The Voice Note Was Fine. The Presenter Wouldn’t Stop Moving.

Consider Maya, a bakery owner preparing a short video for her weekend bread-making class. She already has a clear portrait and a 40-second voice recording. The message sounds warm and natural.

The first generated presenter is much busier. She nods through the opening, raises both hands during the class details, and keeps moving through a quiet pause. Nothing is completely broken, but the performance distracts from the announcement.

With OmniHuman 2.0, this example can begin with the portrait and recording alone. A motion prompt is added only if the first result needs clearer direction.

The goal is not to make the presenter move more. It is to decide when movement helps.

Start With 12 Seconds

Do not begin with the entire recording. Choose a short section that contains a complete thought.

For this example, Maya selects:

“This Saturday, we’re running a small bread-making class. We’ll start with the dough, then bake everything together.”

It includes an introduction, one important detail, and a natural pause. That is enough to judge the basic performance.

Use the clearest available portrait and a clean audio file. Keep the first test simple—no camera movement and no long list of actions.

Create a Baseline

Generate the short section without a motion prompt.

This shows how the audio guides the character before extra direction is added. Watch the result once at normal speed and note only the obvious problems:

The opening feels natural. Both hands move during the class details. The character keeps nodding during the pause.

A baseline prevents unnecessary prompting. If the opening already works, there is no reason to rewrite it.

Give Each Part One Job

Divide the recording into a few speaking beats. Assign an action only where it supports the message.

Audio section

Presenter action

Greeting

Relaxed posture and a small smile

Main detail

One open-hand gesture

Short pause

Hands return to rest

Closing line

Small nod

A focused prompt might be:

A calm waist-up presentation with a stable camera. Begin with a relaxed smile. Use one small open-hand gesture when mentioning the class. Return both hands to rest during the pause. End with a slight nod.

The instructions follow the recording. Each action has a reason to be there.

Avoid filling the prompt with words such as “energetic,” “cinematic,” or “highly expressive” unless the message calls for them. A local class announcement probably does not need the performance style of a stage presentation.

Check the Pauses

Unwanted movement often appears when nobody is speaking.

Mark the quiet moments in the audio:

0.0–3.2s greeting 3.2–3.6s short pause 3.6–8.8s class details 8.8–9.4s longer pause 9.4–12.0s closing line

Watch those pauses again. Look for a hand left in the air, repeated nodding, or a fixed smile that lasts too long.

If the character looks restless, ask for a resting position. Do not hide one unnecessary movement under another.

Compare the Two Versions

You now have two drafts made from the same image and audio:

Version A uses no motion prompt.

Version B adds one short sequence of actions.

Compare them at normal speed.

Check whether:

the gesture appears near the important phrase;

the hands return to rest;

the expression matches the voice;

the camera stays steady;

the character remains visually consistent;

the gestures still look calm when the clip is muted.

Keep Version B only if the added direction improves the message. If Version A feels more natural, use the simpler result.

Once the short section works, continue with the next part of the recording. Review the longer draft again because new pauses and changes in tone can affect the performance.

If one movement still looks awkward, remove it before rewriting the whole prompt.

Use Cleared Images and Audio

Only upload material you are allowed to process. A photo being public does not automatically make it available for animation. A recorded voice should not be used to create a new statement without the speaker’s permission.

For private or client-owned material, check the platform’s current upload and processing terms before using it. Features, settings, and handling policies can change.

If viewers could mistake the result for a real recording or endorsement, make its synthetic nature clear. Take extra care with testimonials, public figures, health advice, financial claims, and client announcements.

To test the workflow, use one cleared portrait and a short section of your own audio. Create a baseline, add one planned gesture, and compare the two results.

The voice note did not need fixing. The presenter simply needed fewer instructions.