Introduction
Give Lipsync 2 a visible-face video and an MP3 or WAV recording to replace the dialogue. Choose sync_mode to handle unequal input lengths, then retrieve the MP4 when the asynchronous task completes.
| Field | Value |
|---|---|
| Input | Video, Audio |
| Output | Video |
| Publisher / tier | Sync / standard Lipsync 2 |
| Input contract | Required |
| Video length | 1–120 s; above-limit video rejected at acceptance |
| New speech | MP3 or WAV audio |
| Timing selection | Five modes; |
| Billing basis | Submitted video duration rounded up to seconds |
| Asynchronous result | MP4 retrieved after task completion |
| Storage period | Generated media deleted after 14 days |
Key features & highlights
The tailor finishes a longer invitation
Our atelier example began with 5.04 seconds of footage and a 6.5-second line. With loop selected, the returned video lasted 6.5 seconds: the source repeated while the mouth delivered the full recording. Its picture measured 864×496 at 24 fps.
A smile and eyebrow lift carry through
The tailor's small smile and raised eyebrow remained in our result after his dialogue changed. Sync says Lipsync 2 learns speaking style from the source without training on that person. Treat this clip as a concrete observation of that behavior, then inspect your own speaker's delivery.
Fit the ending to the edit you need
Use cut_off to stop at the shorter input, silence to pad a shorter recording, or remap to change video timing to fit the audio. Loop and bounce extend short footage by repeating it or alternating playback direction. Each choice uses the same input-video billing rule.
Request parameters
| Parameter | Required | Description |
|---|---|---|
video_url | Yes | Required URL for the existing video; keep the face clearly visible and duration within 1–120 seconds. No default applies. at least 1 characters |
audio_url | Yes | Required MP3 or WAV URL containing the replacement words and voice, with no default. at least 1 characters |
sync_mode | No | Optional in every request; default |
Pricing
$0.06 / second
The submitted video's duration is rounded upward to whole seconds for billing, within a 1–120 second input range. Audio length and sync_mode leave the charge unchanged for that video; clips above 120 seconds are rejected at acceptance.
Use cases
An after-hours shop invitation
Follow the tailor example with an authorized shop clip and a longer recorded invitation. The loop mode can repeat the visual take until the audio ends; check the restart before sharing the result.
A translated presenter segment
Prepare a translated MP3 or WAV for an existing presenter clip. Sync says speaking style can carry across languages; the endpoint matches the supplied recording rather than writing or translating the line.
Dialogue for a generated character
Sync says Lipsync 2 works with animated and AI-generated people as well as live footage. Bring an existing character video and replacement speech, then choose a duration mode for that pair.
A corrected tutorial sentence
Trim the instructor's take to the sentence being changed and record its replacement. Video seconds determine billing, so preparing that short source matters even when cut_off makes the result shorter.
Prompt guide & pro tips
Select the video and voice URLs
Prepare a clear-face source within 1–120 seconds and the finished dialogue as MP3 or WAV. Remove unused video before submission because its duration is billable.
Set
sync_modefor the endingCreate a sync/lipsync-2/lip-sync task with
video_url,audio_url, and an optional timing mode. The defaultcut_offstops at the shorter input; choose a different supported mode when that ending would cut the intended line.Track completion and inspect playback
Use the returned task ID to follow processing by polling or a callback. Save the MP4 within 14 days and review its mouth movement and final frames against the recording.
Standard-tier dialogue replacement
Sync's Lipsync 2 adapts mouth movement in existing footage to an uploaded recording. Sync describes it as a zero-shot model that learns a speaker's delivery style from the source, including when the replacement line is in another language. No speaker-training step is needed. Use sync/lipsync-2/lip-sync for the video-and-audio task; the result is an MP4 containing the supplied voice.
This endpoint exposes three fields. A written script cannot supply the voice, and there is no prompt input for directing the scene. Record or generate your dialogue separately before submitting the media URLs.
Request fields
| Field | What to send | Default and billing effect |
|---|---|---|
video_url | Required URL of a clip with a clearly visible face, 1–120 seconds long. | No default; the source video's duration sets billable seconds. Videos over 120 seconds are rejected at acceptance. |
audio_url | Required URL of replacement dialogue in MP3 or WAV format. | No default; changing audio length does not change the charge for the same video. |
sync_mode | Optional duration rule: cut_off, loop, bounce, silence, or remap. | Defaults to cut_off; choosing a different mode does not change the charge. |
Decide what happens at the ending
| Mode | Duration handling |
|---|---|
cut_off | Ends the result when the shorter input ends. |
loop | Restarts the clip as needed while the longer speech recording plays. |
bounce | Extends footage for longer speech by playing forward, reversing, and repeating that sequence. |
silence | Adds silence after shorter audio so the remaining video can play. |
remap | Changes playback speed to fit the video's complete timeline into the recording's duration. |
Choose loop when a repeat fits the shot. Review the restart in the returned clip. With bounce, check whether reversed movement suits the scene. Use remap when you want the video timeline fitted to the audio; inspect the resulting pace before using it in an edit.
Billing follows the submitted video
Count the input video's seconds and round upward to the next whole second. The accepted video range is 1–120 seconds. Neither audio duration nor the selected sync mode changes the billable quantity for that source. Consult the Sync Lipsync 2 model page for the current per-second price.
A 5.4-second source counted as 6 seconds in our cut_off trial, although the result measured 4.56 seconds. When your recording uses a small portion of a longer take, trim the video before sending it. A mode that shortens the output does not reduce the duration of the source you submitted.
Prepare footage and dialogue
Select the exact passage of video you need and confirm its duration. Keep the face clearly visible. Sync says lipsync-2 needs natural speaking movement in its source; a motionless portion may not produce speaking movement after processing. This publisher caveat matters when evaluating animated or AI-generated characters as well as filmed speakers.
Listen to the complete MP3 or WAV and check its ending against the planned video. The voice comes from that file; this operation does not clone the person in the footage. Use faces, video, and recordings you are entitled to use, ensure each person in the footage is an adult, and avoid impersonation.
Submit, track, and save
Submit the two required URLs with your chosen sync_mode, or omit it to use cut_off. Processing is asynchronous: retain the task ID, then poll for completion or receive the completion callback. Task acceptance means the request has entered processing; retrieve the MP4 after it completes.
Watch the mouth while listening to the new words, and inspect any repeated, reversed, or retimed footage. Download completed media during the 14-day retention period. Generated files are deleted after that period.
Results from our supplied trials
The midnight-tailor-invitation example pairs 5.04 seconds of 864×496 atelier footage with a 6.5-second spoken invitation and sync_mode: loop. In our tests, the video repeated to cover the line and returned a 6.5-second MP4 at 864×496 and 24 fps. The tailor's small smile and eyebrow lift remained visible as his mouth formed the replacement words.
A separate 512×768 source ran 5.4 seconds at 25 fps. With 4.55 seconds of audio and cut_off, the MP4 measured 4.56 seconds and retained those picture dimensions and frame rate. These short trials document their particular outputs; they do not establish a resolution ceiling or a processing-time guarantee.
When to evaluate a related model
For a close shot where facial detail drives your decision, evaluate Sync Lipsync 2 Pro. Sync positions Pro as the higher-precision tier. Its billing counts audio seconds, and its available sync modes exclude silence; check that contract before reusing a request.
Other model contracts to review include Lipsync 3 from Sync, the VEED Lipsync 2.0 option, and Lip Sync v1. Compare their documented inputs and timing behavior with the edit you need rather than assuming their request fields match.
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: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "sync/lipsync-2/lip-sync",
"input": {
"audio_url": "https://example.com/reference.mp3",
"sync_mode": "cut_off",
"video_url": "https://example.com/reference.mp4"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.


