Talking avatars & lip-sync

Make a character speak on Sunra: how lip-sync drives a face from an audio track with Kling 3.0, where the voice comes from, a worked example, source tips, and pairing with dubbing.

A talking avatar is a character whose mouth moves in time with speech. It takes two ingredients: a face (an image or clip of a person) and a voice (an audio track). Lip-sync ties them together so the character looks like it's actually saying the words.

The model that does it

On Sunra, lip-sync runs on Kling 3.0, which is built for dialogue. Its dedicated lip-sync capability drives a character's mouth from an audio track, so a still portrait or a clip can deliver a line convincingly. For a full worked example, see the lip-sync video use case.

When your spokesperson appears across several shots, keep the same face from cut to cut with Kling 3.0 multi-shot storytelling and the multi-shot character consistency workflow.

Where the voice comes from

The audio that drives the lips can come from anywhere in Sunra:

Text to speech — type a script and generate a voice. Best when you're writing the line from scratch.
Dubbing — translate an existing line into another language, then lip-sync the face to match it.
An upload — your own recorded voice.

How it works

Choose Kling 3.0 in the video generator.
Provide the character — a clear portrait or a short clip.
Provide the voice — generated or uploaded audio.
Generate. The model matches the mouth (and natural micro-movements) to the speech.

A worked example

Say you want a spokesperson to introduce a product:

In text to speech, generate the line: *"Meet the new Aero — lighter, faster, yours."*
Upload a clean, front-facing portrait of your spokesperson (real or AI-generated).
Run both through Kling 3.0 lip-sync.

Out comes a short clip of that face delivering the line. Generate the script in short sentences and you can re-roll a single weak line instead of the whole take.

Source tips

Use a front-facing face with the mouth clearly visible — profiles and extreme angles sync poorly.
Keep the audio clean: one speaker, minimal background noise.
Match the energy of the delivery to the face; a calm portrait reading an excited line looks off.
Keep lines short. A few tight sentences sync more reliably than one long monologue.

Lip-sync vs dubbing

They're complementary:

Dubbing changes the language of the audio but leaves the picture alone.
Lip-sync changes the mouth in the picture to match whatever audio you give it.

Localizing a talking-head video? Dub the audio into the target language, then lip-sync the face to the dubbed track — the result looks natively recorded. For the audio side end to end, see Voiceover and audio basics.