Lip sync and talking avatars
The output quality is decided almost entirely by the source image, not by the audio or the settings. Get the character frame right and the rest is straightforward.
Last reviewed
Short answer
How does Dreamina lip sync work?
Supply a character image and an audio track — either uploaded or generated from text — and Dreamina animates the mouth and face to match. It costs roughly 40 credits per 10s of audio. The single biggest quality factor is the source image: front-facing, mouth clearly visible, neutral expression. Angled or smiling faces sync poorly regardless of audio quality.
- Cost
- ~40 credits
- Billed
- per 10s of audio
- Audio
- Upload or TTS
- Best angle
- Front-facing
Preparation
What makes a character image sync well
This table is the whole page, really. Everything else is mechanics.
| Attribute | Works | Fails |
|---|---|---|
| Head angle | Front-facing, square to camera | Three-quarter or profile |
| Expression | Neutral, mouth closed or slightly open | Wide smile, teeth showing |
| Framing | Head and shoulders, face filling a good share of frame | Full body, small distant face |
| Occlusion | Mouth and jaw fully visible | Hand near face, microphone, heavy beard shadow |
| Lighting | Even light across the face | Strong side light, half the face in shadow |
| Resolution | Sharp facial detail | Soft or upscaled-from-small source |
Generate the character for the job
If you are creating the character in Dreamina anyway, prompt explicitly for it — “front-facing portrait, neutral expression, head and shoulders, even soft lighting”. Two credits spent getting the source frame right saves a 40-credit sync that does not work.
Walkthrough
Producing a talking clip
- 1
Prepare or generate the character
2–6 creditsFront-facing, unobstructed mouth. Verify against the table above before going further — this is the step that decides the outcome. - 2
Write and trim the script
FreeCost scales with duration, so every unnecessary sentence has a price. Read it aloud and time it before you generate. - 3
Choose uploaded audio or text-to-speech
FreeUploaded speech carries natural pacing and emphasis. TTS is faster and more consistent but flatter — fine for explainer content, weak for anything with personality. - 4
Generate the sync pass
~40 credits per 10s of audioThe character image stays fixed; only the face is animated to the audio. - 5
Review at full size before regenerating
FreeSync errors that look severe in a small preview often read fine at final size. Check properly before spending again.
Quality
Fixing a sync that looks wrong
Four failure modes cover nearly everything.
| What you see | Cause | Fix |
|---|---|---|
| Mouth barely moves | Source face is angled or too small in frame | Regenerate the character front-facing, head and shoulders |
| Movement out of time with audio | Background noise confusing the waveform | Use a clean recording, or switch to text-to-speech |
| Uncanny, rubbery face | Over-expressive source image | Start from a neutral expression |
| Sync drifts near the end | Clip pushing the duration limit | Split the script and join the segments in an editor |
Real people need consent
Lip syncing a photograph of someone who has not agreed to it is the deepfake mechanism. It breaches the content policy and creates legal exposure that has nothing to do with what the tool will technically allow.
FAQ
Lip sync questions
What does Dreamina lip sync cost?
Around 40 credits per 10s of audio. Because cost scales with audio length rather than complexity, a script that runs thirty seconds costs roughly three times a ten-second one — worth trimming the script before generating, not after.
Can I use my own voice recording?
Yes. Uploaded audio generally produces a more natural result than synthesised speech, because real speech has pacing and emphasis that text-to-speech flattens. Record on anything — clean phone audio beats a poor studio take.
Why does my character barely move its mouth?
Almost always the source image. A three-quarter or profile angle gives the model very little mouth geometry to work with. Front-facing, mouth visible, neutral expression — that is what syncs.
Does lip sync work in languages other than English?
Yes, though accuracy varies. The mouth shapes are driven by the audio waveform rather than by understanding the words, so it handles most languages — but sounds that do not exist in the training distribution sync less convincingly.
Can I lip sync a photograph of a real person?
Technically it will run. Doing it to someone who has not consented is a serious problem — it is the mechanism behind deepfakes, it breaches the content policy, and it carries legal exposure entirely separate from anything the tool permits.
How long can a lip sync clip be?
Limited per generation. For anything longer, split the script into segments, generate each separately from the same character image, and join them in an editor — cutting on a natural pause hides the joins.
Next