Sync Speech to Talking-Head Video — Dubbing & Translation
Align a talking-head video with a speech track to generate lip-synced output. LatentSync 1.6 was added in CNAPS Studio v1.2.2. Supports dubbing, translated narration, and AI voice workflows. Runs on CNAPS infrastructure; no external API key required.
What it does
LatentSync 1.6 takes a talking-head video and speech audio, then generates a lip-synced video. The speech track drives the mouth movement. It fits after narration generation or recording and before final sound mixing. The node was added in CNAPS Studio v1.2.2. It does not translate a script or generate the voice; prepare those inputs first.
Problem it solves
- Dubbing — Replace spoken delivery while aligning the visible mouth to the new track.
- Translated narration — Pair a translated voice track with suitable talking-head footage.
- AI voice workflows — Use Qwen3-TTS speech as the audio input.
- Reusable video production — Apply the same lip-sync workflow to a series of short clips.
Input/Output
- Video input: A talking-head clip with a clearly visible face.
- Audio input: The speech that should drive the lip movement.
- Output: A lip-synced video. Preview and listen to the result before adding background music or effects.
Configuring the Node
Source video — Prefer a clear, stable view of the face and mouth.
Speech track — Use clean speech; prepare the intended timing before lip-sync.
Advanced controls — Keep the node defaults for the first run, then adjust only controls exposed by the Studio node.
How to Use
- Upload the source with Video Loader.
- Prepare speech with Sound Loader or generate it with Qwen3-TTS.
- Connect the video and speech to their matching LatentSync input ports.
- Run a short clip and inspect the mouth throughout the result.
- Complete music, effects, and loudness treatment after approving the lip-sync.
Example Workflow
Flow: Video Loader + Sound Loader → LatentSync 1.6 → Video Viewer.
For generated narration: Text Input → Qwen3-TTS → LatentSync audio input; connect Video Loader to the video input.
Accuracy & Speed
Execution | CNAPS-hosted; no external provider API key required. |
Input quality | A visible, unobstructed mouth makes results easier to judge. |
Quality check | Check fast speech, head turns, teeth, and frame-to-frame consistency. |
Timing | Runtime depends on clip length and processing load; test one short clip first. |
How it compares
Use LatentSync when you already have a video and want its mouth movement to follow speech. Use LTX-2.5 when generating a video from text, imagery, or speech audio. These serve different stages of production.
Technical Details (for engineers)
The ByteDance repository describes audio-conditioned latent diffusion. Version 1.6 was trained on 512 × 512 video to reduce blurriness. Upstream inference steps and guidance settings are implementation controls, not a guarantee of the exact options or output size exposed by CNAPS.
Model Source
- LatentSync — official repository: LatentSync — official repository
- CNAPS integration: v1.2.2 release note
Compliance & Provenance
Provider | ByteDance |
License / service terms | The upstream repository is Apache 2.0; consult its checkpoint and dependency terms. |
Output provenance | Lip-sync alters visible speech. See CNAPS AI Policy for attribution and output labeling. |
For platform classification, acceptable use, and output marking, see the CNAPS AI Policy.
Run it now
Open CNAPS Studio, find LatentSync 1.6 in the Component panel, and start with the example workflow above.