Hedra Prompts and Settings: Expression, Motion and Voice

Hedra (hedra.com) animates a portrait to speak an audio track. Once you have a good image and clean audio, the remaining quality comes from three things: the prompt, the expression and motion controls, and the voice and model you choose. This guide explains what each one does, how they interact and what to write down so you can reproduce a performance you liked.

What the prompt does in Hedra

In most image and video tools the prompt describes what you want to see. In Hedra the image already shows the character, so the prompt is a performance note. It tells the Character model how the person should come across while speaking. The useful vocabulary is the vocabulary a director uses with an actor:

• Mood: calm, warm, serious, excited, reassuring, playful, confident.
• Energy: gentle, measured, animated, lively, subdued.
• Movement: small natural head movements, nods along with key points, mostly still, leans in slightly.
• Gaze: steady eye contact, occasionally glances down as if reading, looks off to the side thoughtfully.
• Face: slight smile, raised brows on questions, relaxed jaw.

One line is enough. A prompt such as warm, confident presenter; slight smile; small natural head movements; steady eye contact gives the model clear direction without contradictions. Avoid describing the character’s appearance, clothing or background; the image already fixes those and repeating them wastes the prompt. Avoid camera language too, since the framing comes from your crop.

Expression and motion

Hedra exposes controls for how animated the face is and how much the head and upper body move. The exact labels and ranges may differ in the current interface, but the behaviour is consistent.

Expression scales facial performance: the size of mouth shapes, brow movement, smiles and blinks. Low values give a composed, newsreader delivery. High values give a lively, emphatic performance that suits enthusiastic content but can tip into exaggeration, especially on illustrated characters. Photoreal faces tolerate more expression than stylised ones.

Motion scales head turns, nods, shoulder movement and breathing. Low values keep the character planted, which is right for tight crops and for characters that must stay centred for captions or overlays. High values add life to wider framings but increase the chance of drift, wobble or the head sliding towards an edge.

The two interact. High expression with low motion reads as an intense, direct address. Low expression with high motion reads as restless or distracted. For most presenter content, expression slightly above middle and motion slightly below middle is a reliable combination; for narration where the face is small on screen, lower both.

Voice and model choices

If you type a script rather than uploading audio, Hedra generates the speech with a text-to-speech voice. The voice matters more than any slider: it sets pace, warmth and the emotional cues the face follows. Audition three or four voices on the same two sentences before committing, and listen for how they handle your actual script’s longest sentence. Punctuation is your pacing tool: full stops create breaths, commas create short pauses, and splitting one long sentence into two makes most voices sound more natural.

Hedra offers more than one Character model version. Newer versions generally improve lip sync and expressiveness, so default to the latest, but keep a note of which one you used. If you return to a project months later and the results look different, the model version is the first thing to check.

Settings at a glance

Credit costs depend on length, resolution and model; check the platform’s current pricing page.

A tuning routine that works

1. Lock the portrait and a 10-15 second test audio. Do not change them again until the end.
2. Generate once with the prompt alone and middle values on both controls. This is your baseline.
3. Change only expression: one step down, one step up. Watch the three clips with sound.
4. Keep the best expression value and repeat for motion.
5. Only now adjust the prompt wording if something specific is still wrong, for example too many nods.
6. Render the full-length audio with the final values.

Write down the portrait filename, audio filename, prompt, expression, motion, voice and model version for the version you keep. A presenter series only looks consistent if each episode uses the same values. If you are starting from the beginning, the Hedra talking character guide covers choosing the image and audio first.

Common problems and fixes

• Over-acting. Expression is too high or the prompt stacks emphatic words. Lower expression and cut the prompt to one mood word.
• Flat delivery. The voice, not the face, is usually the cause. Try a different voice or add punctuation before raising expression.
• Head slides towards the frame edge. Lower motion and re-crop the portrait with more room on that side.
• Mouth shapes look too big. Common on illustrations. Lower expression and, if needed, choose a portrait with a smaller mouth relative to the face.

Try it

Portrait: fixed (your master image).
Script: Three things changed how I plan my week. The first is simple: I write tomorrow down tonight. The second takes five minutes. The third, honestly, took me years to learn.
Prompt: warm, measured, confident; slight smile; small natural head movements; steady eye contact; brief nod on each numbered point.
Expression: medium-high. Motion: medium-low. Voice: a calm mid-range voice at natural pace.

Remove the phrase about nodding on each point and regenerate; if the nods disappear you have confirmed the prompt is steering movement, and you can decide whether you want that emphasis or a stiller delivery.