Clean narration: model and settings

Pick the right model, start from sane slider values and prep your text so numbers read correctly.

Good ElevenLabs narration depends on three choices, in this order according to the docs: the voice, the model, then the settings. Getting the first two right saves a lot of slider fiddling.

Choose the model for the job

Eleven v4 is the current top-quality model for content such as voiceovers and character voices. Eleven v3 is the expressive model for dramatic delivery. Multilingual v2 is described as the most stable choice for long-form generation, useful for long explainers. Flash v2.5 is fast and cheap, which suits drafts and automation, but as a smaller model it handles numbers and abbreviations less well.

Start from the defaults

A sensible starting point is stability around 50, similarity around 75 and style exaggeration at 0. Lower stability gives a wider emotional range but can turn erratic or rushed; higher stability is consistent but can drift into monotone. If you want a livelier read, lower stability a little and generate several times rather than pushing the slider to an extreme. Similarity set high can copy noise from a poor voice sample. Note that not every slider exists on every model: v4 keeps stability and similarity but drops style and speed.

Prepare the text

Write numbers, dates, prices and abbreviations out in words, for example "twenty twenty-six" or "three dollars fifty". Phone numbers, currencies, units and URLs are the usual failure points, especially on smaller models. For brand names or unusual words, v4 accepts IPA wrapped in slashes, while older models need phonetic respelling or a pronunciation dictionary.

Regenerate smartly

Output is non-deterministic, so the same settings give slightly different takes. On the website you get two free regenerations when the text, voice and model stay the same, within two hours and without refreshing the page; you may change sliders in between. Use those to audition the hook before generating the full script.

Export

Download MP3 or WAV straight after generating, or later from the history panel. The advanced menu adds higher-bitrate MP3, M4A and FLAC.