Back to Blog
AI Music

AI Sound Effects Generator: Tools, Limits and Prompts

An AI sound effects generator turns text into SFX. Compare ElevenLabs, Stable Audio 3, Adobe Firefly and open models on length, licence and prompts.

8 min read
By OndaLabs Team

An AI sound effects generator creates short sounds, such as a door slam or rain, from a text description. In 2026 the main options are ElevenLabs Sound Effects, Stable Audio 3 Small SFX and Adobe Firefly. Their lengths and licences differ, so check each one before you use a sound in a paid project.

This guide compares the tools, explains their limits and gives prompt patterns that work across them. Facts are as of October 2026.

What an AI sound effects generator makes

Most tools take a short written description and return a clip of one to a few seconds up to around 30 seconds. People use them for:

  • One-off hits, such as an impact, a whoosh or a notification sound.
  • Ambience, such as wind, a crowd or a city street.
  • Loops for games, video or podcast beds.
  • Foley-style details, such as footsteps or cloth movement.

Sound effects are not the same as music. For a full track, use a music tool; see our AI music generators guide.

AI sound effects generators at a glance

| Tool | Maximum length | Commercial rule (as stated) | Where it runs | |---|---|---|---| | ElevenLabs Sound Effects | Up to 30 seconds per generation | Free plan is personal use only; paid accounts can use effects royalty-free in commercial projects | ElevenLabs app and API | | Stable Audio 3 Small SFX | Not listed on the model card | Community License lets you own and commercialise outputs; organisations over $1 million in annual revenue need the Enterprise License | On your own device, with open weights | | Adobe Firefly Generate Sound Effects | Not stated in the announcement | Adobe calls the audio "commercially safe"; the announcement gives no indemnification details | Adobe Firefly, generally available since 2026-08-20 | | MMAudio (research model) | 8 seconds by default | Checkpoints are CC-BY-NC 4.0; the README says they are not guaranteed suitable for commercial use | Your own GPU | | AudioGen (Meta AudioCraft) | Not stated in the sources we read | Model weights are CC-BY-NC 4.0, which is non-commercial | Your own hardware |

ElevenLabs Sound Effects: up to 30 seconds

ElevenLabs' sound effects docs say each generation can run up to 30 seconds. You can set the duration. If you do not, the length is estimated from your prompt. The API reference adds that a set duration costs 40 credits per second. The pricing page says a typical generation costs about 200 credits, and the help centre notes that cost depends on your settings and on whether you use the website or the API. It is not influenced by the text you enter.

Two features matter for production:

  • Looping: The docs say effects can be played on repeat without perceptible start or end points. Looping is supported only by the eleven_text_to_sound_v2 model. The docs suggest generating 30 seconds of soft rain and looping it.
  • Formats: Output is MP3 for all effects, and WAV at 48 kHz for non-looping effects. A 192 kbps MP3 needs the Creator tier or above, and 44.1 kHz PCM needs Pro or above.

The licence is simple to read but still needs checking. ElevenLabs says that as long as you have a paid account, you can use its sound effects royalty-free in commercial projects. The Free plan gives 10,000 credits a month for personal use only.

Stable Audio 3 Small SFX: runs on your own device

Stability AI's Stable Audio 3.0, released 2026-05-20, includes a model called Small SFX. Stability says it "generates sound effects on-device, such as on phones and consumer-grade laptops." Its weights are open on Hugging Face, alongside Small and Medium.

The Hugging Face model card describes:

  • Text-to-audio output with variable length.
  • Inpainting for targeted edits.
  • English-language prompts.
  • 0.6 billion parameters, and generation in a few seconds on a MacBook Pro M4, according to the card.
  • A Stability AI Community License with Gemma Terms of Use components.

The card points to Stability's licence page for commercial use. Read that page before a paid project, because the Community License has an annual-revenue threshold for organisations.

Adobe Firefly Generate Sound Effects

Adobe's Generate Sound Effects uses the Firefly Audio Model to create sounds that "match the action, timing and energy of your content." Adobe made it broadly available on 2026-08-20, alongside Generate Music and Generate Speech. Adobe calls the audio "commercially safe." That is Adobe's own claim. Its announcement gives no clip length, pricing or indemnification details, so do not assume them.

Open research models and their licence limits

Two research models are useful to know, and both have licence limits:

  • MMAudio is a video-to-audio model from CVPR 2025. It can also make audio from text when no video is given. Its default output and training length is 8 seconds. The code is MIT-licensed, but the checkpoints are CC-BY-NC 4.0, and the README says the models are not guaranteed suitable for commercial use. The default model needs about 6 GB of GPU memory at 16-bit.
  • AudioGen, from Meta's AudioCraft repository, is described as a state-of-the-art text-to-sound model. Its code is MIT-licensed, but its model weights are CC-BY-NC 4.0.

The split between code and weight licences is the trap. Check the weights before a commercial project.

How to write sound effect prompts

ElevenLabs' docs give practical advice that works with most text-to-sound tools:

  1. Keep it concise. For example: "Thunder rumbling in the distance."
  2. Describe sequences in order. For example: "Sword being drawn, then clashing with another blade."
  3. Use audio terms. The docs list words such as impact, whoosh, ambience, loop, drone and glitch.
  4. Ask for loops explicitly. Add "loop" and a duration, and then test the seam.
  5. Use a music tool for full tracks. The docs say full music production should use the Music API. For a short beat loop, they show an example such as "90s hip-hop drum loop, 90 BPM."

| Goal | Pattern | Example from ElevenLabs docs | |---|---|---| | One event | Subject and action | Thunder rumbling in the distance | | A sequence | Event, then event | Sword being drawn, then clashing with another blade | | A loop | Sound, duration, loop | Soft rain for 30 seconds, looped | | A rhythm | Genre, tempo | 90s hip-hop drum loop, 90 BPM |

Sound effects, music and voice: which tool for which job

  • Effects: A dedicated sound effects tool, such as ElevenLabs, Stable Audio 3 Small SFX or Adobe Firefly.
  • Music: A music generator, such as those in our AI music generators guide.
  • Voice: A text-to-speech tool. Narration and dialogue are different from effects, and they need voice models rather than sound models.

Some tools cover more than one job, but the limits differ. Check the length and licence for each job separately.

Sound effects and voice in OndaLabs

OndaLabs is an AI voice studio, so it does not generate sound effects. Its tools are text to speech, two-speaker dialogue, speech to text, custom voices, image generation and music generation. Use a dedicated effects tool for effects, and OndaLabs for voice and music.

Text to speech includes 30 prebuilt voices and expressive tags. Dialogue is available from the Free plan up. Music generation is on the Creator plan ($22 a month, 400,000 credits) and above, at 2,500 credits per clip and 5,000 per song. Commercial use is included on Starter and above. You can try text to speech, build a dialogue, make music or compare plans on the pricing page.

Frequently asked questions

What is the best AI sound effects generator?

We do not rank tools on sound quality, because our sources include no controlled test. Choose by the things you can check: maximum length, licence, and whether the model runs on your device or in the cloud. For many projects, the licence is the deciding factor.

Is there a free AI sound effects generator?

ElevenLabs has a Free plan with 10,000 credits a month, but it is for personal use only. Stable Audio 3 Small SFX has open weights you can download, subject to the Community License. MMAudio and AudioGen weights are non-commercial, so they are not suitable for paid work.

How long can AI-generated sound effects be?

ElevenLabs generates up to 30 seconds per effect, and MMAudio's default length is 8 seconds. Stable Audio's model card does not list a maximum for Small SFX. For longer scenes, generate several clips or use a loop with a seam check.

Can I use AI sound effects commercially?

It depends on the tool and plan. ElevenLabs says paid accounts can use effects royalty-free in commercial projects. Stable Audio's Community License allows commercial use with a revenue threshold for organisations. Research model weights from MMAudio and AudioGen are non-commercial. Read each licence before publishing, and treat this as general information only.

Can AI make looping ambience?

Yes, with ElevenLabs. Its docs say looping is supported on the eleven_text_to_sound_v2 model and show a 30-second soft rain example looped. Test the loop point in your editor before you use it, because a seam can be audible.