ElevenLabs produces a voice. Cutroom produces the ad around it.

Control over delivery, a large library of voices, cloning, and output good enough that listeners stop noticing it is synthetic.

Many languages from one script, which matters when the same audiobook, course or product tour has to exist in nine markets.

And an API, so voice can sit inside a product rather than inside a person's afternoon.

Consistency is the underrated part. One voice across two hundred assets is a brand decision, and only a voice library holds it.

For an audio deliverable that is the correct buy, and nothing here competes with it.

Here is what a clean read still needs. Pictures, captions, pace, a first line that interrupts, an export sized for a feed.

That is four more jobs, and a voice engine correctly claims none of them.

Cutroom does every one of them from a single 100-credit batch, and generates a read at 70 credits a minute when nobody can film.

What comes back is a 9:16 MP4 with captions burned in and coverage sitting on the words you marked.

The read arrives clean. Then it needs pictures, captions, pace, a first line that interrupts, and an export sized for a feed.

That second stage is where the hours live, and it is the stage a voice engine correctly does not attempt.

There is a harder problem underneath. A generated read is even by design, and evenness is the tell that makes a claim sound rented.

In direct response the hesitation is doing work. The half-second before a number is what makes the number sound arrived at rather than written.

That is why the strongest input for a paid ad is usually a read you already gave, in your own voice, with the roughness intact.

Cutroom is built around that recording. The voice module exists for the case where no recording is possible.

The half-second before a number is what makes the number sound arrived at rather than written.

Bring that recording here and the roughness survives into the export, which is the point of it.

Upload a talking-head take of up to three minutes. It returns transcribed, drafted and captioned as a vertical MP4.

Highlight a phrase and a clip lands over exactly it. Swap the clip, or search millions of free ones.

Delete a line and the cut rebuilds around the gap. Change the pace and the whole piece re-cuts.

Emphasise the word your claim turns on so it pops in the captions. Fix a mis-heard word and the burned-in text follows.

Where nobody can film, the avatar module takes a photo you own plus a generated voice-over and produces a lip-synced presenter.

That combination costs 320 credits a minute for the render and 70 for the voice, on top of the 100-credit batch.

Coverage is capped by pace: 40 percent on chill, 45 on normal, 52 on fast. The presenter stays the spine of the ad.

The route in is a video take of up to three minutes. The route out is a 9:16 MP4.

There is no audio-only export. This cannot hand you a WAV at any price.

There is no voice cloning and no library of characters to audition. One generated voice per job, priced per minute.

There is no translation route. One take, one language, one finished file.

There is no way to reuse one generated voice across a campaign. Each job generates its own.

There is no timeline underneath and nothing of yours gets stamped on the file.

Those are the boundaries. Inside them sits a finished ad, which is what the next section prices.

Winners run at 5 to 8 percent of creatives, per Motion's analysis of 550,000+ Meta ads.

Six a month draws under one winner. Thirty draws two, and the difference is what a second attempt costs.

Here that is 20 credits a finished minute and about four minutes of attention, from a take already batched.

One finished minute from a fresh take is about 120 credits. An avatar minute adds 320 on top.

Seven days and 300 credits with no card covers two finished ads before any decision is made.

Keep the voice specialist for anything where sound is the deliverable. It wins that outright.

Bring the read here when the deliverable is a file you upload to a feed.