ElevenLabs Sound Effects: 10 Credits a Second, Not 40

ElevenLabs Sound Effects turns a text prompt into a short sound. We made 25 sounds with a Free plan API key and read the credit counter after each call. A fixed-length clip cost 10 credits a second. An auto-length clip cost a flat 50 credits, even at 20 seconds. Both are a quarter of ElevenLabs' published prices, which likely assume the app's four takes per click.

Measured on 11 Nov 2026 between 10:42 and 11:04 Vietnam time, from a Windows 11 PC (WSL), with an ElevenLabs Free plan API key. ElevenLabs' sound effects docs, its API reference and its pricing page were read the same morning.

At a glance

Key results

• Fixed length: 10 credits per second. 1 second cost 10 credits, 5 seconds 50, 10 seconds 100. The docs say 40 credits per second.
• Auto length: 50 credits flat. With no duration_seconds, the model chose lengths from 1 to 20 seconds, and every clip cost 50 credits. The pricing page estimates 200 credits per generation.
• The Free plan's 10,000 credits buy about 200 auto clips, or 1,000 seconds at fixed length. Our whole test of 25 sounds used 1,130 credits.
• Fast. Most clips came back in 1.3 to 2.6 seconds, a 20-second clip in 3.6. One 5-second clip took 9 seconds.
• Write prompts in English. Two Vietnamese prompts came back almost silent. The same scenes in English came back as the sound we asked for.
• Free plan output is not for commercial use. ElevenLabs' help center says the Free plan "cannot be used for any commercial purpose", and shared clips must credit elevenlabs.io in the title.

Hands-on test

Setup

• Account: ElevenLabs Free plan, 10,000 credits a month, one API key.
• Endpoint: POST https://api.elevenlabs.io/v1/sound-generation, called from Python with httpx.
• Models: eleven_text_to_sound_v3 (SFX v3, the newer one) and eleven_text_to_sound_v2 (the default, which the docs say will be removed on 1 March 2027).
• Prompts: "Heavy wooden door creaking open", "Rain against a window in a quiet apartment", "Footsteps on gravel, then a metallic door opens", plus two scenes written in English and in Vietnamese.
• How we counted credits: we read GET /v1/user/subscription before and after each call. The counter lags by about 40 seconds, so we waited until it stopped moving. GET /v1/usage/character-stats split the first batch by model and agreed with the totals.
• How we judged sound: we measured length and loudness with ffprobe and ffmpeg. Gemini 4 Argon then listened to each file twice and scored the match to the prompt from 1 to 5. That is a machine judge, not human ears.

How to call the Sound Effects API

1. Create an API key in your ElevenLabs account. A free account is enough. RanThis is an independent affiliate of ElevenLabs and may earn a commission if you sign up or upgrade through this link. Our ElevenLabs API key guide shows where the key lives.
2. Send the prompt and the model:

curl -X POST "https://api.elevenlabs.io/v1/sound-generation?output_format=mp3_44100_128" \
  -H "xi-api-key: $ELEVENLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Rain against a window in a quiet apartment", "model_id": "eleven_text_to_sound_v3"}' \
  --output rain.mp3

3. Add "duration_seconds": 5 for an exact length, between 0.5 and 30. Leave it out to let the model choose.
4. Add "loop": true for background sounds you want to repeat.
5. Set model_id every time. Without it you get v2, which is on its way out.

What each call really cost

Three rules come out of this:

• The model does not change the price. v2 and v3 cost the same.
• Looping costs nothing extra.
• Auto length is the cheap option for anything over 5 seconds. A 20-second rain bed cost 50 credits on auto. Asking for 20 seconds would cost 200.

ElevenLabs' pages disagree with these numbers and with each other. The sound effects docs say "40 credits per second when duration is specified". The pricing FAQ lists "Sound Effects 200 credits per generation" as an approximate cost. The API pricing page says sound effects "are billed per generation". We were charged a quarter of the first two figures, and fixed lengths were billed by the second. The likely reason is on the Sound Effects product page: in the ElevenLabs app, each generation gives you "four samples". The API returns one sound per call, so 50 credits is a quarter of the app's 200. The per-second figure is off by the same factor of four, but no page says why. Prices can change without notice, so read your own counter after a few calls.

To turn credits into dollars: Starter is 6 USD a month for 30,000 credits, so one credit is 0.02 cents. At our rates an auto clip costs about 1 cent and one fixed second about 0.2 cents.

Prompts, quality and limits

• English prompts worked, Vietnamese did not. "Heavy rain on a corrugated metal roof" gave 20 seconds of steady rain, average loudness −21.7 dB. "Mưa rào trên mái tôn", the same scene in Vietnamese, gave 15 seconds that were almost silent, −51.5 dB on average. Gemini heard birds and then nothing. The motorbike scene went the same way. We only tested Vietnamese, but write your prompts in English for any language.
• Some prompts need several takes. The rain prompts scored 3 or 4 out of 5 in every pass. The door prompt was unreliable: across seven clips the judge heard a rattle, metal clicks and once a dog. Budget for two or three takes. At 10 credits a second that is cheap.
• v2 hit full scale. Both v2 clips peaked at 0.0 and −0.2 dB, the ceiling, so they may distort. The v3 clips peaked between −1.1 and −28.6 dB. The docs say v3 is "less likely to clip on high-frequency sounds", and our numbers fit that.
• The length limits are 0.5 to 30 seconds. The API refused 0.3 and 31 with a clear error and charged nothing. The capabilities page says the range starts at 0.1 seconds. The API reference and the API itself say 0.5.
• Formats: WAV and 192 kbps came back on Free. The API reference says 192 kbps MP3 needs Creator or above. Our Free key received both a 192 kbps MP3 and a 48 kHz WAV at the normal 10 credits.
• Four at once is fine. Four 1-second requests sent at the same time all returned 200. The models page lists a sound effects concurrency limit of 4 for the Free plan.

What worked

What worked

• Speed. Two seconds for a 10-second clip is fast enough to generate sounds on demand inside an app.
• Auto length. A flat 50 credits for up to 20 seconds of ambience is the best deal we found in the API.
• Clean errors. Bad lengths were rejected before any charge, with a message that names the valid range.
• The usage endpoint. GET /v1/usage/character-stats with breakdown_type=model shows sound effects spend by model, by the hour.

Where it fell short

What didn't work

• The docs and pricing pages. Three pages give three different prices, and none matched what we were charged.
• Vietnamese prompts, as above. The API still charged 50 credits for each near-silent clip.
• The credit counter lags. /v1/user/subscription took about 40 seconds to show a charge, and reading it too often returned 429 Too Many Requests.
• Specific objects. "A door creaking open" came back as many different sounds. Scenes like rain were more reliable than single objects.

Our verdict

Use ElevenLabs Sound Effects through the API if you need short sounds on demand for videos, games or apps. It is fast and costs less than ElevenLabs says: 10 credits a second at fixed length, 50 credits for any auto clip. Leave duration_seconds out for ambience, set it for short one-shots, write prompts in English, use v3, and plan for a couple of takes per sound. Prototype on the Free plan, but move to a paid plan before you publish anything that earns money, because Free output has no commercial license. The partner link in step 1 above opens a free account.

FAQ

Is the ElevenLabs sound effects generator free?

Yes, within the Free plan's 10,000 monthly credits. In our API test that was enough for about 200 auto-length clips or 1,000 seconds at fixed length. In the web app the generator makes four sound effects per click, according to ElevenLabs' sound effects guide; we did not measure what the app charges. The Free plan does not include a commercial license.

How many credits does a sound effect cost?

In our test on 11 Nov 2026, 10 credits per second when you set the length, and 50 credits per clip when the model picks the length. ElevenLabs' docs say 40 credits per second, and its pricing FAQ estimates 200 per generation.

v2 or v3: which ElevenLabs sound effects model should I use?

Use v3. In our test both cost the same: 50 credits for an auto-length clip and 50 for 5 seconds. Both v2 clips peaked at 0.0 and −0.2 dB, the ceiling, while the v3 clips peaked between −1.1 and −28.6 dB. On the door prompt at auto length, the judge scored v2 2 out of 5 in both passes and v3 3 out of 5. v2 is still the default when you leave out model_id, and ElevenLabs' models page says it will be removed on 1 March 2027.

Can I use ElevenLabs sound effects commercially?

On a paid plan, yes. ElevenLabs' help center says the Free plan "cannot be used for any commercial purpose" and that published Free plan content must credit "elevenlabs.io" or "11.ai" in the title. Its sound effects page suggests that Free sounds can be used commercially with attribution. When two official pages disagree, follow the stricter one.

How long can an ElevenLabs sound effect be?

From 0.5 to 30 seconds per generation. For longer backgrounds, set loop to true; ElevenLabs says the clip then loops smoothly.

How do I download ElevenLabs sound effects?

With the API, the response is the file. The API reference describes the reply as "The generated sound effect audio, encoded in the specified output_format", so save it straight to disk, as --output rain.mp3 does in the curl command above. We got MP3 and WAV files this way. In the web app, ElevenLabs' guide says to open the History tab, click the download icon and choose MP3 (44.1kHz) or WAV (48kHz). We did not test the app.

Claude, the AI assistant, ran the API calls, measured the files and wrote this page. Credits come from ElevenLabs' own counter and usage endpoint. Quality scores come from Gemini 4 Argon listening to the files, not from human listeners. Each ElevenLabs claim was checked against the pages linked above. For text to speech prices see our ElevenLabs pricing test.