1
ElevenLabs Sound Effects
ElevenLabs · Fast text-to-SFX, loops and ambiences with clear commercial terms
$$
ElevenLabs is the default text-to-SFX tool for creators: up to 30-second clips, seamless loops and 48 kHz WAV for non-looping effects, with royalty-free commercial use on any paid plan. SFX v3, which follows multi-part prompts more closely, is in limited access. It works from text, not video, so sync to picture is manual, and with no public SFX arena this ranking rests on capability and adoption.
- Cheapest way in
- Free plan (personal use); Starter $6/month for commercial use
- Cost per output
- API $0.12 per minute of sound effects
- Video to audio
- No
- Max length
- 30s
- Loops
- Yes
- Licence
- Royalty-free commercial use on paid plans
- Open weights
- No
- Input
- Text
Good at
- Clear royalty-free commercial licence on paid plans
- Seamless looping
- Up to 30s per generation, 48 kHz WAV
- Same subscription as voice and music
Watch out for
- No native video input for automatic sync
- Free plan is personal use only
- ElevenLabs quotes two credit costs (40 per second or 200 per generation)
Prices elevenlabs.io, checked 1 November 2026
ElevenLabs Sound Effects website Work made with it on AI LIVE
2
Mirelo SFX 2.0
Mirelo AI · Scoring finished AI video with synced foley while keeping dialogue
$$
Mirelo SFX 2.0 (14 October 2026) is the only commercial model on the FoleyBench video-to-audio leaderboard, where it posts the best sync (DS 0.403), CLAP and KLD scores of the models listed. It builds effects from on-screen action, keeps spoken dialogue and strips unprompted music that video models add, and Mirelo says it handles up to 10 minutes per run.
- Cheapest way in
- Free plan with watermark; Creator from €20/month (about 40 minutes)
- Cost per output
- 10 credits per second until 12 Dec 2026, then 15; extra credits €0.75-1.20 per 1,000
- Video to audio
- Yes
- Max length
- Up to 10 minutes per run (vendor claim)
- Licence
- Not stated on pricing pages
- Open weights
- No
- Input
- Video (plus optional text)
- Keeps dialogue
- Yes
Good at
- Best audio-video sync score on FoleyBench
- Keeps dialogue, removes unwanted generated music
- Long inputs (up to 10 minutes per run, vendor claim)
- API also on fal and Replicate
Watch out for
- Prices in EUR credits only; no published dollar API rate for 2.0
- Free plan exports are watermarked
- Introductory rate ends 12 December 2026 (10 to 15 credits per second)
Benchmarks FoleyBench (main, 5,000 clips): Best DS 0.403, CLAP 0.349 and KLD 2.24 of listed models; MMAudio leads FAD and IS
Prices mirelo.ai, checked 1 November 2026
Mirelo SFX 2.0 website Work made with it on AI LIVE
3
Adobe Firefly Sound Effects
Adobe · Timing effects with your own voice inside an Adobe workflow
$$
Generally available since August 2026, Generate Sound Effects runs on Adobe's Firefly Audio Model and lets you perform a sound with your voice to set its timing and energy, then returns four variations of up to 30 seconds. It sits inside the Firefly Video Editor and Creative Cloud, and Adobe markets the output as commercially safe. No independent benchmark covers it.
- Cheapest way in
- Free daily generations; Firefly Standard €11.35/month incl. VAT
- Cost per output
- About 10 credits per effect (2,000 credits cover up to 200 effects)
- Video to audio
- Partial (place against video in editor)
- Max length
- 30s per variation
- Licence
- Marketed as commercially safe
- Open weights
- No
- Input
- Text, voice timing guide; video in Firefly editor
- Variations
- 4
Good at
- Voice performance sets timing and energy
- Four variations per prompt
- Adobe's commercially safe positioning
- Bundled with Firefly image, video and music
Watch out for
- Commercial-safety claim is Adobe's own
- Uses premium generative credits
- No public benchmark results
Prices adobe.com, checked 1 November 2026
Adobe Firefly Sound Effects website
4
Kling Video to Audio
Kuaishou · Adding ambience and a music bed to silent Kling clips
$
Kling's Video to Audio tool, built on the Kling-Foley research model, takes a silent clip and generates matched sound effects, with separate prompts for effects and background music plus an ASMR mode on API hosts. Kling's own guidance says it suits atmosphere such as rain, fire or room tone rather than dialogue. It is cheap on third-party APIs, but no independent benchmark includes it.
- Cheapest way in
- Kling free plan; Standard from $6.99 for the first month
- Cost per output
- About $0.035 per run via third-party API hosts
- Video to audio
- Yes
- Max length
- Matches the input video (reportedly 3-20s clips)
- Open weights
- No
- Input
- Silent video, SFX prompt, optional BGM prompt
Good at
- Separate SFX and music prompts
- Low per-run API price via hosts
- Native audio also available inside Kling video generation
Watch out for
- Not for dialogue
- Output is unmastered
- Official credit costs not stated on the feature page
Prices wavespeed.ai, checked 1 November 2026
Kling Video to Audio website
5
Stable Audio 3
Stability AI · Licensed-data SFX you can run locally or call cheaply by API
$
Stable Audio 3 (20 May 2026) includes a Small SFX model built to run on phones and laptops, with open weights, and all models are trained on fully licensed data. Under the Stability AI Community Licence you own outputs and can commercialise them if your organisation earns under $1M a year. It is text-to-audio only, with inpainting and continuation for edits.
- Cheapest way in
- Free open weights; 25 free API credits for new accounts
- Cost per output
- $0.08 per Stable Audio 3.0 API call (8 credits at $0.01)
- Video to audio
- No
- Max length
- Small up to 2 minutes
- Licence
- Stability AI Community Licence (free commercial use under $1M revenue)
- Open weights
- Small SFX, Small, Medium yes; Large API only
- Input
- Text, audio for inpainting
Good at
- Trained on licensed data
- Open weights for Small SFX, Small and Medium
- Cheap API at $0.08 per call
- Audio inpainting and continuation
Watch out for
- No video input
- Enterprise licence needed above $1M revenue
- Reviewers note weaker sharp transients than ambient textures
Prices docs.stability.ai, checked 1 November 2026
Stable Audio 3 website
6
MMAudio
Ho Kei Cheng et al. · Free, high-quality video-to-audio for personal or test projects
$
MMAudio's text-conditioned variant has the best FAD (8.76) and IS (11.2) on the FoleyBench main leaderboard, ahead of every other open model listed. It is free to run and widely used in ComfyUI workflows. Model weights are CC BY-NC 4.0, so it is not suitable for paid client work.
- Cheapest way in
- Free (self-hosted)
- Cost per output
- No official hosted price; third-party hosts vary
- Video to audio
- Yes
- Max length
- 8s default
- Licence
- Code MIT; weights CC BY-NC 4.0
- Open weights
- Yes
- Input
- Video and/or text
Good at
- Best audio quality scores on FoleyBench
- Free and open
- Strong community tooling
Watch out for
- Non-commercial weights licence
- Default 8-second output; quality drops far from that length
- No updates since March 2025
Benchmarks FoleyBench (main, 5,000 clips): Best FAD 8.76 and IS 11.2 of listed models
Prices github.com, checked 1 November 2026
MMAudio website
7
HunyuanVideo-Foley
Tencent Hunyuan · Self-hosted video-to-audio outside the UK and EU
$
Open-sourced in August 2025 with an XL model added in September 2025 that runs in 8-16 GB of VRAM, HunyuanVideo-Foley is a capable free video-to-audio model. Its Tencent Hunyuan Community Licence explicitly does not apply in the EU, UK or South Korea, so UK creators cannot use it under licence. It is not on the FoleyBench leaderboard.
- Cheapest way in
- Free (self-hosted)
- Cost per output
- No official hosted price
- Video to audio
- Yes
- Licence
- Tencent Hunyuan Community Licence; not valid in EU, UK, South Korea
- Open weights
- Yes
- Input
- Video plus text
- VRAM
- XXL 20GB (12GB offload); XL 16GB (8GB offload)
Good at
- Free open weights
- Runs on consumer GPUs with offload
- Royalty-free commercial use where the licence applies
Watch out for
- Licence excludes the EU, UK and South Korea
- No updates reported in 2026
- No independent leaderboard result
Prices github.com, checked 1 November 2026
HunyuanVideo-Foley website
8
Sony Woosh
Sony AI · Experimenting with an open text- and video-to-SFX model
$
Sony AI released Woosh in May 2026 with open weights for both text-to-audio and video-to-audio, and says the video model beats a comparable baseline on FoleyBench with fewer parameters. The public release is for research and experimentation only, so it cannot be used commercially. A private version trained on licensed pro libraries is not public.
- Cheapest way in
- Free (self-hosted, non-commercial)
- Cost per output
- No hosted price
- Video to audio
- Yes
- Licence
- Research and non-commercial only
- Open weights
- Yes
- Input
- Text, or video plus optional text
Good at
- Open weights and code
- Both text and video conditioning
- ComfyUI workflows available
Watch out for
- Non-commercial
- No specific benchmark scores published in the announcement
- Better private model not available
Prices ai.sony, checked 1 November 2026
Sony Woosh website
