Introduction
Send up to 10,000 characters to elevenlabs/v4/text-to-speech and choose from 21 premade voices. Guide the spoken delivery with inline audio tags, select a language and audio format, then review the finished recording for your narration or character line.
| Field | Value |
|---|---|
| Input | Text |
| Output | Audio |
| Speech invocation name | elevenlabs/v4/text-to-speech |
| Input capacity and cues | Required text, up to 10,000 characters, with inline audio tags |
| Voice cast and default | 21 premade choices; Roger is the default voice |
| Export choices | 9 formats; |
| Observed v4 request | November 13, 2026: Daniel, 84 characters, en, seed 7, |
Key features & highlights
Put a cue beside the spoken line
Eleven v4 accepts square-bracket audio tags within text, including [whispers] and [laughs]. Begin with a passage whose wording already conveys the intended mood, then add a cue where the delivery should change. Our test used one tag, so audition additional combinations before building a full scene around them.
Audition within a 21-voice cast
The voice_id selection supplies 21 premade voices, with Roger selected by default. Compare Daniel, Sarah or Lily against the same passage before revising the script. ElevenLabs says stability shapes variation and similarity_boost shapes adherence to the chosen voice; clone_strength does not alter premade voices here.
A Daniel take with recorded timings
On November 13, 2026, our 84-character v4 call used Daniel, language en, seed 7 and mp3_22050_32. It returned 5.17 seconds of MP3 audio and finished in about 4.7 seconds. That observation is a reference for one request, with no claim about longer scripts or every voice.
Choose the reading and the audio format
Language selection offers auto or 91 codes, while output_format has nine options and defaults to mp3_44100_128. Use text_normalization_mode to decide how numbers and abbreviations are expanded. Check a passage containing your actual names and dates before producing the rest of the script.
Request parameters
| Parameter | Required | Description |
|---|---|---|
text | Yes | Supply the spoken script in this required field, with no more than 10,000 characters. Square-bracket cues such as [whispers] belong beside the words they should direct. at least 1 characters · at most 10000 characters |
voice_id | No | Choose from 21 premade voices; Roger is selected by default. Audition the same passage with Daniel, Sarah or Lily before settling on the cast. |
language | No | Select auto for detection or one of 91 language codes for an explicit choice. For a short English line, en makes the intended language clear. |
stability | No | ElevenLabs says stability controls variation between generations, with lower settings allowing more emotional range. |
similarity_boost | No | ElevenLabs says |
output_format | No | Choose among nine audio formats, with |
text_normalization_mode | No | Use auto, on or off to control expansion of numbers and abbreviations. If a date remains ambiguous, write the intended spoken form into the script and audition it. |
clone_strength | No | Changing |
seed | No | Keep a fixed seed while comparing a single edit to the script or settings. It supports controlled comparisons without guaranteeing the same recording. |
Pricing
$0.096 / 1,000 characters
Eleven v4 usage is metered by characters in text. Calculate the estimate using the character-based price and unit displayed on this page.
Use cases
A lesson with an exact instruction
Submit "Open the worksheet and underline the verbs in the second paragraph" with Daniel and language en. Use the returned narration as the spoken instruction beside that course exercise.
A product walkthrough
Send "Select Export, choose the destination folder, then confirm the file name" with Sarah. The resulting voice-over can accompany a screen recording of those three actions.
A quiet game character line
Give Lily the text "[whispers] The guard is still outside the gate." The output is a single character take that you can audition against the game scene.
A localized app instruction
Provide your approved translation of a sign-in instruction in text and select its language code. The endpoint returns spoken audio for that language version; translation remains a separate preparation step.
An advertisement closing line
Use "Bring your next idea to the workshop this Saturday" as the script and audition a premade voice. Place the selected recording at the close of a promotional video after checking the delivery.
Prompt guide & pro tips
Create the text input
Place your narration in text and use a short passage containing any difficult names or numbers. Keep its total length at or below 10,000 characters.
Select a premade cast member
Set
voice_idto a supported choice, then choose language andoutput_formatfor the intended recording. Leaveclone_strengthout when working with these premade voices.Invoke and review the recording
Call elevenlabs/v4/text-to-speech with the JSON input described below. Listen to the audio file before reusing the selected settings for another script section.
ElevenLabs v4 API for directed narration
ElevenLabs v4 API exposes Eleven v4, ElevenLabs' text-to-speech model, as elevenlabs/v4/text-to-speech. Supply a script and a premade voice to produce a single-speaker recording for a lesson, product video or character line.
ElevenLabs says v4 targets content creation and long-form audio in its September release note. This endpoint accepts up to 10,000 text characters; custom voice IDs and multi-speaker dialogue input are unavailable here.
Request example: an English character line
Call elevenlabs/v4/text-to-speech with the following JSON as the endpoint input. This example specifies Daniel, English, an MP3 format and seed 7; its script is illustrative rather than the text used in our timed test.
{
"text": "[whispers] Keep the lantern covered until we reach the bridge.",
"voice_id": "Daniel",
"language": "en",
"text_normalization_mode": "auto",
"output_format": "mp3_22050_32",
"seed": 7
}
Use the page's API request example to submit these input fields with your credentials. The public invocation name above identifies the model; this input object does not describe the surrounding HTTP request envelope.
Response example: the observed audio result
Our separate 84-character Daniel request produced an MP3 recording lasting 5.17 seconds. The following is a description of that completed audio result, rather than a JSON response schema or a claim that the illustrative script has the same duration.
Audio: MP3
Selected output_format: mp3_22050_32
Recording duration: 5.17 seconds
Request completion: approximately 4.7 seconds
Parameters for a single-voice recording
The required input is text; the other fields select the voice, language, delivery or export behavior. Use these fields to prepare your narration.
| Parameter | Accepted setting or role | Practical choice |
|---|---|---|
text | Required; up to 10,000 characters | Write the spoken script; place [whispers] or [laughs] beside the relevant passage. |
voice_id | 21 premade voices; default Roger | Audition Daniel, Sarah or Lily with the same excerpt. |
language | auto or one of 91 language codes | Choose en for an English take or select the intended language code. |
stability | Delivery variation | Adjust against a fixed passage; see the publisher explanation below. |
similarity_boost | Adherence to the selected voice | Listen for the voice characteristics you want to retain. |
clone_strength | No effect on premade voices | Omit it for this endpoint's cast. |
text_normalization_mode | auto, on, off | Check how dates, quantities and abbreviations are spoken. |
output_format | Nine options; default mp3_44100_128 | mp3_22050_32 matches the format in our v4 measurement. |
seed | A seed for comparing changes | Reuse it when adjusting one element; identical audio is not promised. |
Pricing and script length
Billing follows the character count of text, so use the rate and billing unit displayed on this page to estimate usage. Audio duration is a result measurement, not the metering unit for this endpoint. Plan separate script sections when your narration exceeds the 10,000-character request limit.
Getting a performance you can publish
Start with a short excerpt containing the same names, numbers and tone as the intended production. A lesson instruction such as “Open the worksheet and underline the verbs” gives you a focused passage to audition before recording the surrounding explanation.
- Choose a voice before rewriting the delivery cues.
- Add an inline tag where the performance should shift, then listen to that transition.
- Set the intended language and check number readings with your chosen normalization mode.
- Keep a seed fixed while changing a single setting, and save the settings beside the selected take.
ElevenLabs says lower stability allows greater emotional variation and similarity_boost controls adherence to the original voice in its voice settings documentation. Judge the resulting delivery by listening to your passage rather than assuming a setting fits every voice.
Troubleshooting an awkward read
Revise the relevant input when a recording sounds wrong: spell out an ambiguous date, remove competing cues or audition another premade voice. ElevenLabs says complex numbers and abbreviations can be mispronounced in its speech guidance. If a request fails, check that text is present and within the limit, and that the voice, language and format match the choices on this page.
Test evidence and model selection
Our November 13, 2026 calls provide two concrete samples with different inputs. Each contained one audio tag; these results are observations rather than a controlled comparison.
| Variant | Input and voice | Language / seed | Format | Audio length | Completion |
|---|---|---|---|---|---|
| Eleven v4 | 84 characters; Daniel | en / 7 | mp3_22050_32 | 5.17 seconds | About 4.7 seconds |
| Eleven v4 Turbo | 75 characters; Sarah | Not specified in the fact sheet | mp3_44100_128 | 3.40 seconds | About 7.6 seconds |
Choose v4 for reviewed final takes, or consider Eleven v4 Turbo for drafts, high-volume work and latency-sensitive flows. The Turbo measurement above does not reverse that use-case guidance: script, voice and export settings differed. If you need a custom voice or multi-speaker input, evaluate an alternative that explicitly documents those inputs before planning an integration.
Recording rights
Publish recordings from scripts and voices you are entitled to use, without impersonating a real person. Review the completed audio before publishing it as part of a course, application or promotional video.
Provide text and select voices you have the right to use for the recording. Do not use the generated speech to impersonate a real person.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "elevenlabs/v4/text-to-speech",
"input": {
"text": "Welcome back. Your files finished rendering a minute ago and are ready to download.",
"language": "auto",
"voice_id": "Roger",
"output_format": "mp3_44100_128",
"text_normalization_mode": "auto"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.




