval system.

Illustrative scene-voice assignment worksheet

These are illustrative exercise rows. Save a copy and record your own observations; the example is not an executed acceptance test.

Download review worksheet (CSV)