AI voice generator: type text, choose a voice, get an MP3

This is a straightforward text-to-speech converter: you write or paste text, choose an engine and a voice, and the tool sends that text to the selected provider and hands back an audio URL. Nothing about the input text is analyzed for meaning, translated, or rewritten — what you type is (mostly) what gets spoken. The one exception is Gemini's silent 4,000-character cutoff, which quietly drops anything past that point rather than reading it.

It is not a voice cloning tool — none of the four engines let you upload a sample of your own voice to imitate; you are choosing from each provider's fixed voice roster. It is also not a dubbing tool: there is no way to feed it an existing video or audio track and get a re-voiced version back in the same interface — that workflow exists separately as Translate & Dub, which chains this same TTS step after a translation call and before a talking-avatar animation step. If what you actually want is a photo that speaks your translated script, that combined flow is the right tool, not this one.

It is also not a script writer. Unlike VdoBloom's faceless-channel or creator-reel tools, this page assumes you already have the words — its only job is turning them into audio.

ElevenLabs Multilingual V2 and Turbo 2.5 are the two options under the "ElevenLabs" group in the provider dropdown, and they share the same 21-voice list. The difference between them is not exposed in the UI copy beyond the model names — Turbo 2.5 is ElevenLabs' faster-generation model, Multilingual V2 its more general one — but pricing, voice list and the 5,000-character cap are identical between the two, so switching between them is a quality/speed trade-off, not a cost one.

Gemini 3.1 Flash TTS is the only engine with a language selector visible in the UI — a 24-language dropdown (English, Hindi, Tamil, Telugu, Marathi, Bengali, Gujarati, Kannada, Malayalam, Punjabi, Urdu, French, German, Spanish, Italian, Portuguese, Japanese, Korean, Chinese, Arabic, Russian, Indonesian, Turkish, Vietnamese) that also appears for xAI TTS. It is the largest voice roster (30 named voices, e.g. Zephyr, Puck, Kore, Autonoe) and the broadest language support of the four, at the cost of the silent 4,000-character truncation.

xAI TTS is the smallest option: six voices (una, leo, eve, ara, sal, rex) described in the UI as "expressive." It shares the same 24-language dropdown as Gemini but the code has no enforced character ceiling for it specifically — practically, very long inputs will still be constrained by whatever xAI's own API accepts, which isn't asserted anywhere in this product's code.

ElevenLabs, by contrast, shows no language dropdown at all in the UI. It takes an optional ISO 639-1 language code under the hood, but the interface leaves language detection to the model itself rather than asking you to pick one — in practice this means you type in whatever language you want spoken and the multilingual model handles it without an explicit selector.

The most direct use is a standalone voiceover for something being edited elsewhere — a YouTube video, an explainer, a product demo — where you need a clean narration track as an MP3 to drop into your own timeline. Because output is character-priced rather than duration-priced, a 30-second read and a 3-minute read of the same character count cost the same, which makes it easy to budget a script before recording.

It is also the audio engine underneath two other VdoBloom features that need spoken lines rather than a standalone file: Translate & Dub uses this same endpoint (defaulting to ElevenLabs Multilingual V2) to voice a translated script before animating a photo with it, and the faceless-channel and creator-reel flows generate scripts that ultimately need a voice. If you are building one of those, you do not need to visit this page separately — but understanding its per-1,000-character pricing tells you what the voicing portion of those combined jobs actually costs.

For anything needing more than 5,000 characters of ElevenLabs output — a longer article read-aloud, for instance — the practical answer in the current product is to split the text into chunks and generate each separately, since there is no automatic chunking or multi-part stitching built into this tool.

All four engines charge per 1,000 characters of input text, rounded up to the next whole thousand — a 1-character request and a 1,000-character request cost the same, because both round up to one unit. There is no minimum-generation discount and no volume discount for longer text; the rate per 1,000 characters is flat.

Stay under 5,000 characters on ElevenLabs and under 4,000 on Gemini — going over on ElevenLabs gets you a clear error and no charge, but going over on Gemini silently drops the tail end of your script with no warning, so it is worth checking the character count in the textarea before generating, especially for a longer narration.

If a script needs to run long, split it into two or more generations at natural sentence or paragraph breaks and stitch the resulting MP3s together in your video editor — the tool does not do this automatically, and there is no "continue from here" parameter exposed in the interface even though the underlying ElevenLabs API supports previous/next-text context.

Try a voice on a short line first. Because pricing is per 1,000 characters rounded up, a 50-character test costs the same single unit as a 999-character one, so testing a voice choice on a short sentence before committing to the full script is effectively free relative to just generating the whole thing once.