Seedance AI Lip Sync Feature: How to Use in VideoGen
How the Seedance lip sync feature works and how to use it in VideoGen: write dialogue the model can sync, fix drift, and know when an avatar fits better.
To use the Seedance AI lip sync feature, write the spoken line inside quotation marks in your prompt, describe how it is delivered, and Seedance 2.0 generates the picture, the speech, and the matching mouth movement in one pass. In VideoGen, you write that prompt in a storyboard scene or a single clip request on a quality tier that lists Seedance 2.0 Mini or Seedance 2.5.
This guide covers how to write that dialogue, the steps in VideoGen, example prompts, common fixes, and when an avatar clip fits better.
How Seedance lip sync works
Seedance 2.0 is a multimodal video model from ByteDance. It takes text, images, video, and audio as inputs and generates clips up to 15 seconds long with synchronized stereo audio. Lip-sync is part of that same generation.
Because the audio is native, the prompt is where you control the words, the delivery, and the sound around them.
The Seedance 2.0 page lists the model's specs and shows examples.
Write dialogue the model can sync
Follow these rules when you write a spoken line:
- Put the spoken words in quotation marks. Everything outside the quotes is direction, not speech.
- Describe the delivery. "With a quick laugh," "in a calm, low voice," or "half-whispering" tells the model how the face should move.
- Keep lines short. One or two sentences per clip is a safe target. If you need more, split the script across clips.
- Use one speaker per clip. If a scene needs two people, number the shots so each shot has one speaker.
- Name the surrounding sound. Room tone, traffic, or a kitchen hum gives the audio a bed and makes the speech sit in the scene.
- Write the line in the language you want to hear. If the result is unclear in that language, VideoGen's model collection also includes other lip-sync capable models: Kling 3.0 (5 languages), Happy Horse 1.1 (7 languages, from a still image), and FLUX 3 (more than 14 languages).
Step by step in VideoGen
- Start a new project and pick a workflow. Storyboard-to-video fits a planned set of scenes, and in a storyboard scene the spoken script may fill the whole clip. You can also request a single clip from the editor.
- Pick a quality tier that lists Seedance in its "Powered by" line. As of October 2026, Seedance 2.0 Mini and Seedance 2.5 appear in every video clip tier. VideoGen picks a model per request from that tier and can fall back to another provider for the same inputs.
- Set the clip length. Seedance 2.0 Mini accepts 4 to 15 seconds in VideoGen and Seedance 2.5 accepts 4 to 30 seconds. VideoGen never shortens a request to fit a model; a longer request moves to a model that supports that length.
- Write the prompt in five layers: subject and action, environment and lighting, camera and framing, style and mood, audio and dialogue. The dialogue layer is where the quoted line and the delivery note go.
- Add references if you need a consistent character or product. Give each a role, such as "use image 1 for the speaker's outfit and the shop interior."
- Generate, then review. Watch the mouth against the words, check that the line finishes before the clip ends, and confirm the audio matches the scene.
- Place the clip in the timeline, enable captions across the project, add background music, and export.
For the full prompt structure, read how to use Seedance 2.0. The Seedance prompts page has copy-paste dialogue prompts.
Example prompts
A single line to camera:
A florist stands behind a counter of fresh peonies in soft morning light. Medium shot, slight push-in. She wipes her hands on her apron, looks at the camera, and says, "These came in an hour ago," with a warm, unhurried voice. Quiet shop ambience, a bell rings once as the door opens.
Two speakers, split across numbered shots:
Two-shot sequence:
1. Close-up of a mechanic holding up a worn brake pad. He says, "This is why it squeaks," in a flat, matter-of-fact tone.
2. Reverse shot of the customer nodding and saying, "Fix it," with a short laugh.
Keep the garage ambience and the radio playing continuous across both shots.
Fix common lip-sync problems
- Drift: the mouth starts in time and falls behind. The line is too long for the clip. Cut the dialogue to one sentence or split it across two clips.
- Mumbling: the mouth moves but the words are unclear. Add a delivery note and shorten the sentence. Remove competing sounds that sit on top of the voice.
- Two speakers talking over each other: give each speaker a numbered shot, or keep one speaker per clip.
- Too much text for the clip length: write dialogue a person could say comfortably in the clip's length, with room at the start and end.
- The face changes between takes: describe the character the same way in every prompt (age range, hair, clothing) and state what must stay unchanged.
Change one thing per re-generation so you can tell what fixed the problem.
Seedance lip sync or an avatar clip?
Both produce a talking person. They fit different jobs.
- Choose Seedance lip sync for short, cinematic dialogue inside a scene: a line to camera, a character moment, a product beat with sound. Camera movement, environment, and speech come from one generation.
- Choose an avatar workflow for longer scripts and a specific person on screen. VideoGen's avatar workflow pairs an actor image (stock or your own) with speech audio from your script, and routes to HeyGen or Hedra per request based on the actor and the generation requirements. If the actor image is not suitable for HeyGen (the face is small, turned, or obstructed), VideoGen generates a front-facing headshot of that actor from its existing references and retries. See the HeyGen page and the AI avatar generator.
- Both kinds of clip can sit in the same timeline.
Frequently asked questions
Does Seedance 2.0 need a separate voice recording for lip-sync?
No. Seedance 2.0 generates the speech and the mouth movement together from the prompt. You write the line in quotation marks and describe the delivery.
Can Seedance 2.0 lip sync a line in Spanish or German, or only English?
Write the line in the language you want to hear and review the result in that language. If it is unclear, VideoGen's model collection also includes FLUX 3 (lip-synced dialogue in more than 14 languages), Happy Horse 1.1 (7 languages), and Kling 3.0 (5 languages).
What happens if my spoken line is too long for a 15-second Seedance clip in VideoGen?
Seedance 2.0 Mini accepts 4 to 15 seconds in VideoGen. If you request a longer clip, VideoGen does not shorten it; it moves the request to a model that supports the length, such as Seedance 2.5 at 4 to 30 seconds.
Should I use Seedance lip sync or an avatar clip for a 60-second talking-head script?
Use the avatar workflow. It pairs an actor image with speech audio from your full script and routes to HeyGen or Hedra, so one face can deliver the whole minute. Use Seedance lip sync for short, cinematic lines inside a scene.
Start a project and test a dialogue clip inside a timeline where you can add captions, music, and the rest of the video.
Anton Koenig
