How to Make a HeyGen Avatar Move: Motion Prompts

A talking avatar that never moves its hands looks like a webcam left on. HeyGen's Custom Motion lets you direct how the avatar acts: facial expression, posture, hand gestures and where it looks. It works with the Avatar IV and Avatar V engines on your custom avatars, and the difference between a stiff presenter and a convincing one is mostly in how you write those prompts. This guide covers both engines, the vocabulary HeyGen recommends, the hard limits, and how to keep motion natural over a long script.

Choosing the engine

In AI Studio, click your avatar, select the person icon in the upper-right corner and choose the avatar engine.

• Avatar IV takes free-text motion prompts you write yourself, applied per scene, so movement can change with the script. HeyGen describes it as interpreting vocal tone, rhythm and emotion to produce head tilts, pauses and micro-expressions, and as the best engine for cartoon, animated or non-human faces.
• Avatar V, HeyGen's newest and highest-fidelity engine, adds preset tabs for Expression, Gesture and Gaze under Advanced Settings. Use them on their own or combine them with your own instructions.

Avatar IV's credit use depends on the look: HeyGen lists 16 credits per minute for a Photo Look and 31 for a Video Look, charged only for the seconds of avatar generation. Check your plan's current rates before long renders.

What motion prompts can and cannot do

HeyGen is equally clear about what is out of scope: camera motion (zoom, pan, dolly), changing location, props or actions such as drinking coffee or picking up a phone, standing up or walking around, and background or lighting changes. Those belong elsewhere in the editor, or to a different kind of avatar.

Writing a motion prompt

HeyGen's structure for Avatar IV is body part + action + emotion or intensity, for example: Right arm raises in a wave, enthusiastic and friendly.

• One body gesture per prompt, optionally with one facial expression: "Avatar smiles softly while raising a hand."
• Short two-beat sequences are fine: "Avatar nods confidently, then points forward." More than that ("waves, crosses arms, then points and shrugs") is too much.
• Keep it under two short clauses for stable, lifelike motion.
• No timings. The engine handles pacing; you do not need to specify seconds.
• Experiment with wording. HeyGen notes that small changes in wording can make big differences.

The 10-second limit and long scenes

Custom motion has a maximum duration of 10 seconds. Applied to a longer scene, the motion is generated for the first 10 seconds and then loops or holds. HeyGen's recommendation is to use default motion, with no custom prompt, for the main body of a long scene and to apply short custom prompts only to the gestures you want to highlight. In practice:

1. Split the script into scenes at natural beats.
2. Leave most scenes on default motion.
3. Add a custom gesture where it supports the words: a point at "this feature", a hand on heart at "thank you".
4. Use stillness prompts for dense or serious passages so the avatar does not fidget.

Intercutting default motion with short, targeted gestures looks more natural than one prompt stretched over a minute, and avoids the visible loop.

Matching gestures to the script

Gestures read as natural when they arrive with the words they support, the way a presenter's hand rises just before the important word. A few pairings that work in explainers and sales videos:

• Introductions: "Avatar smiles warmly and gives a small wave" on the greeting only.
• Lists: "Avatar counts on the fingers of one hand, focused" on a scene that lists three points.
• The key benefit: "Avatar leans in slightly and points forward, confident".
• Thanks or values: "Avatar places a hand on heart, sincere".
• Data or caveats: "Avatar keeps hands still, calm and serious".

Keep a short library of prompts that worked for your avatar, with the engine used. Because small wording changes can change the result, reusing proven wording is faster than writing new prompts for every video.

When you need real movement

If the avatar must walk, sit down or move through a space, motion prompts are the wrong tool. A walking or motion digital twin is trained from footage of the person doing it: HeyGen's filming guidance asks for at least three minutes, head centred, soft even light, a steady camera, no other faces in shot and a range of natural expressions. For camera-directed, multi-shot scenes, HeyGen also offers cinematic avatar and prompt-to-video features. See the avatar video guide for the wider workflow.

Fixing common problems

• Nothing happens. Name a visible action (point, nod, wave) rather than a feeling alone.
• Gestures look frantic. Cut to one gesture and add "composed" or "less expressive".
• The gesture repeats. The scene is longer than 10 seconds; split it or drop the custom prompt for the remainder.
• The prompt is ignored. Check the engine: motion prompts work with Avatar IV and V on custom avatars.

Try it

Avatar smiles warmly and opens both hands towards the camera, welcoming and relaxed.

Apply it to a short opening scene of under 10 seconds, leave the following scenes on default motion, and add "Avatar nods, then points forward" only at the line that introduces your main point. Compare it with a version that uses a single prompt for the whole video.