Skip to main content

Uncensored — This model declares no content filter. SpicyAPI forwards the request without adding a platform content scan; legal and consent duties remain with the customer.Browse every model at this filtering level
google/gemini-2.5-pro-tts/text-to-speech

Gemini 2.5 Pro text-to-speech reads a script in one of 30 voices and returns a 24 kHz mono WAV. Accent, style and pace take plain-language direction, bracketed cues like [sighs] are performed, and a dialogue can use up to two voices.

From $0.00045/s10% below the publisher's list price
Run model

Workbench

Input

Text to speech
Provide at least one of theseRequirement met
127/5000
The words to be spoken, exactly as written. Cues in square brackets such as [sighs] are performed, not read. Up to 5,000 characters. For a conversation between two voices, use dialogue instead. at least 1 characters · at most 5000 characters

About 11 sec of audio (rough estimate)

Dialogue0/20
A scripted conversation for up to two voices, used instead of text: up to 20 lines, each with speaker (a name from speakers) and text of up to 250 characters. Lines are performed in order. Billed per second of audio returned, like text. at least 1 items · at most 20 items
Speakers0/2
The cast for dialogue: one or two entries, each with a name the lines refer to, a voice_id, and optionally accent, style, pace and profile (a short description of who is speaking, which is not read aloud). Anything a speaker leaves out falls back to the top-level accent, style and pace. Does not change the price per second. at least 1 items · at most 2 items
One of 30 prebuilt voices, each with a fixed character. Defaults to Kore. It voices text; in dialogue each speaker sets its own voice_id.
Style
WhisperEmpatheticVocal smileNewscasterPromo hypeDeadpan
Overall manner of the reading. Leave empty to use the voice's own default manner. In dialogue it applies to speakers that set none.
Pace
NaturalRapid-fireUnhurriedStaccato
Rhythm of the reading: natural, rapid_fire (fast and clipped), the_drift (slow and lingering) or staccato (short, separated phrases). Omit it to keep the voice's own rhythm; in dialogue it applies to speakers that set none. Slower pacing yields more seconds of audio, which is what you pay for.
English accent for the reading. Defaults to neutral. In dialogue it applies to speakers that set none.
0/1000
Free-text direction on how to read: pace, mood, distance from the microphone, who is listening. Shapes the delivery and is not spoken. Up to 1,000 characters. at most 1000 characters

Never read aloud. It only shapes how the script is performed.

0/500
Where and to whom the words are spoken, such as “a quiet hotel room after midnight”. Shapes the delivery and is not spoken. Up to 500 characters. Does not change the price per second. at most 500 characters
Temperature
Variation between takes, from 0 to 2: higher values make the delivery less predictable. Leave it unset to use the model's default. Does not change the price per second. no lower than 0 · no higher than 2
Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run
Example result
0:00 / –:––
Avatar
Speech to text
Lip sync

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

Direct a reading the way you'd brief a voice actor

Gemini 2.5 Pro TTS turns up to 5,000 characters of script into a single WAV file, spoken by one of 30 prebuilt voices. Four controls shape the performance: voice_id picks the voice, style sets an overall manner such as whisper or newscaster, accent narrows an English reading to one of eight varieties, and instructions takes a sentence or two of plain-language direction. Cues written in square brackets, such as [sighs], are acted out rather than read aloud. SpicyAPI adds no content filter, and in our tests the model read our strongest monologue and narration test scripts in full. You're billed per second of audio actually returned.

Open the playground
Voices
30 prebuilt, each with a fixed character
Script length
Up to 5,000 characters per request
Direction
voice_id, style, accent (English) and instructions
Output
One WAV file, 24 kHz, 16-bit, mono
Longest clip
About 655 seconds (16,384 audio tokens)
Billing
Per second of audio, 10% below Google's list price
Direction
Plain-language notes instead of a settings sheetstyle picks a manner such as whisper or deadpan, accent narrows an English reading to eight varieties, and instructions takes a sentence or two about pace, mood and distance from the microphone. None of it is spoken; it only shapes how the fixed voice performs the script.
Delivery
Bracketed cues are performed, not readWrite [sighs], [whispering] or [laughing] inside the script and the model acts the cue out instead of speaking the words, so a single text field carries both the words and part of the performance.
Content
No content filter, scripts read as writtenSpicyAPI applies no content filter of its own. In our tests the model read all three of our strongest test scripts, a first-person monologue, an audiobook-style passage and a scene carried by bracketed cues, in full, without skipping words or cutting the clip short.

What the Gemini 2.5 Pro TTS API does well

  • Four dials instead of one voice knob

    voice_id, style, accent and instructions work together: pick a voice, set an overall manner, narrow the accent for an English script, then add a short note for anything the first three don't cover. Changing the mood of a line doesn't require switching voices.

  • Delivery notes that never get spoken

    instructions takes up to 1,000 characters of plain-language direction — pace, mood, who's listening, how close to the microphone — and none of it ends up in the audio. Script and direction stay cleanly separated.

  • Bracketed cues acted out inside the script

    [sighs], [whispering], [laughing] and similar cues are performed at the point they appear in the text rather than spoken as words, so a scene's beats can be blocked directly in the script.

  • One lossless file, ready for downstream tools

    Output is a single WAV file at 24 kHz, 16-bit, mono — no compression artifacts to clean up before mixing, editing or feeding into another model.

  • Billed only for the seconds you get back

    A hold covers the longest clip the model can produce while the task runs, then the charge settles to the actual length of the returned audio once it's measured. A failed request isn't charged.

  • Unfiltered scripts read as written

    SpicyAPI adds no content filter. In our tests, our strongest first-person and narrated test scripts were read in full, and bracketed cues such as [breathy] were performed, with nothing skipped or cut short.

Who Gemini 2.5 Pro TTS is for

  • Audiobook and serial fiction producers

    One voice, kept consistent across chapters

    Pick a voice and a short instructions note once, then reuse both across an entire book. The same voice with the same direction reads consistently from chapter to chapter.

  • Chat and companion app builders

    A character voice with its own mood

    style and instructions turn a flat script into something that sounds spoken to one listener — a whisper style with a note about closeness to the microphone reads very differently from the same line in newscaster style.

  • Uncensored audio publishers

    Unfiltered scripts read without edits

    In our tests the model read our strongest first-person and narrated test scripts in full, performing bracketed sounds and breaths rather than skipping them. What you publish, and to whom, stays your call.

  • Video and animation pipelines

    A voice track to hand to a lip-sync model

    Generate the line here, get back a WAV file, then pass it to a talking-avatar or lip-sync model so a face performs it. The audio's length sets the length of the clip downstream.

How to call the Gemini 2.5 Pro TTS API

  1. Write the script in text

    Up to 5,000 characters, exactly as it should be spoken, including the punctuation you want to set the pauses.

  2. Pick a voice_id

    One of 30 prebuilt voices. Defaults to Kore if you leave it out.

  3. Add style and accent if the default manner isn't right

    style sets an overall manner such as whisper or newscaster; accent narrows an English reading to one of eight varieties. Leave either out to use the voice's own default.

  4. Direct anything else in instructions

    A sentence or two about pace, mood or distance from the microphone, up to 1,000 characters. It shapes the reading and is never spoken.

  5. Read the WAV back once the task settles

    The file is 24 kHz, 16-bit, mono, kept on SpicyAPI storage for 14 days. Billing settles to the actual seconds returned once the audio is measured.

Gemini 2.5 Pro TTS worked examples

  • A whispered late-night voice note

    style set to whisper and a short instructions note about closeness to the microphone turn a plain line into something that reads as spoken quietly, close up, to one listener.

    {"text": "Hey... I know it's late, but I couldn't stop thinking about you. [sighs] Call me back when you get this.", "voice_id": "Enceladus", "style": "whisper", "instructions": "Close to the microphone, quiet, a little breathless, as if speaking only to the person listening."}

  • A romance audiobook chapter opening

    A steady narrator voice, kept constant across a whole book, needs only voice_id and a short instructions note about pacing to carry it from one chapter to the next.

    {"text": "The rain hadn't stopped since morning, and neither had the letter sitting unopened on her desk. She knew, the moment she saw the handwriting, exactly who it was from.", "voice_id": "Charon", "instructions": "Warm, unhurried narration, like reading aloud to someone curled up nearby."}

  • A flirty game line with a specific accent

    accent narrows the reading to one English variety, useful for character voices aimed at a specific market, while style adds a knowing warmth to the line.

    {"text": "You actually came back for me? [whispering] I was hoping you would.", "voice_id": "Aoede", "accent": "british_rp", "style": "vocal_smile"}

Gemini 2.5 Pro TTS API questions

Treat style, accent and instructions as three separate levers rather than one settings panel: pick style for an overall manner, reach for accent only when the script is in English and a specific variety matters, and save instructions for the details neither covers — how close to the mic, who's listening, how tired or amused the voice should sound. Keep the note short; one or two sentences steer a reading more reliably than a long list of adjectives. If a scene needs a beat — a breath, a pause, a laugh — write it into the script itself as a bracketed cue rather than describing it in instructions, since cues are performed exactly where they appear.

All Gemini models8

Audio models3

Text models5