The Sound Is Already Done: AI Video That Needs No Audio Pass

Most people who have made a short video know that the video is only the first half. Then comes the other half: export, open an editor, go looking for music that fits and is cleared to use, hunt down a couple of effects, line them up against the picture, discover the timings are slightly off, fix that, export again. It routinely takes longer than making the footage did.

That half is not part of this. A clip from Fellowi Video comes back with its sound already in it, made at the same time as the picture and for the same scene.

What actually arrives

One MP4, with a video track and an audio track already combined inside it. Not a video and a separate soundtrack to line up, not a stem set to mix down. One file that plays with sound in any player, uploads to any platform, and sends to anyone without instructions.

The switch is on by default in the studio, and it does not change the price. Cost is model times resolution times duration; the sound is not in that arithmetic anywhere. A four-second 720p turbo clip is 500 coins with the audio on and 500 coins with it off, so leaving it on is free and turning it off saves nothing.

Worth saying plainly, since the reflex from other tools is to assume otherwise: there is no separate audio credit, no separate audio generation step to wait through, and no second render.

The prompt is the mixing desk

Because the sound is generated for the same scene, it is steered by the same sentence. Our guide to video prompts lists sound as the fifth thing a good prompt answers, and this is why: a clause about what you should hear is worth as much as one about what moves.

Say nothing and the model picks something plausible for the scene, which is often fine and occasionally not what you had in mind. Say one thing and it usually listens:

• “rain against the window, muffled traffic below”
• “a quiet room, a clock ticking, nothing else”
• “wind through pines, distant water”
• “low ambient music, no percussion”
• “a busy cafe behind her, cups and conversation, no music”

Two habits worth borrowing from the picture side. Name one thing rather than five: the soundscape has the same attention budget as the image, and a list of six sounds gets you a muddy average of all of them. And name the absence when it matters, because “no music” and “nothing else” are the rare cases where saying what you do not want works, since silence is a real state of a soundtrack rather than an object that has to be drawn.

When to turn it off

There are three honest cases, and none of them is about quality.

You are scoring it yourself. If the clip is going into a sequence with its own music, generated ambience underneath will only fight the track you chose. Turn it off and keep the timeline clean.

You need a specific piece of music. A brand's audio identity, a track you have licensed, a song the piece is built around. Nothing generated will substitute for a particular recording, and it should not try.

Nobody is going to hear it. Vertical social video autoplays muted, and most viewers never unmute. Our post on vertical video for social makes the same point from the other direction: if the first second has to work silently, spend your prompt on what moves rather than on what it sounds like.

What it does not replace

Treat what comes back as the scene’s own sound rather than as a voice actor working from your script. If a clip depends on exact words landing at exact moments, or on a cut falling precisely on a beat, you are still doing that in an editor, and you should. The honest description of this feature is that it removes the routine half of audio work, not all of it.

What it removes is nonetheless most of it. For the enormous number of clips that just need to not be silent, the step that used to cost an hour now costs nothing and happens while you are not looking.

Both paths give you it: describe a scene from scratch with text to video, or animate a still you already have. If you are unsure which you want, the comparison settles it in a sentence. The full set of dials, including the four models and the durations up to thirty seconds, is in the post about the control surface. Either way, play the result with the volume up before you decide to change anything.