Choosing and using a lipsync model

Fabric 1.0 for an image that talks, Lipsync 2.0 for new audio on existing footage — what each takes, and how to get good results.

VEED has two lipsync models, and the one you want depends on what you start from. A still image: Fabric 1.0 makes it talk. Existing footage: Lipsync 2.0 re-syncs it to new audio.

Which one

	Fabric 1.0	Lipsync 2.0
You send	An image and an audio track	A video and an audio track
You get	A talking video	The same video, re-synced to the new audio
Use it for	Avatars, mascots, personalised videos	Dubbing, translation, replacing a voice track
Price	$0.08–$0.15 per second of the generated video	$0.07 per second of the generated video

Both are asynchronous: submit, then poll until the job settles. The quickstart walks that loop.

Fabric 1.0: an image that talks

audio_urlurlrequired

URL of the audio track to lip-sync to.

Must be an http(s) URL.

image_urlurlrequired

URL of the source image to animate.

Must be an http(s) URL.

resolutionenumrequired

Output video resolution.

Accepts720p480p

terminal

# Submit a job
curl -X POST "https://api.veed.io/v1/fabric-1.0" \
  -H "Authorization: Bearer $VEED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "audio_url": "https://static-assets.veed.io/api-examples/fabric-input.mp3",
  "image_url": "https://static-assets.veed.io/api-examples/fabric-input.png",
  "resolution": "720p"
}'

# Poll until COMPLETED (response includes data.job_id)
curl "https://api.veed.io/v1/fabric-1.0/{job_id}" \
  -H "Authorization: Bearer $VEED_API_KEY"

resolution is one of 720p, 480p, and the price follows it. The model takes speech, not text: to make the image say a script, generate the audio with a text-to-speech tool and send its URL.

Lipsync 2.0: new audio for existing footage

audio_urlurlrequired

URL of the new audio track to lip-sync the video to.

Must be an http(s) URL.

video_urlurlrequired

URL of the source video to dub.

Must be an http(s) URL.

terminal

# Submit a job
curl -X POST "https://api.veed.io/v1/lipsync-2.0" \
  -H "Authorization: Bearer $VEED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "audio_url": "https://static-assets.veed.io/api-examples/lipsyncv2-input.mp3",
  "video_url": "https://static-assets.veed.io/api-examples/lipsyncv2-input.mp4"
}'

# Poll until COMPLETED (response includes data.job_id)
curl "https://api.veed.io/v1/lipsync-2.0/{job_id}" \
  -H "Authorization: Bearer $VEED_API_KEY"

To translate a video, send the translated speech with the original footage. One speaker per video: more than one is not supported.

Getting good results

These hold for both models.

Send clean speech. One voice, with no music or background noise. If the recording is noisy, run it through Clean Audio first.
Show the face clearly. Front-facing, well lit, and not covered by hands, a microphone or hair.
Use a sharp, high-resolution source.
For Lipsync 2.0, avoid fast movement. A speaker who turns away or moves quickly gives worse results.
Try a short clip first, before you send a long job.

If a job ends FAILED, its error.code says why, and each poll endpoint lists what every code means:

GET

/v1/fabric-1.0/{job_id}

GET

/v1/lipsync-2.0/{job_id}