Turn Written Scripts into Performed Speech with Grok TTS
Last verified: November 3, 2026
Grok TTS puts real vocal reactions into synthetic speech. Type [laugh], [sigh], [breath], [tongue-click] or [hum-tune] at an exact spot in a script and the voice performs it right there, so the narration sounds acted instead of read aloud. xAI released the model as Grok Text to Speech . It opened the API to developers on March 16, 2026 and built it on the same speech stack that powers Grok Voice, Tesla vehicles and Starlink customer support . Its job is simple: turn written text into natural, expressive spoken audio.
What sets it apart is its two kinds of speech tags. Bracketed inline tags add a single sound: a short or long pause, a chuckle, a giggle, crying, an inhale or exhale, a tsk or a lip smack. Wrapping tags change a whole phrase. They can make it soft, whispered or loud, build or ease the intensity, raise or lower the pitch, slow it down or speed it up, make it sing-song or sung, or add emphasis . Used together, they let one script hold an aside, a nervous laugh and a slow, soft sign-off, all written into the text itself.
xAI built its standalone speech endpoints for voice agents, accessibility tools, podcasts and interactive audio . Here, the model runs as a text-to-speech generator with five preset Grok voices and a Language menu that can detect your script's language automatically.
Explore Grok Imagine's Models
Speech Engine Specs at a Glance
What the generator accepts in each run, and the speech tags the model understands inside your script.
Script length per run
Up to 15,000 characters
Preset voices
Eve, Ara, Rex, Sal, Leo (Eve is selected by default)
Language options
Auto-detect plus 20 languages and regional variants
Inline sound tags
14 bracketed sounds, from [pause] to [hum-tune]
Phrase-level style tags
12 wrapping styles, including whisper, higher pitch, build intensity and singing
Voice and language pairing
Any built-in voice can speak any supported language
Check These Before You Press Generate
Six quick checks that keep speech tags working as cues and keep long or regional scripts sounding right.
Spell sound tags exactly as xAI lists them
Inline tags use square brackets and fixed names such as [laugh], [long-pause], [tongue-click] and [hum-tune]. Copy them character for character instead of writing your own variants like [laughs] or [pause long], so each cue matches the supported list.
Wrap whole phrases with style tags, and close each one
Styles such as whisper, soft, slow or singing go around a stretch of text, with an opening tag before it and a matching closing tag after it. Wrap the complete phrase you want changed, not a single word, and make sure every tag you open is closed.
Time your beats with [pause] and [long-pause]
Put pauses where a speaker would naturally stop: after a punchline, before a reveal, between instructions. Use them instead of strings of dots or dashes, and save [long-pause] for scene or topic changes so the timing doesn't drag.
Choose the regional variant when your audience expects one
Auto detects the language from your text. If Spanish copy is meant for Mexico rather than Spain, or Portuguese for Brazil rather than Portugal, select that exact option in the Language menu instead of leaving the choice to detection.
Split anything longer than 15,000 characters at natural breaks
Break audiobooks, courses and long scripts at chapter, scene or lesson boundaries. Keep the same Voice and Language for every part, so each section is spoken by the same preset speaker.
Read Enhance Prompt suggestions before you accept one
Every word in the Text box is spoken. Before you pick a suggested rewrite, check that it keeps your exact wording, names, figures and speech tags, because your script only changes when you choose a suggestion.
Grok TTS or Gemini 3.1 Flash TTS: Which Voice Generator Fits Your Script?
Both generators run on Vidofy and turn typed text into spoken audio. The rows below show what each one offers here (Text box size, Voice and Language menus, clips per run), plus what xAI and Google document about each model's expressive tags and release status.
| Feature/Spec | Grok TTS | Gemini 3.1 Flash TTS |
|---|---|---|
| Text box limit per run | Up to 15,000 characters | Up to 40,000 characters |
| Preset voices in the Voice menu | 5 presets: Eve — energetic, Ara — warm, Rex — confident, Sal — smooth, Leo — authoritative | 30 named voices, from Achernar to Zubenelgenubi |
| Voice selected by default | Eve — energetic | Kore |
| Language menu | Auto + 20 options, including Arabic (Egypt), Arabic (Saudi Arabia) and Arabic (UAE) | Auto + 87 options, including Arabic (Egypt) and Arabic (World) |
| Expressive tags in the script | Inline sound tags such as [laugh], [sigh], [breath] and [hum-tune], plus wrapping tags for whisper, pitch, pace, intensity and singing | 200+ audio tags such as [laughs] and [whispers]; tags are English-only and can be combined with text in other languages |
| Release status | Generally available ; API opened March 16, 2026 | Preview, launched April 15, 2026 ; marked by Google as a legacy preview model |
| Audio clips per run | 1 audio clip | 1 audio clip |
Match the Generator to Your Script
Script length and long-form production
Gemini 3.1 Flash TTS's Text box holds more than twice as much as Grok TTS's, so a long chapter or a full course module can go in as a single run. With Grok TTS, plan long-form work as a series of sections (chapters, scenes or lessons) and keep Voice and Language the same across them. For ads, explainers, support prompts and short narration, either box holds the whole script comfortably.
Casting range versus regional precision
Gemini's menus cast the wider net. It has six times as many named voices and more than four times as many language options, which helps when you want to audition many speakers or need a language Grok's list leaves out, such as Swahili or Thai. Grok TTS keeps casting to five presets, each labelled with its character in the menu, so choosing a voice is quick. Its Language menu also lists Saudi and Emirati Arabic as separate options next to Egyptian Arabic.
Directing delivery with tags
Both models read expressive tags typed straight into the script, so the workflow is the same: write the line, then mark where the performance changes. Grok TTS separates one-off sounds in square brackets from style tags that wrap a phrase. Gemini 3.1 Flash TTS draws on a large vocabulary of descriptive tags. Use the spelling each model documents. Grok's list has [laugh], for example, while Gemini's uses [laughs].
Pick by Script Length, Cast Size and Target Locale
Use this quick guidance to pick the best option for your workflow.
How to choose: Choose Gemini 3.1 Flash TTS if your script won't fit in 15,000 characters, if you want more than five voices to audition, or if your language or locale isn't on Grok's list. Its Text box, Voice menu and Language menu all offer more. Choose Grok TTS if you want to set Arabic (Saudi Arabia) or Arabic (UAE) as the language, since Gemini's menu doesn't list those options. Otherwise the two work the same way: both read expressive tags typed into the script, both offer Auto language detection, and both return one audio clip per run.
Turn a Script into Speech in Four Moves
Four steps take you from a typed script to a finished audio clip.
Write your script in the Text box
Paste or type up to 15,000 characters, adding bracketed sound tags such as [pause] or [laugh] wherever you want them heard. If you're starting from a rough idea, press Enhance Prompt under the box. Your text only changes if you pick one of its suggestions.
Choose a speaker from the Voice menu
Eve — energetic is selected by default. Switch to Ara — warm, Rex — confident, Sal — smooth or Leo — authoritative to fit the role your script needs.
Set the Language menu
Leave it on Auto to detect the language from your text, or pick an option such as Portuguese (Brazil) or Arabic (UAE) when a specific regional variant matters.
Press Generate
Each run returns one audio clip. Listen to it, adjust your tags, voice or wording, and generate again for a new take.