A Line of Polish Speech With Kling 3.0: Lip Sync Checklist

Lip sync is judged by ear and by eye at the same time. A mouth that moves well is worth nothing if the Polish sentence is hard to follow. Begin small: one invented character, one short line. Keep in mind what this does: it generates speech together with the video. It is not a tool for copying a real person's voice.

Model and what it costs

The model here is Kling 3.0, at from $0.0963 per second right now. Turning on audio changes the estimate, so look at the pricing page or the calculator for your exact settings.

• No subscription, and your balance does not expire.
• A failed generation is free.
• A finished clip is charged even if you decide not to use it.

Setting up the line and the shot

1. Write the exact sentence in Polish, inside quotation marks, in the prompt. For a first attempt, avoid names and numbers, which are harder to pronounce.
2. One speaker, invented, facing the lens. Nothing in front of the mouth: no hands, cups or microphones.
3. Ask for the delivery you want: Polish speech, slow and calm, no background music.
4. Turn on generated audio and check the new estimate before you send the task.

The request body

Send this to POST /v1/generate. If you add a reference image, it must be a URL the API can reach.

{
  "model": "kling-3.0",
  "input": {
    "prompt": "An invented cartoon baker looks into the camera and says in Polish: \"Świeże pieczywo czeka na ciebie.\" Slow, calm voice, clear mouth movement, no music, no captions",
    "duration": 5,
    "generate_audio": true
  }
}

Reviewing the result

Two passes work best:

• Ears first. Put on headphones and listen without looking. Can you understand every word?
• Eyes second. Mute the clip and watch. Does the mouth start moving when the speech starts, and stop when it ends? The first and last syllables are where drift shows up.

Before you publish:

• Captions are typed and checked by you, and match what is actually said.
• Any ad claim in the line means what you intended in Polish.

Questions

Is audio on enough to get good Polish?

No. The switch only makes the model generate sound. Pronunciation and timing still need your review on every clip.

Can I see the price before I pay for a run?

Yes. Studio shows it before you send the task, and the API returns it as estimated_cost. Check the model schema at the same time.

Where are all the Kling 3.0 inputs listed?

On the Kling 3.0 model page and in the generate endpoint docs.