A voice over your film is not a person in it

There is a question that sounds like it is about sound and is actually about casting. Somebody wants a sentence said over their product film — the name of the thing, what it is for, one line — and the first control they find is the Sound picker's Voiceover option. It works. A line goes over the film. But if what they pictured was a person on screen saying it, they have chosen the one option where nobody appears, and the film comes back with a voice and no speaker.

That is not a mistake anyone should feel bad about, because the vocabulary genuinely collides. A spoken line is a spoken line. What differs is who, if anyone, is visible — and, less obviously, whether that person exists tomorrow.

Three doors, and only one of them remembers anyone

The three are separate surfaces, reached from different places, and they are not variations of a single feature.

• Voiceover, in the Sound picker. One spoken line laid over the film's own sound, which is turned down underneath it — or over silence, if you switch that bed off. Nobody is on camera. Leave the script blank and a line is written for you from the film's beat sheet. The script runs to roughly what a forty-second film holds at three words a second, so it is the only one of the three that scales with a longer film.
• A creator format, among the ad formats. Several of the creator styles put a generated person on camera saying a short hook. That person is produced for that film and is not saved anywhere. Make the film twice and you get two strangers.
• Say a line, in Model Studio. A model from your roster — one you created and kept — delivered to camera, speaking, as one ten-second take. The speech is the film's own audio rather than a track laid on top. This is the only door where the person is a thing you own and can send again.

The cap on the line is telling you the format

A spoken take accepts between three and one hundred and sixty characters. That upper bound looks like a form limit and is not one; it is the length of the film, written as a number of characters. Ten seconds holds about twenty-five words at three words a second, and a hundred and sixty characters is twenty-five words with room for punctuation. Hand it more and the mouth stops matching the words.

Which makes the cap useful rather than annoying. It settles an argument about the script before the script is written: a spoken take is one sentence, said once. If what you have is three sentences, the honest options are to cut it to one, or to use the voiceover door, where the budget is several times larger because the film can be four times longer. Trying to fit a paragraph into a ten-second mouth produces the specific failure a viewer notices fastest.

A spoken take is also 9:16 by default and 16:9 if you ask. The square and four-by-five placements are not offered on this door, which is the same reasoning as the length: a face delivering a line wants the frame a face is watched in.

The pinned version is who is speaking

This is the property worth knowing, and it is the reason the roster door exists at all rather than being a variant of the creator formats.

A model you save has versions. One of them is pinned, and a spoken take uses the pinned one — not the newest. Refine the model next month, add a version, change the hair, and the film you made in October still has October's person in it. The reference the film was conditioned on is copied onto that shoot when you queue it, so it is not a pointer to a model that can move underneath it. It is a snapshot.

Read in one direction that is a safety property: nothing you do later edits a film you already published. Read in the other it is the whole feature. Fifty product films with the same face in them is a brand asset, and it is only possible if the face is a saved thing rather than a fresh roll of the dice. The creator formats cannot give you that, because they were never storing anybody. Deciding between the doors is therefore less about this film than about whether there is a second one.

A spoken take keeps its own sound, and that is not configurable

One consequence catches people, and it is better met here than in the Studio. A spoken take renders with the video model's own audio, because the speech is that audio. There is no separate voice track laid over a silent picture. So the Sound picker's choices do not apply to it — you cannot put a bed under a spoken take, and you cannot ask for it silent, which would be a spoken take with the speech removed.

Separately, and for a different reason: Nitaris does not supply a music track to put under anything. That decision is about licence rather than craft, and the answer on sound sets out why it stays with you, in the editor you publish from.

Which door the film actually needs

Three questions settle it, and they are in this order on purpose.

• Does anyone need to be visible? If the point you are making is about what the object is or how it behaves, nobody does, and a voiceover over product footage is the cheapest and least risky of the three. An earlier post, “a face is one more thing that has to be right”, counts what a viewer has to accept before they accept a film, and the count goes up when a face appears.
• Is this the only film? If it is, and somebody should be visible, a creator format is the shorter path — there is nothing to create first. A roster model is a thing you build before you can use it.
• Will the same person be back? If yes, that is the roster, and it is worth the setup. Continuity across a catalogue is the one thing the other two doors structurally cannot do, however good any individual film is.

The honest limit

None of this changes the argument for being careful about faces. A model you saved and pinned is still a generated person, and a viewer inspects a generated face more closely than anything else you can put in a frame — the risk that post describes is not reduced by the person being reusable. What the roster changes is narrower: it makes the face a decision you make once and then keep, instead of one the renderer makes again every time.

So the question to ask is not which option sounds best. It is who should be speaking, and whether you intend to hear from them again.