learnVoice & sound
AI narration over silent clips: when it beats voiced renders, priced per 30 seconds
Render silent, narrate on top, or voice the clip in the render? Per-engine credit math for a 30-second piece, the word budget per shot, and when narration loses.
by reeldouble · October 29, 2026 · 6 min read
Two ways to put a voice on an AI video
There are two ways to get a voice onto a generated clip. The engine can make the sound in the same pass as the picture (a voiced render). Or the engine renders a silent picture and a narrator speaks over it afterwards (narration over silent clips).
Search "add AI narration to video" and page 1 is voiceover tools: upload a clip, paste a script, pick a voice. They all assume the picture already exists. None of them ask the question that comes first when you are generating the picture yourself: should the render have been silent to begin with, and what does that choice save or cost on each engine?
That question has a different answer per engine, because engines bill audio differently. We answer it here for a 30-second piece, with the numbers from our rate file, and then cover the part that decides whether narration works at all: writing it to fit the shots.
Which engines charge less for silence
| engine | 720p voiced | 720p silent | silent saves |
|---|---|---|---|
| Kling O3 Pro | 30 | 24 | 20% |
| Veo 3.1 Fast | 30 | 20 | 33% |
| Seedance 2.5 | 70 | 70 | nothing |
| Seedance 2.0 | 45 | 45 | nothing |
| Seedance 2.0 Fast | 50 | 50 | nothing |
Credits per second. Seedance prices by resolution only, so on any Seedance model the silent render costs the same as the voiced one. If you plan to narrate over a Seedance shot, render it voiced anyway: the ambient sound comes free and you can duck it under the narrator.
One 30-second narrated piece, five ways
Every engine has a different longest take, so the same 30 seconds splits into different shots:
| build | shots | silent | voiced |
|---|---|---|---|
| Veo 3.1 Fast | 8 + 8 + 8 + 6 s | 600 | 900 |
| Kling O3 Pro | 15 + 15 s | 720 | 900 |
| Seedance 2.0 | 15 + 15 s | 1,350 | 1,350 |
| Seedance 2.0 Fast | 15 + 15 s | 1,500 | 1,500 |
| Seedance 2.5 | one 30 s take | 2,100 | 2,100 |
At 720p. The cheapest narrated piece on our shelf is Veo silent at 600 credits, a third under voiced. Kling silent saves 180 credits on the same 30 seconds. On Seedance the saving is zero.
One Veo catch: when you attach character references, Veo always renders and bills 8 seconds. With a recurring character in every shot, the Veo build becomes four 8-second shots (32 seconds, 640 silent) and you trim the last two seconds in the edit.
The mixed build we actually use
Pure narration is rare in our own work. Most narrated pieces still have one or two moments where a character says a line on screen. Those shots need a voiced render, because a narrator cannot move a character's mouth. Everything else can stay silent.
A 30-second piece on Kling O3 Pro, six shots of 5 seconds:
| shot | audio | credits |
|---|---|---|
| establishing, narrated | silent | 120 |
| character line on camera | voiced | 150 |
| b-roll, narrated | silent | 120 |
| b-roll, narrated | silent | 120 |
| character line on camera | voiced | 150 |
| close, narrated | silent | 120 |
| total | 780 |
That is 120 credits under all-voiced (900) and 60 over all-silent (720), and nobody's mouth moves without a sound. The voiced shots use the character's banked voice, so the on-camera lines and the rest of the series sound like the same person. Our voice across episodes page covers how that voice is kept.
Write the narration to the shots, not the other way round
The failure we see most in narrated AI video is a script written first and shots cut to fit it. Generated shots come in fixed lengths (Veo in 4, 6 or 8 seconds, everything else in whole seconds from 4), so the shot list should come first and the narration should be written into it.
Brisk narration runs about 2.5 words a second; slower delivery about 2.2. Leave half a second of air at the start and end of each shot so a line does not begin on the first frame or clip at the cut.
| shot length | words, brisk | words, slow |
|---|---|---|
| 4 s | about 7 | about 6 |
| 6 s | about 12 | about 11 |
| 8 s | about 17 | about 15 |
| 15 s | about 35 | about 31 |
A 30-second narrated piece holds roughly 70 words. If your draft is 110, the piece is 45 seconds, and that costs 50% more on every engine. Cut the words before you render, not the shots after. Words per shot has the long version of this table.
Where narration loses
A character speaks on camera. Narration cannot sync a mouth. Render that shot voiced.
The scene needs its own sound. A door slam, footsteps, a laugh from inside the room. Voiced renders make that sound in the same pass; faking it in post means sourcing and timing every effect by hand. On Veo the 10-credit-a-second difference usually buys more than it costs here.
The piece is on Seedance anyway. No saving from silence, so take the sound.
The narrator would be the character. If the voice is meant to be the on-screen character thinking aloud, keep it in the same banked voice. A different narrator voice reads as a second person.
The order we work in
- Shot list with lengths, matched to the engine's lengths.
- Narration written into each shot at about 2.5 words a second.
- Mark the shots where a character speaks on camera: those render voiced.
- Render the rest silent on Kling O3 Pro or Veo 3.1 Fast.
- Export the cut, then lay the narration under it in your audio editor and duck the voiced shots' ambience under it.
The quote under the render button changes the moment you flip the audio toggle, so you can see the saving before you spend it. Build the shot list in the studio, render one silent and one voiced version of the same shot, and listen to which one your narration sits on better.
FAQ
Is it cheaper to narrate over silent AI clips than to render voiced clips?
On Kling O3 Pro and Veo 3.1 Fast, yes: silent is 24 and 20 credits a second on ReelDouble against 30 voiced. On Seedance models silent and voiced cost the same, so there is no saving.
How much does a 30 second narrated AI video cost on ReelDouble?
At 720p, rendered silent: 600 credits on Veo 3.1 Fast, 720 on Kling O3 Pro, 1,350 on Seedance 2.0 and 2,100 on Seedance 2.5. Voiced versions cost 900 on Veo and Kling and the same as silent on Seedance.
How many words of narration fit in an AI video shot?
About 2.5 words a second for brisk delivery, minus half a second of air at each end. That is about 7 words in a 4-second shot, 17 in an 8-second shot and roughly 70 in a 30-second piece.
When should an AI video shot be voiced instead of narrated?
When a character speaks on camera, since narration cannot move a mouth, or when the scene needs its own sound like footsteps or a door. Keep those shots voiced and render the rest silent.
Does turning off audio save credits on Seedance?
No. Seedance bills by resolution, not audio, so a silent Seedance clip on ReelDouble costs the same as a voiced one. Render Seedance shots voiced and duck the ambient sound under your narration.
