Pricing and buying
What "free" actually means on an AI voice generator
Every voice tool has a free tier and no two mean the same thing by it. Here are the four limits to look for, and which one is the trap.
6 min read ·
"Free" on a voice generator is doing one of three jobs: it is a demo, a trial, or a genuinely usable tier for small volumes. The word does not distinguish them, so you have to.
Limit 1: volume
The obvious one, and usually the least important. A free tier gives you some minutes or characters per month. Whether that number is generous matters less than whether it is enough to *evaluate* the tool, which takes only a few minutes of audio.
Ours is three minutes and 3,000 characters a month, which is honestly a trial rather than a working tier. We would rather say that than describe it as free forever and let you discover it in week two.
Limit 2: quality
This is the one to watch. Some free tiers run a smaller, older or faster model than the paid tiers, which means the audio you evaluate is not the audio you would buy. That makes the free tier useless for its main purpose.
Ask directly whether free-tier output comes from the same model as paid. If a vendor will not say, assume it does not.
Ours does not, and this is the place to say so plainly. Mayantar's free tier runs a smaller English-only model; the paid tiers route to different models per language — an Indian-language model for Hindi and the other Indian languages, and a multilingual model for the rest. So the free tier tells you whether generated narration suits your format. It does not tell you what Hindi or Spanish will sound like on a paid plan, because the free tier cannot generate them at all.
If that matters to your decision — and for anyone publishing in an Indian language it should — the honest advice is to judge us on the language samples on the language pages, which are generated by the models the paid tiers actually use.
Limit 3: rights
Most free tiers restrict commercial use, and most people do not read that before publishing a monetised video. This is the limit most likely to cause an actual problem, because it surfaces after you have built something on it.
It is also usually reasonable. The thing to avoid is not the restriction; it is discovering it in month three.
Limit 4: watermarking and attribution
Free output is often watermarked — sometimes audibly with a spoken tag, more often inaudibly. An audible tag makes free output unpublishable, which is the point of it. An inaudible watermark affects nothing you can hear and is worth keeping regardless of tier, for the provenance reasons that matter more every year.
Check which kind you are getting. They are described with the same word and are not remotely the same thing.
How to read a free tier in two minutes
- Is the model the same as paid? If unclear, assume not.
- Is commercial use allowed? If not, is that fine for what you are doing right now?
- Is the watermark audible or inaudible?
- Is there a card required? If yes, it is a trial, and set a calendar reminder.
- Is the volume enough to test properly — a few minutes — rather than enough to work?
A free tier that answers well on questions 1 to 4 and poorly on 5 is a good free tier. One that answers well on 5 and badly on 1 is a marketing number.