How much one take can read depends on the model you pick on Voice. The counter under the script shows the characters you've written against that model's limit, and turns red when you reach it.
The three models
| Model | One take reads up to | Languages | Audio tags |
|---|---|---|---|
| Eleven v4 | 10,000 characters | 94 | Yes |
| Eleven v4 Turbo | 10,000 characters | 94 | Yes |
| Eleven v3 | 5,000 characters | 74 | Yes |
- Eleven v4 is the current flagship, the most expressive, and marked Best in the model menu.
- Eleven v4 Turbo is the same generation made faster, for when speed matters more than nuance. A thousand characters cost 4 credits on it, against 8 credits on Eleven v4.
- Eleven v3 is the previous flagship. It performs audio tags in its own manner, and reads half as much in one take.
The model menu shows what your script costs on each one before you pick.
A script longer than one take
The script field stops at the model's limit. If you switch to Eleven v3 with more than 5,000 characters written, Generate stays off and the bar suggests Eleven v4 or a shorter script. For anything longer than 10,000 characters, split the script into takes.
The line under the script also estimates how long it runs aloud, at a narrator's pace of 150 words a minute. Audio tags are performed rather than read, so they don't count toward that time, though they do count as characters.
Dialogue
A dialogue, made with Add speaker, holds up to 10,000 characters across all its lines, on any of the three models, with up to 10 voices and 100 lines. It's read in parts and joined into one file.
Languages
Leave the language on Auto-detect and it's worked out from the script's words, or pick one from the list to have the take read in it. Eleven v4 and v4 Turbo read 94 languages. Eleven v3 reads 74 of them, and the Language list offers only those when v3 is picked.
The 20 that only v4 and v4 Turbo read are Amharic, Asturian, Burmese, Cantonese, Fula (Pulaar), Kamba, Lao, Luganda, Maltese, Maori, Mongolian, Occitan, Odia, Shona, Sorani Kurdish, Tajik, Uzbek, Wolof, Yoruba and Zulu. If one of them is picked when you switch to Eleven v3, the bar says v3 doesn't read it.
What the voices and settings do
Each voice in the Voice list shows its traits and, where it's known, its accent; play a sample before you pick. How a voice reads is set two ways:
- In the script, with audio tags such as [whispers] or [laughs], CAPITALS for emphasis and an ellipsis for a pause. Emotion is written here.
- Under Settings, with Stability, from Expressive to Steady, and Clarity, from Loose to Close to the voice. Lower stability performs audio tags more freely.
Formats
- MP3 128kbps: standard quality, a smaller file.
- PCM 16kHz, 22kHz and 24kHz: raw audio, from telephony quality to higher quality, downloaded as WAV.
Where else a voiceover's length matters
- A voiceover card on a board reads up to its model's limit, counting its own words and the text wired into it together.
- A talking clip on a board takes a voiceover of up to 15 seconds on Kling Avatar and Kling Avatar Pro, and up to 59 on OmniHuman 1.5.
- A lip sync of a clip you have takes up to five minutes of voiceover.
- A finish card takes a voiceover of up to 60 seconds.
See Sound in your video: voices and lip-sync and Put a voiceover or a song in a video.
Related
Still stuck?
A person reads every message.
Usually within 4 hours on a working day.
Have an account? Log in and write to us from Help in the app.