A product video without sound feels unfinished, and a product video with the wrong sound feels cheap. Here is how audio actually works in what Artvizon generates, and the small number of things you control.
Sound arrives with the video
This is the part that surprises people coming from template editors, where audio is a separate track you drop in afterwards. Here the video models generate the audio along with the picture, so what comes back already has an ambience appropriate to the scene: the room tone of an interior, the surface sound of something being set down, the general acoustic weight of the place the shot is in.
The practical consequence is that you are not choosing a music bed from a library. You are describing a scene, and the sound belongs to it.
Chapters are stitched without an audible seam
A multi-chapter video, 24, 32 or 36 seconds, is built from 12-second chapters. The naive way to assemble those is to butt them together, and the result is a click or a sudden change of room at every join, which is one of the fastest ways to make a video feel amateur.
The chapters are stitched with audio continuity across the transitions instead, so a viewer hears one continuous piece rather than three clips in a row. It is the kind of detail nobody notices when it is right and everybody notices when it is wrong.
Voice-over is optional and worth thinking about
You can add a spoken voice-over, written in the language you are chatting in, so it is composed in that language rather than translated into it. The difference shows up mostly in rhythm: translated narration tends to run long and land oddly on short sentences.
Whether you want one depends almost entirely on the placement.
- TikTok is largely watched with sound on. Narration works here, and often carries the ad.
- The Meta feed is largely watched with sound off. A video that only makes sense when heard will fail for most of its audience. Build for silence and treat audio as a bonus.
- Etsy and marketplace listing videos frequently autoplay muted on a loop. Narration is close to wasted there.
- Landing pages should autoplay muted by default, because a page that makes noise unprompted loses more visitors than it keeps.
The general rule: make the video work silent, then let the audio improve it.
Setting the mood
One sentence in the chat is usually enough. "Calm and premium" produces something different from "energetic, fast, street", and both produce something different from silence, where the model chooses for you.
Be concrete rather than adjectival where you can. "Quiet interior, morning" gives the model more to work with than "nice atmosphere".
What to check before you publish
Play it once with sound and once muted. If the muted version makes no sense, add on-screen text rather than assuming the narration will be heard. If the version with sound is louder or more abrasive than the platform's surrounding content, it will feel like an intrusion, which is the one thing an ad in a feed cannot afford.
Related reading: Why multi-chapter video outperforms single-clip ads and How the AI agent writes your script.