Hedra Talking Character Guide: Image and Audio to Video

Hedra (hedra.com) turns a single image into a video of that character speaking. You supply the face, you supply the words as audio or as text that Hedra voices for you, and its Character models animate the mouth, eyes, head and shoulders to match. It is the fastest route from a still portrait to a presenter, narrator or animated mascot, and it works on illustrations and stylised characters as well as photographs.

How Hedra talking characters work

Hedra’s Character models are built for speech. Rather than animating a whole scene from a prompt, they take a portrait and a speech track and generate the facial performance that fits: lip movement synchronised to the phonemes, blinks, brow movement, small head turns and breathing motion. Because the clip length comes from the audio, you do not pick a duration; you pick what is said.

There are three inputs. The image decides the character’s appearance and framing. The audio decides timing, rhythm and emotion in the voice; you can upload a recording or generate one from a script with one of Hedra’s text-to-speech voices. The prompt and controls decide performance: how expressive the face is, how much the head and body move, and any direction about mood. Get the first two right and the third is fine-tuning.

Step by step: your first talking character

1. Choose or make the portrait. Front-facing or slightly angled, eyes open, mouth closed or neutral, good even light, nothing covering the lower face. Leave headroom and shoulder room so motion has space. A portrait generated in an image model works just as well as a photo, as long as the face is clear.
2. Prepare the speech. If you are uploading audio, record in a quiet room, trim silence from the start and end and keep it short for a first test, around ten to twenty seconds. If you are using text-to-speech, write the script with punctuation that signals pauses, then pick a voice that suits the character.
3. Set the frame. Choose the aspect ratio for where the video will be shown (square or vertical for social, landscape for presentations). Check the crop preview so the face is not cut off.
4. Add performance direction. A short prompt such as calm, friendly, slight smile, small natural head movements. Then set the expression and motion controls to moderate values for the first run.
5. Generate and review. Watch with sound on at full size. Check lip sync on plosive sounds (p, b, m), check that the eyes blink naturally and that the head does not drift out of frame.
6. Iterate on one variable. If the face is too still, raise expression. If the head wobbles, lower motion. If the voice is wrong, change the voice and keep everything else.

Choosing the image

Most failed clips come from the portrait, not the settings. Avoid:

• Profile or extreme angles. The model has to invent the hidden half of the face.
• Open mouths and big grins. A neutral mouth gives the model a clean starting point for speech shapes.
• Hands near the face, microphones, scarves. Anything that overlaps the jaw will smear.
• Heavy texture over the face. Dense patterns, netting or strong grain confuse the mouth tracking.

For a recurring character, generate a few portraits in the same session and keep the one that produces the best lip sync as your master image.

Audio or text-to-speech?

Recorded audio gives you a real performance: pacing, emphasis and breath that the face will follow. Use it when a person’s delivery matters, for example a founder message or a narrated story. Text-to-speech is faster and consistent across many clips; use it for explainer series, multilingual versions or when you have no voice talent. Whichever you pick, the quality of the audio sets a ceiling on the clip. Clean, close-miked speech syncs better than room-echo or music under voice. If you need music, add it in your editor after rendering, not in the audio you upload.

Settings at a glance

Generation uses credits that depend on clip length and resolution; check the platform’s current pricing page. Control names may differ in the current interface. For a closer look at the performance controls, see Hedra prompts and settings.

Common problems and fixes

• Lips lag or lead the audio. Trim leading silence from the audio file and re-export it at a standard sample rate.
• The face looks frozen. Raise expression a step and add warm, engaged, natural to the prompt.
• The head drifts or bobs. Lower motion and make sure the portrait has headroom.
• Teeth or mouth look wrong. Use a portrait with a closed, neutral mouth and try a different Character model version.

For alternatives, see AI talking avatar tools compared.

Try it

Image: head-and-shoulders portrait of a woman in her forties, short grey hair, navy jumper, soft window light, plain warm grey background, neutral expression, mouth closed.
Script (text-to-speech): Hello, and welcome. Over the next two minutes I will show you three small changes that make your mornings calmer. Let us begin with the first one.
Prompt: warm, calm, friendly presenter; gentle smile; small natural head movements; steady eye contact.
Expression: medium. Motion: medium-low. Aspect ratio: 16:9.

Regenerate with expression one step higher and compare the two side by side with sound on; most characters land between the two, and that value becomes your default for the series.