Three layers of sound
| Layer | What it is | How to write it |
|---|---|---|
| Speech | the character’s words | in quotation marks: The woman says: «Привет!» (“Hi!”) |
| Sound effects | specific sounds of events | Sound: a door creaks, glasses clink |
| Background | the atmosphere of the place | Ambient: quiet café chatter, rain |
| Music | the mood | soft piano music or no music |
Rule: one line, one main effect and one background. A sound mess of ten sources comes out unintelligible.
Speech
Medium shot, a young man in a bright kitchen looks into the camera and says: «Это заняло у меня пять минут». Static camera, soft daylight. Ambient: quiet kitchen sounds. No music.
«Это заняло у меня пять минут» means “It took me five minutes”: the line stays in Russian so the character speaks Russian.
- The phrase fits into the length of the clip: for 6 to 8 seconds that is one or two short sentences.
- In Russian the lips match worse than in English: keep the lines short.
- For a talking head lasting a minute, a voiceover or a series of joined clips works better: AI talking avatar.
Sound effects and background
Wide shot, a rainy city street at night, a person with an umbrella walks past. Sound: footsteps on wet pavement, a car passes by. Ambient: steady rain. No music.
Describe specific sounds: not «city sounds», but «footsteps on wet asphalt, a car passes by».
Music
Slow push in on a decorated Christmas tree in a cozy living room. Music: soft festive piano melody.
If you do not need music, write no music: otherwise the model often adds its own.
When to turn sound off
- Reels and music videos: a track will go on top.
- Ads with a narrator: the voice is recorded separately.
- Marketplace video covers: they play without sound: product video with AI.
- Drafts: you do not need sound to check the motion and the shot.
What sound costs
KADR (kadr-ai.ru) prices for 6 seconds in 1080p, October 2026:
| Model | Without sound | With sound |
|---|---|---|
| Veo 3.1 Lite | 32 tokens | 53 tokens |
| Seedance 1.5 Pro | 46 | 86 |
| Veo 3.1 Fast | 64 | 80 |
| Kling v3 | 71 | 106 |
| Veo 3.1 | 132 | 264 |
| Seedance 2.0 | sound included in the price | 362 |
| Seedance 2.5 | sound included in the price | 537 |
This is not a public offer. With Veo 3.1 and Seedance 1.5 Pro sound roughly doubles the price, so turn it off wherever you do not need it.
A separate voiceover
If you need a voice on top of a finished clip, KADR has an «Audio» tab: text of up to 2500 characters is voiced by the ElevenLabs models Flash v2.5 (fast and cheap), Multilingual v2 (the most natural speech) and v3 (more emotion). You can take a standard voice, one from the library, or create your own from a text description. A minute of speech (about 1000 characters) costs about 89 rubles on Multilingual v2 and about 44 on Flash. The «Attach to video» button passes the finished voiceover to Seedance («Standard» mode) as an audio reference, and the character’s lips move in time with it. For a talking person from a photo there is OmniHuman 1.5: a photo plus your audio.
Which model to choose for sound
- Speech and detailed sound: Veo 3.1, Veo 3.1 Fast.
- Speech with the facial expressions of a real person from a photo: Kling v3.
- A long scene with sound at no extra cost: Seedance 2.5.
Ready prompts with sound: Veo 3.1 prompts. About the voices of real people: AI voice cloning.
In short
Video models voice the clip themselves: the line in quotation marks, sound effects and background as separate phrases, «no music» if you do not need music. With most models sound raises the price, with Veo 3.1 it doubles it, so for Reels, music videos, ads with a narrator and video covers it is cheaper to turn it off. With Seedance 2.5 sound is already included in the price.
