AI video prompt guide

daVinci-MagiHuman

A practical prompt structure for avatar-style video: expressive, accurate facial performance and lip-sync, while the character and scene stay stable.

Designed forAvatar-style talking video

Main aimFacial performance and lip-sync

Key techniqueKeep the torso and background still

What is daVinci-MagiHuman?

daVinci-MagiHuman is designed for avatar-style video generation. Its prompts should prioritise high-fidelity facial performance and lip-sync while keeping the rest of the image stable. This helps avoid the warping or mangling that can happen when too much of a scene is asked to move.

Why use a strict prompt structure?

A clear structure tells the model exactly what should move and what should not. The character can speak and show controlled facial expression, while the torso, camera position and background remain fixed. It is particularly useful for portraits, presenters, characters delivering dialogue and other close-up speaking scenes.

1. Main body

Write 150–200 words in a clinical, precise style.
Start with the character’s appearance and immediate surroundings.
State the overall emotional disposition, such as aggrieved, elated or stoic.
Explicitly say that the torso and position remain stationary.

2. Facial dynamics

Describe facial movements chronologically as the speech progresses.
Include details such as brow movement, blinks, jaw movement and the shape of the lips.
Focus on precise articulation and natural lip-sync.
Avoid metaphors, poetic language and narrative framing.

3. Stable environment

Describe the background as rigid and static.
Avoid moving details such as rustling leaves, passing cars or active crowds.
Keep the framing and camera position fixed.
Use a simple scene when facial performance is the priority.

Dialogue and sound

Use: Dialogue: <character description, language>: "Dialogue content"
Keep the description short and clear, for example: middle-aged man with glasses, English.
Add a separate Background Sound: paragraph.
For silence, use exactly: Background Sound: "No prominent background sound"

Copy-and-edit template

Character/Image: [Describe your image or character]

Scene/Mood: [Describe the emotion and setting]

Dialogue: [Insert the text they are saying]

Audio/SFX: [Describe background noise or write "none"]

Then ask for a 150–200 word clinical performance description. It should begin with the character’s overall state and emotional disposition, state that the torso and environment are fixed, and describe facial movements in chronological order. Finish with the separate Dialogue and Background Sound lines.

Example: A woman in her late 20s with shoulder-length blonde hair, wearing a white linen blazer, sits in a brightly lit modern office. Her disposition is one of suppressed frustration. Her torso and shoulders remain completely stationary within the frame, and the desk and shelving behind her remain rigid. As she begins to speak, her upper lip curls slightly and her nostrils flare. Her brow furrows towards the centre. Her mouth opens with precise movements to articulate each syllable, and her jaw drops slightly on emphasised words. Her eyes remain forward and she blinks once slowly mid-sentence. As she finishes, her lips press into a thin line.

Dialogue: <blonde woman in white blazer, English>: "We have gone over this three times already."

Background Sound: The faint, rhythmic hum of an air-conditioning unit.