How to Use Seedance AI Lip Sync: Talking Videos Guide
Seedance 2.0 generates talking characters with synced lips, while ByteDance's OmniHuman turns a photo + audio into a speaking avatar. Here's which to use and how.
"Seedance lip sync" actually covers two different tools from ByteDance, and picking the right one saves a lot of frustration. Let's clear that up first, then walk through each.
The two kinds of lip sync
- Seedance 2.0 itself generates talking characters with synced lip movement as part of normal video generation. Its official partner listing describes dialogue as "clear with precise lip-sync." So if you're making a scene where a character happens to speak, Seedance handles the mouth movement inside the same generation.
- OmniHuman is ByteDance's dedicated talking-avatar model — a separate product, not Seedance. You give it one photo + an audio clip, and it outputs a person speaking that audio with matched lip movement, expression, and body motion. It's available through Dreamina's AI Avatar / Lip Sync tools.
Rule of thumb: generating a video that includes speech → Seedance 2.0. Making a specific photo talk from an audio file → OmniHuman. Both are ByteDance, but they're different models.
Making a talking clip with Seedance 2.0
Because lip sync is built into generation, you drive it through the prompt (and optionally an audio reference):
- Describe the speaker and the line. Name who's talking, their tone, and the setting — e.g. "A news anchor at a desk, calm and authoritative, saying a short welcome, subtle head movement."
- Add an audio reference if you want the mouth timed to specific speech — in Dreamina's Multiframes mode, upload the audio and reference it with
@audio1(see how to use Seedance in CapCut & Dreamina). - Keep motion believable. Over-the-top gestures fight with clean lip sync; steady framing reads better.
Step by step: a talking clip in Dreamina
- Open video generation in Dreamina (or CapCut's Dreamina-powered video tool) and pick Seedance 2.0.
- Describe the speaker and setting — who they are, the mood, the framing.
- Put the line in quotes. Dialogue written into the prompt is what triggers generated speech with matching lip movement.
- Optional: add an audio reference. In Multiframes mode, upload a voice clip and reference it with
@audio1to time the mouth to specific speech. - Generate, then check the mouth. If the lips drift, simplify the motion — steadier framing, less gesturing — and re-run.
Two prompt shapes that work:
- A news anchor at a desk, calm and authoritative, steady medium shot, saying "Markets opened higher this morning across Asia." Soft studio lighting.
- Close-up of a weathered fisherman on a dock at dawn, warm rim light, he says quietly "The sea gives, and the sea takes." Gentle handheld sway.
When to reach for OmniHuman instead
If your starting point is a still portrait and a voiceover — a spokesperson, a UGC-style talking head, a character reading a script — OmniHuman is purpose-built for that. It preserves the person's identity from the single photo while animating speech and expression, which is more reliable than trying to force a full scene generator to hold a face perfectly.
Seedance lip sync vs OmniHuman at a glance
| Seedance 2.0 lip sync | OmniHuman (Dreamina AI Avatar) | |
|---|---|---|
| Input | Text prompt (+ optional audio reference) | One photo + an audio clip |
| Output | A full generated scene where characters speak | That exact person speaking your audio |
| Best for | Scenes with dialogue, ads, short-film moments | Talking heads, spokespeople, UGC-style clips |
| Where | Dreamina, CapCut, ponpon, API | Dreamina's AI Avatar / Lip Sync tools |
Try it
The fastest way to learn what reads well is to generate and iterate. Grab a character or portrait prompt from the library, then run it on ponpon, which runs the Seedance model line. For the underlying prompt structure — subject, camera, lighting, style — see how to use Seedance.