TTS API Free: What the $0.50 Trial Covers (Text Only)

The short answer

If your goal is a tts api free of charge that turns text into spoken audio, this is not that service. The endpoint at https://api.xtts.cc/v1 speaks the OpenAI chat format and returns text from one model, uncensored. It has no audio output of any kind.

What it does offer for free is a trial. After you sign up with Google or with an email and password, the API key page has a button that adds $0.50 of credit to your account. You can claim it once per person, it stays valid for 7 days, and it does not ask for a card. That credit pays for the text half of a speech pipeline: writing scripts, rewriting them for the ear, splitting them into lines per character, and checking them before you spend money on synthesis elsewhere.

The table below puts the facts in one place.

Item	Detail
Trial credit	$0.50, one click on the key page
Validity	7 days from the claim
Card required	No
Trial key limits	2 concurrent requests, 60 requests per minute
Output	Text only
Price after the trial	$0.15 per 1M input tokens, $0.50 per 1M output tokens

How far $0.50 goes

Tokens are the billing unit. Input tokens are what you send, output tokens are what the model writes back. Output costs more per token, so long generated scripts dominate the bill, while long prompts with short answers are cheap.

Two extremes show the range. If you used the whole trial on input alone, $0.50 / $0.15 per million would buy about 3.33 million input tokens. If you used it on output alone, $0.50 / $0.50 per million buys exactly 1 million output tokens. Real calls mix both, so the practical number sits between those two.

Three realistic jobs

Each job below is priced from the two rates. The formula is input tokens x $0.15 / 1,000,000 plus output tokens x $0.50 / 1,000,000.

Job	Input / output tokens	Cost per call	Calls in a $0.50 trial
Short line rewrite for speech	200 / 300	$0.00018	about 2,777
Summary read-out of a long document	3,000 / 1,500	$0.0012	about 416
Full narration script	800 / 4,000	$0.00212	about 235

Check the first row: 200 x 0.15 / 1,000,000 = $0.00003, and 300 x 0.50 / 1,000,000 = $0.00015, which adds up to $0.00018. In the last row, 800 input tokens cost $0.00012 and 4,000 output tokens cost $0.002, so one script costs $0.00212. Divide $0.50 by that and you get 235.8, so 235 whole scripts.

The token counts are estimates for illustration. A script of around 3,000 words is typically a few thousand tokens in English, but your exact figure depends on the text. Run one real request and read the usage field in the response to calibrate.

Claim the trial in four steps

1. Open the API key page and sign up with Google, or with an email and password.
2. Create your key and copy it somewhere safe. Treat it like a password.
3. Press the trial button on the same page. The $0.50 appears as balance.
4. Send a first request and confirm the balance moved.

Here is a request you can paste into a terminal. Replace the placeholder with your key:

Copycurl https://api.xtts.cc/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "uncensored",
    "messages": [
      {"role": "user", "content": "Rewrite for reading aloud: The Q3 report shows a 12% rise."}
    ]
  }' -i

The -i flag prints response headers. Look for X-Balance-USD, which reports what is left in your account, and X-Request-Id, which is handy if you ever contact support. The quickstart shows the same call in Python and Node.

Because the trial key is capped at 2 concurrent requests and 60 per minute, keep scripts small and sequential while you test. Hitting a limit returns HTTP 429 with the code rate_limited or concurrency, and a failed request is not charged.

Where the free credit is best spent

A trial is a test, so spend it on the questions you cannot answer by reading docs. Three are worth the money.

Does the model write in the voice you need?

Narration for a bedtime story, a noir audio drama and a product walkthrough read differently. Send the same brief with three system prompts and compare. Because the model is tuned to answer without content refusals for lawful adult use, fiction with dark themes is not cut off halfway, which is a frequent problem with stricter models. The hard line is any sexual content involving minors, which is never produced.

Does your cleanup prompt hold up?

Take twenty messy samples from your real data: abbreviations, figures, product names, markdown. Run them through a normalizing prompt and read the results aloud yourself. Fix the prompt until the failures stop, and only then start paying for synthesis elsewhere. This cheap loop saves expensive re-recordings later.

Does streaming fit your design?

Streaming over server-sent events lets your voice stage start on the first sentence. A few streamed calls are enough to confirm the pattern works with your engine. Remember that a streamed request holds its concurrency slot until the stream ends, which matters when the trial key allows only 2 at once.

Mistakes that waste trial credit

Seven days and fifty cents are easy to burn without learning anything. Four habits help.

• Do not paste your whole manuscript first. Output tokens cost the most. Test on one chapter, or even one page, until the prompt behaves.
• Set max_tokens while experimenting. It is optional, but a cap of a few hundred tokens stops a runaway answer from spending a few cents on text you will throw away.
• Keep the system prompt short. It is billed as input on every call. A 1,000-token system prompt sent 500 times is 500,000 input tokens, or $0.075. That is small, but it adds up when you run many experiments, and shorter prompts are easier to debug.
• Read the usage field. Every response reports prompt and completion tokens. Log them from your first call so your estimates come from measurements and not from guesses.

The trial window is a fixed 7 days from the claim, so claim it when you are ready to test, not the moment you sign up. If you want to prepare first, read the quickstart and write your test prompts ahead of time.

What it costs after the trial

There is no subscription. You top up prepaid credit with any whole amount from $20 to $500, and the credit never expires. Larger top-ups come with a bonus: +5% from $50, +10% from $100 and +15% from $200. A $100 top-up therefore gives $110 of credit, and a $200 top-up gives $230.

You can pay with crypto (USDT on TRC20 or USDC on Base). The pricing page has the current table.

To make that concrete, suppose you generate 100 full narration scripts per day for a month. That is 3,000 scripts at $0.00212 each, or $6.36. A $20 top-up would cover about 9,400 of those scripts. Your own volume will differ, but the scale is useful: text generation is a small line item next to most speech synthesis bills, so the sensible place to economize is usually the voice side.

Failed or refused requests are not charged, and prompts are not used for training.

Free options for the audio side

Since this service cannot give you free audio, here is how people commonly get it without a hosted bill. Some speech models are open source and can be run on your own computer or server, which costs hardware time instead of per-character fees. Operating systems also ship basic built-in voices. Hosted speech services from other companies sometimes include free tiers or monthly allowances.

All of these have limits and terms that change, including license conditions for commercial use and caps on usage, so read each provider's current pages before you build on one. This site gives no figures for them. A clean split works well in practice: use a free or self-hosted engine for the audio, and the pay-as-you-go text API here for scripts. If you are weighing both, start with the trial so the text side costs nothing while you test the audio side.

More guides

• TTS API Python: Prepare Text for Speech with the SDK
• TTS API Demo: Live Request and Response Examples
• Voice Cloning API: OpenAI Compatible TTS