Avatar IV motion prompts

Write short, single-gesture motion prompts and plan scenes so custom motion does not visibly loop.

What a motion prompt does

With Avatar IV and photo looks you can add a short custom motion prompt that tells the avatar how to move while it talks: a nod, a wave, a point, a smile. It is free text, but it works best with a simple pattern: body part, then action, then emotion or intensity. For example: "Right hand lifts in a small wave, warm and friendly."

Keep it to one gesture

The most common mistake is stuffing several moves into one prompt. Stick to one clear, visible gesture plus at most one facial expression, or a short two-step sequence such as "nods, then points at the camera". Keep it under two short clauses. A chain like wave, cross arms, point and shrug usually comes out muddled.

Leave timing and camera out

Do not write seconds or timestamps; Avatar IV handles pacing by itself. Motion prompts are also not the place for camera moves, scene or location changes, props or walking. Those are handled elsewhere in the editor, if at all.

Plan around the loop

Custom motion is generated for a limited window of up to about ten seconds per scene, after which it loops or holds. In a 30 to 60 second Short, one long scene with a custom gesture can start to look repetitive. Split the script into shorter scenes, let the default motion carry most of it, and save custom gestures for highlights like the hook or the call to action.

Audio drives the face

The avatar takes a lot of its energy from the voice. Flat, monotone audio gives a stiff performance whatever the prompt says, so fix the voice first: try a livelier voice, adjust the speed, or upload your own expressive recording. On photo looks, a simple prompt such as "talks excitedly" can add life, but it will not rescue dull audio.

Iterate cheaply

Small wording changes can noticeably change the result, so test variations on short scenes. Custom motion uses extra credits and every re-render costs more, so check the speech first, then each scene, and only then render the full video.