ElevenLabs TTS: the voice worth paying for, on the lines that need it
Text-to-speech stopped sounding robotic years ago. What still separates one voice from another is acting - whether a line lands with the stress you intended - and that is the only thing worth paying a premium per character for.
- Cost
- 14 PT
- Per
- ~100 chars
- Tier
- premium
- From
- $23/mo
ElevenLabs is the voice to reach for when delivery carries the line - the pause before a reveal, emphasis landing where you wrote it. It is also the priciest voice here, and it bills by the character rather than by the render, so a long script adds up fast. Draft cheaper, finish here.
What ElevenLabs TTS is best at
- Client-facing narration where a flat read would undercut the script - brand films, launch videos, anything with a producer listening.
- Lines that turn on emphasis: a beat before a reveal, a rising question, a name said with weight.
- Character dialogue in audio drama or game scenes, where two speakers have to sound like different people rather than one voice at two speeds.
- Final renders, after the script is locked and the timing has been proved out somewhere cheaper.
Where it falls down
- The highest per-character rate of any voice on the platform, with no volume discount - a script costs exactly its length, every time you re-render it.
- No voice cloning. Renders use a preset ElevenLabs voice, so there is no upload-a-sample step and nothing here holds a voice model of you.
- English-first: the render path uses ElevenLabs’ monolingual English model, so a non-English line comes back spoken with English phonetics.
- Delivery is not exposed as controls - you pick a voice and send text, with no stability or style sliders, so a read you dislike is a rewrite rather than an adjustment.
- Billing has a one-unit floor, so a three-word pickup costs the same as a full hundred-character line. Above that it bills by the characters you actually send.
- Premium voice carries its own monthly ceiling on top of the token budget. Cross it and renders continue on OpenAI TTS instead of failing, and the reply tells you it switched.
Prompts for ElevenLabs TTS
Copy one, swap the subject, keep the structure - the constraint clauses are what make these land.
Narration that breathes instead of racing
ElevenLabs TTSSo. The short version is that we were wrong about the timeline. Not a little wrong. Wrong in the direction that cost people a weekend. Here is what changed, and here is what happens if you already shipped.
Punctuation is the only pacing control plain text-to-speech gives you, so write the pauses as full stops: a one-word sentence buys a beat that no instruction about tone can, while the same words joined by commas run together.
Numbers and acronyms read the way you meant them
ElevenLabs TTSSet the oven to four hundred degrees. Bake for twenty-five minutes, then rest the tray for five. If the top has not browned by minute thirty, move it up one rack - not two.
Spell out digits, units and initialisms as the words you want heard, because every engine guesses at how to say a numeral, and a single misread figure in a delivered render costs the price of the whole take again.
A line that has to land on one word
ElevenLabs TTSI never said she took the money. I said she took the money. There is a difference, and the difference is the whole case.
You cannot mark stress directly, so build a sentence that only makes sense when the intended word is stressed - contrast in the surrounding text is what makes an expressive voice place emphasis instead of averaging it out.
Dialogue inside narration, without the whole paragraph over-acting
ElevenLabs TTSHe waited until the lift doors closed. "You told them," he said. Not a question. She watched the floor numbers count down and said nothing at all.
Keep quoted speech short and the tag plain - "he said" rather than "he snarled" - because an expressive voice reads emotion out of the tag and then colours the surrounding narration with it too.
Craft notes
- 01Prove the script somewhere cheaper. Draft on OpenAI TTS to catch wording and pronunciation problems, then spend the premium characters only on the take you are keeping.
- 02Batch pickups. The one-unit floor means four three-word retakes cost four units, while the same four lines sent in one pass bill as the characters they actually are.
- 03Proofread for homographs before you spend: read, lead, live, bow and wound are decided by context, and a wrong guess is only audible after you have paid.
- 04Rewriting beats re-rolling. With no style controls, the fastest way to change a delivery is to change the sentence - split it, shorten it, or move the contrast word later.
- 05Check what actually produced the file. Once the month’s premium-voice allowance is gone, renders keep working on the cheaper voice, and the difference is easy to miss on a laptop speaker.
When to use something else
ElevenLabs TTS is one of 3 voice models on the same Platform Token budget, so switching costs nothing but the tokens.
The read only has to be clear rather than good - internal walkthroughs, timing passes, draft cuts, anything nobody outside the team will hear.
You are generating speech at volume, where the per-character rate decides whether a feature is worth shipping at all rather than how it sounds.
Frequently asked questions
It is metered by input length in Platform Tokens at the highest voice rate here, and the cost is shown before the render runs. Plans including it start at $23/month.
ElevenLabs TTS, on the same bill as everything else
14 PT per ~100 chars, shown before you spend it. ElevenLabs TTS is on the $23/month plan and up.
Start free1 free prompt · no credit card · cancel any time