Introduction
Provide video_url for a 1–300 second face video and audio_url for the replacement speech. The resulting MP4 retains source picture dimensions and frame rate, with the speech re-encoded as mono AAC. A shorter recording trims the clip; a longer one extends its ending with frames played back.
| Field | Value |
|---|---|
| Input | Video, Audio |
| Output | Video |
| Model publisher | sync. · lipsync-3, also called sync-3 |
| Accepted pair | Face video via |
| Source range | 1–300 s |
| Returned media | MP4; source resolution and frame rate; mono AAC audio |
| Timeline behavior | Audio-led length, with trimming or backward frame extension |
| Charging basis | Rounded-up input-video seconds |
| Observed processing | Approximately 76 s and 173 s in two short-clip runs |
| Excluded inputs and controls | Still images, built-in TTS, |
Key features & highlights
Keep the selected take in your workflow
The source clip supplies the face video and the replacement recording supplies the words. lipsync-3 returns a re-synced MP4, making a corrected presenter line or an authorized translated lesson a concrete input pair rather than a text prompt.
Plan the ending around the speech
A 6.48-second speech track extended our 5.00-second source to 6.48 seconds by playing picture frames back from the end. Use audio close to the intended shot length when that reversal would distract from the delivery, and review the tail before placing it in an edit.
Retain picture dimensions, check the mix
Both measured clips kept their source resolution and frame rate: 1280×720 at 24 fps and 1358×680 at 25 fps. The output soundtrack is mono AAC, so check your final audio mix separately from picture dimensions.
Evaluate difficult facial transitions
sync. says sync-3 uses scene context and handles obstructions automatically in its model documentation. Test a representative head turn or hand crossing in your own footage; our two measured runs establish duration and processing observations, not a scored comparison of those visual situations.
Request parameters
| Parameter | Required | Description |
|---|---|---|
video_url | Yes | Required face-video URL; the source must be 1–300 seconds, and its duration rounds up to whole seconds for billing. at least 1 characters |
audio_url | Yes | Required URL for the replacement speech; WAV and MP3 were used in our tests. at least 1 characters |
Pricing
$0.162 / second
Billed per second of the input video, rounded up to a whole second, for clips of 1 to 300 seconds. The length of the audio does not change the charge, clips outside the range are rejected before any charge, and failed generations are free.
Use cases
Localize a recorded lesson
Pair lesson.mp4, showing an authorized instructor, with a separately recorded translated lesson.wav. Receive an MP4 whose mouth movements follow that supplied speech; prepare the translation and recording before the lip-sync call.
Correct a product announcement
Submit the original presenter take with a WAV recording of the corrected product name and sentence. The returned video matches the new wording, so you can review the revised line before replacing it in the edit.
Prepare regional ad versions
Reuse a spokesperson clip with separate approved voice-over files for each market. Each submitted video/audio pair produces its own re-synced MP4, and each request’s billing follows the source-video duration.
Re-voice an interview excerpt
Combine an authorized interview excerpt with a replacement MP3 speech track. Inspect the resulting MP4’s final frames when the replacement runs longer: the picture extends backward from the source ending to cover the audio.
Prompt guide & pro tips
Select and trim the face video
Choose a take with a readable mouth and measure its duration before upload. Keep the submitted clip within 1–300 seconds and remove footage you do not need to bill.
Finish the replacement speech
Prepare the spoken line as an audio file; our tests used WAV and MP3. Match its length to the intended edit if you want to avoid trimming or backward extension.
Submit the video and audio URLs
POST /api/v1/jobs/createTask with model set to sync/lipsync-3/lip-sync and input containing
video_urlandaudio_url. Include your Bearer API key and use the request example below as the JSON body.Retrieve and inspect the MP4
Save the returned taskId and query /api/v1/jobs/recordInfo?taskId=YOUR_TASK_ID for completion. Download the finished asset and compare consonants, mouth transitions and the ending with your speech file.
Sync Lipsync 3 API for existing footage
Sync Lipsync 3 API runs sync.’s lipsync-3, also called sync-3, to re-sync a face video to a replacement speech track. Send video_url and audio_url to sync/lipsync-3/lip-sync; the deliverable is an MP4 retaining the source resolution and frame rate, with mono AAC audio.
sync. says the model uses wider scene context and detects obstructions automatically in its model documentation. The tests below measure media properties, duration and processing time.
Submit a lip-sync request
Create a task with the public model name and the two media URLs inside input. Save this example as request.json and replace the placeholder URLs:
{
"model": "sync/lipsync-3/lip-sync",
"input": {
"video_url": "https://example.com/presenter.mp4",
"audio_url": "https://example.com/replacement.wav"
}
}
Submit that file using your SpicyAPI key:
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
--data-binary @request.json
Save data.taskId and query GET /api/v1/jobs/recordInfo?taskId=YOUR_TASK_ID with the same authentication. Check task state for completion.
Response shape
A completed task exposes the video through data.output.assets. This excerpt omits other fields and uses placeholder identifiers and URLs; media measurements match the WAV test:
{
"code": 200,
"msg": "success",
"data": {
"taskId": "job_example",
"model": "sync/lipsync-3/lip-sync",
"state": "succeeded",
"output": {
"assets": [
{
"url": "https://example.com/dubbed-result.mp4",
"mime": "video/mp4",
"width": 1280,
"height": 720,
"durationSeconds": 4.58
}
]
}
}
}
Parameters and available controls
This endpoint takes a source-video URL and a speech-audio URL. Omit publisher-specific controls.
| Field | Required | What it controls | Preparation |
|---|---|---|---|
video_url | Yes | Source face video, 1–300 seconds; resolution and frame rate follow this clip | Trim unused footage and inspect the visible mouth. |
audio_url | Yes | New speech and the finished timeline length | Prepare dialogue separately; our runs used WAV and MP3. |
Image input, built-in TTS and sync_mode are not offered by this endpoint. sync. documents additional input workflows in its own API; its request examples do not define the fields accepted here.
Output length and pricing
Input-video seconds determine billing, rounded up; use the price shown on this page for the current rate. Audio length does not change the billing quantity, even when the returned MP4 is shorter or longer than the source. Out-of-range clips are rejected before charging, and failed generations are not billed.
The replacement recording determines how much picture is returned. Shorter speech trims the clip; longer speech extends it by playing source frames back from the ending. Match speech to the shot length when backward motion would distract; there is no exposed mode switch.
What our short-clip tests showed
The November 13, 2026 fact sheet reports two short-clip tests with these output lengths and processing times. They do not establish long-video performance or a visual-quality ranking.
| Source video | Speech input | Returned MP4 | Processing |
|---|---|---|---|
| 5.04 s, 1280×720, 24 fps | 4.55 s WAV | 4.58 s, 1280×720, 24 fps | About 76 s |
| 5.00 s, 1358×680, 25 fps | 6.48 s MP3 | 6.48 s, 1358×680, 25 fps | About 173 s |
Both runs kept the source picture size and frame rate, and the output audio was mono AAC. The WAV case shows that encoded-video duration can differ slightly from audio duration.
Prepare a useful quality trial
Start a quality trial with a representative speaking shot and isolated dialogue. sync. recommends visible, well-lit faces and speech without music or overlapping speakers in its troubleshooting guide.
- Select a short take with permission to use both the footage and voice.
- Finish the replacement line and compare its duration with the picture.
- Review lip closures, head turns and any hand crossing around the mouth.
- Check the tail for reversed motion and listen to the mono AAC mix.
A lesson video paired with a translated recording is a localization input; a presenter take paired with a corrected product sentence is a dialogue-repair input. Prepare translations and speech before calling this endpoint.
Limits and troubleshooting
Check source duration and the two URL fields before investigating a failed request. The video range is 1–300 seconds; remove unsupported fields such as sync_mode rather than copying a publisher request unchanged. For a media-fetch problem, verify that each URL serves the intended file and remains accessible during processing.
An unexpectedly short result usually calls for checking the supplied speech length. Backward motion at the ending is the documented extension behavior when the speech outlasts the source. Keep the original files beside the MP4 when diagnosing timing.
Choose lipsync-3 or a sibling
Choose a lip-sync model by reviewing the same authorized input pair on each candidate.
| Option | When to evaluate it | Evidence here |
|---|---|---|
| lipsync-3 | Existing face video plus speech | Two measured short-clip runs. |
| lip-sync/v1 | Lower-cost lip-sync on SpicyAPI | No side-by-side quality or latency test. |
A workflow requiring image input, integrated TTS or selectable sync modes needs a separately documented API with those features. Use footage and speech you are entitled to use, and never use a re-voiced result to impersonate a real person.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "sync/lipsync-3/lip-sync",
"input": {
"audio_url": "https://example.com/reference.mp3",
"video_url": "https://example.com/reference.mp4"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.


