Lip-sync a photo to audio with the Percify API

Turn a portrait and an audio file into a talking avatar video with POST /v1/run and infinitetalk-fast or infinitetalk. Inputs, pricing and examples.

To make a photo talk through the Percify API, call POST /v1/run with modelId set to infinitetalk-fast or infinitetalk, an image URL of a face and an audio URL of the speech. Percify returns a generation id, and the finished lip-synced video appears in output.urls when the generation succeeds. It is billed per second of audio.

POST https://api.percify.io/v3/playground/v1/run

Which model should I use?

Model id	Resolution	Best for
infinitetalk-fast	480p	Drafts, tests and high volume
infinitetalk	480p or 720p (default 720p)	Final renders. 720p costs twice as much as 480p.

Body

modelId · string · required

infinitetalk-fast or infinitetalk.

input.image · string · required

Public https URL of the face to animate, such as a .jpg or .png portrait with a clearly visible face. The video keeps this image’s aspect ratio.

input.audio · string · required

Public https URL of the speech, as .mp3 or .wav. Percify reads the file’s length to price the run, so it must be reachable.

input.resolution · string · default: "720p"

infinitetalk only: 480p or 720p. infinitetalk-fast ignores it.

input.seed · integer · default: "-1"

Random seed. -1 picks a random one.

webhook · string

Optional https URL to receive the result. See Webhooks.

Send an Idempotency-Key header too, so a retry never charges twice.

Example

cURL

curl -s -X POST https://api.percify.io/v3/playground/v1/run \
  -H "Authorization: Bearer $PERCIFY_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: intro-video-v1" \
  -d '{
    "modelId": "infinitetalk",
    "input": {
      "image": "https://example.com/presenter.jpg",
      "audio": "https://example.com/intro.mp3",
      "resolution": "720p"
    }
  }'

Then wait for it with GET /v1/generations/{id}?wait=45. When it succeeds:

Succeeded (trimmed)

{
  "success": true,
  "data": {
    "id": "b2d7e0f4-5c1a-4e93-8f26-0a9c3d7e1b45",
    "modelId": "infinitetalk",
    "status": "succeeded",
    "output": {
      "type": "video",
      "urls": ["https://cdn.percify.io/media-assets/playground/b2d7e0f4-5c1a-4e93-8f26-0a9c3d7e1b45-0"]
    },
    "error": null,
    "mediaDurationSec": 10.4
  }
}

mediaDurationSec is the audio length Percify measured and billed, and creditsSpent holds the credits charged.

How much does lip-sync cost?

Both models are billed per second of the audio you send, and infinitetalk at 720p costs twice its 480p price. For the exact credits, send the same body to POST /v1/estimate first; it measures your audio file and never charges. Credits are refunded if the run fails.

Errors specific to lip-sync

Message starts with	Fix
Missing required field: image or audio	Send both URLs.
audio URL is not reachable	Make the file public and use https.
Unsupported media format	Convert the audio to .mp3 or .wav.
Image rejected by content moderation	Use a different portrait. Nothing was charged.
resolution must be one of: 480p, 720p	Use one of those values, on infinitetalk only.

Long clips take longer to render. A generation fails and refunds if it is still running after 15 minutes plus 5 seconds per second of audio, or after 30 minutes at most. See Async jobs.

Related

Talking avatar pipeline

Generate the face and the voice too.

Voice cloning and speech

Make the audio track with zonos2.

Get a generation

Wait for the video URL.

Lip-sync troubleshooting

When the mouth does not match the audio.