Yapper API
Models & Inputs

Video lip-sync

Lip-sync models that align an uploaded source video to uploaded audio.

POST /api/v1/processes with "type": "video-lipsync". Pick any model — the API does not plan-gate models. GET /api/v1/models returns the same capabilities at runtime.

Training behavior

  • max is a trained model. A Max request without trainingId trains on the source video before generating, so the first run takes longer. Poll the process; once its trainingId appears, save it and pass it to later Max requests for the same trained source to skip training and start faster.
  • lipdub-v2, pro, and sync-lipsync-v3 are single-shot models. They generate directly from the supplied video and audio, require no training, and should not receive trainingId. For a lipdub-v2 source video longer than the safe 59-second cap, Yapper creates or reuses a cached variant containing its first 59 seconds.

Input fields

FieldTypeRequiredDescription
sourceVideoAssetIdstringyesTeam video asset id from /assets, /assets/import, or the direct-upload flow.
audioAssetIdstringyesTeam audio asset id from /assets, /assets/import, /assets/uploads, or /audio/speech. Its measured duration determines the exact quote and charge.
trainingIdstringnoSync-Max only. Reuse training from an earlier Max process; omit on the first Max run to train from the source video. Do not send for lipdub-v2, pro, or sync-lipsync-v3.
disableAutoRetriesbooleannoDisable automatic same-model retries. Defaults to false, so automatic retries are enabled.
endTimeSecondsnumbernoOptional source-video trim end in seconds. Omit to use the full remaining clip.
startTimeSecondsnumbernoOptional source-video trim start in seconds. Omit to use the beginning.

Example

Add "dryRun": true to get the exact credit cost without starting, and always send an Idempotency-Key header on real starts.

POST /api/v1/processes
{
  "type": "video-lipsync",
  "model": "sync-lipsync-v3",
  "input": {
    "sourceVideoAssetId": "video_asset_id",
    "audioAssetId": "audio_asset_id"
  }
}

Models

Cost depends on the input — use dryRun: true for an exact quote.

ModelNameDescription
lipdub-v2LipDub v2Single-shot LipDub model with direct video and audio ingestion. No training is required; longer source videos automatically use a cached 59-second safety-capped variant.
maxSync-MaxTrain-once model for highest quality. The first run trains on the source video; reuse its trainingId for faster later runs.
proSync-ProSingle-shot model with no training required. Quicker, with some artifacts on harder videos.
sync-lipsync-v3Sync-v3Single-shot model with no training required, stronger alignment, and more natural motion.

On this page