How AI voiceover for video actually works (the file, not the timeline)

The search box sold you a timeline

Type AI voiceover, how to add voiceover to a video, or faceless YouTube and the pages that come back are mostly the same product: an editor that generates the voice, the captions, the avatar, and sometimes the B-roll in one click. Those pages are not wrong about their own tools. They are incomplete if the next thing you actually need is a take.

A voiceover is audio. Picture is picture. Mixing them is an edit. Collapsing all three into one button is convenient until the script changes, the footage is already locked, or the mouth on screen does not match the words.

The product page for this job is YouTube and Video. This post is the craft version: the three jobs hiding under one search, audio-first versus picture-locked, and what YouTube's own pages actually say. The click path for CapCut, Premiere, iMovie, and the rest already lives under how to do a voiceover.

Three jobs, not one generate button

People type the same phrase for three products. Mixing them up is how a Short gets written like a lecture, or a locked cut gets a two-minute take that will not fit.

Audio-first, then picture

If the video does not exist yet, generate the voice first. Measure it. Cut the picture to the take. That is the boring method that keeps narration and shots on the same clock.

House rule of thumb: about 1,000 characters of script is about a minute of audio. A 90-second voiceover is about 1,500 characters on the YouTube and Video page. Speed varies by voice, so treat it as an estimate. If the cut needs to land on a sentence, generate that sentence as its own file.

The words still have to be speakable. We already wrote that fight in how to write a voiceover script. This post assumes you have a spoken column, not a two-column camera script pasted into the box.

Picture-locked is a different clock

If the footage is already cut, the voice has to fit the picture. Stretching a take to cover a locked sequence is how you get a rush, then you blame the voice.

Cut the sentence that explains the company founding date. Or generate the line on MAI Voice 2, which is the engine on this roster with real style and speed controls. The Product Demos worked example is the same script at launch energy or tutorial calm, set on purpose rather than luck.

Do not lay a generated read over a person whose mouth is already moving. That is not a sync problem we can fix, because we do not lip-sync. Cut to B-roll for those lines, or record the person.

Shorts are one line

A Short is not a compressed lecture. Pace is the job. Generate the hook on its own. Audition it at full energy. Cut anything that takes a breath before the point.

On this roster the starting engine for that read is Grok Voice: five English personas, a confident characterful take that can hold a retention hook without sounding like a screen reader. When the line needs an acted cue, move that line, not the whole video, onto Gemini Flash.

Long-form is the opposite: a steadier read, with chapter breaks scripted as pauses so YouTube chapters land on clean audio boundaries. Same studio, different clock.

Same file, two engines

The usual failure is directing a draft, or drafting a keeper.

Write the words on Kokoro until they are actually the words. It is the unlimited draft engine: a clean, plain read, fast enough that rewriting a sentence is cheaper than arguing about the take. When a line has to sound like a person leaning in, move that line onto the engine that matches the job.

For most YouTube narration that engine is Grok Voice. For a cue in brackets, Gemini Flash. For a demo that needs intensity without a cue, MAI Voice 2. The longer routing argument is in pick the engine for the job. The short definition of a cue is what is a TTS cue.

Beds under a faceless cut are a second export. Generate the voice here, generate a quiet instrumental in Sound and Music, mix them in the editor. Duck the music under the voice. Do not ask the music model for the narration.

YouTube's pages, not the rumor

Two different YouTube rules get mashed into one internet sentence. Disclosure is about whether a viewer might think something real happened. Monetization is about whether the channel looks mass-produced. They are not the same form.

As of October 2026, YouTube's disclosing use of GenAI content page requires a disclosure when realistic AI content could mislead: a real person appearing to say or do something they did not, real events altered, a realistic scene that did not occur, or music that is the main focus of the video. The same page lists cloning one's own voice to create voiceovers or dubs among the examples that do not need that disclosure. A roster voice narrating your own footage is not the 'real person speaking' case. When in doubt, tick the box in upload. YouTube states that disclosing AI use does not limit a video's reach or its eligibility to earn. We are not YouTube. Read their page. Verify before you rely.

As of October 2026, YouTube's channel monetization policies still do not ban synthetic narration. What they reject is inauthentic content: mass-produced, templated videos, including AI-generated work that gives the impression of a factory and adds no original insight. An AI-voiced channel is judged on the research, writing, and editing. The voiceover is a production choice. Originality is the policy. Same caveat: store rules move. Check the live page before you file.

Where it still fails

A fluent take can still be the wrong file. These are the misses we see when people treat the search phrase as a spec.

• Pasting a two-column camera script, so the engine reads CAMERA PUSHES IN.
• One ten-minute generate for a cut that needed twelve files.
• A Short written like module three of a course.
• A picture-locked demo stretched until the voice sounds late for its own sentences.
• Narration laid over a talking head, then blaming us for the mouth.
• Cues on a plain-read engine, then blaming the voice for a flat take.
• Asking the studio to export captions, visemes, or a finished video.
• A clone of a real YouTuber (we do not offer that, and acceptable use would not either).

Honest limits

We generate files. We do not sit inside CapCut, Premiere, or YouTube Studio. The platform click path is how to add a voiceover to a video, or the YouTube spoke if that is the host. The output here is audio you own: commercial use, no watermark, same as the rest of the studio (you own what you make here).

No lip-sync. No visemes. No caption file. No avatar. No in-editor timeline. The console API tab is still labeled soon; we will not pretend a public render endpoint exists.

Cue-following is Gemini Flash only. The other engines on the roster read plainly. We strip brackets so they do not speak the direction. We do not pretend they performed it.

A generation has a character cap. Split a long video. Do not paste a 40-minute script into one box and call the truncation a voice problem.

Premium generation sits inside a flat monthly allowance (published free-tier premium budget: 10,000 characters, about 10 minutes by house math). Kokoro drafting stays unlimited so you can rehearse a hook without burning keeper budget on every micro-edit. Paid plans raise the flat allowance; they do not put you on a per-character anxiety meter.

Voice cloning is not a live path today. When it ships it will be English-first with a consent attestation. Until then, roster voices and personas are the honest path.

Cut one this week

The shortest proof is a video you already owe, not a roundup of vendors.

• Pick a 90-second explainer or a 12-second hook. Count it.
• Write the spoken column only. If a sentence is hard to say, cut it.
• Draft the words on Kokoro until they are the words.
• Move the keeper onto Grok Voice for a narration, or Gemini Flash if the line needs a cue.
• Export. Drop the file on a track in the editor you already use.
• If the picture is locked, cut the script until it fits. Do not stretch the take.
• Listen once without looking at the page. If a word is wrong, fix the page, not the voice picker.

Open the studio, not the avatar

If you came here from an AI voiceover or add-voiceover-to-video search, the honest product is a spoken file you can hear this afternoon. Open YouTube and Video, or start in Text to Speech if you already have the lines.

When you want the exact clicks in an editor, stay on voiceover. When you want the full studio, create a free account. Written to be heard includes the faceless cut, as long as we do not lie about what that file is for.