Stable Audio 3 AI Music Generator (Stable Audio 3 Medium)
Stability AI's instrumental music model, cut to the exact second, up to six minutes.
Stable Audio 3 is the music model from Stability AI, the company behind Stable Diffusion, reached here through fal.ai. It makes instrumental music only, and its defining feature is length: you set the duration in seconds and the track is made to that length, from a 3-second sting up to 6 minutes.
That makes it the model for music that has to fit something else. A bed under a 27-second product video, a 90-second loop for a hold line, a six-minute background for a webinar or a shop floor: set the seconds to the edit and the track ends with it, with no fading out over a hard cut.
Stability says the model is trained on licensed music, which is part of why commercial teams reach for it. It does not sing and does not take lyrics, and on this platform it is told to keep vocals out, so anything with a voice in it belongs on Lyria. For a sound you already like, read a reference track with Audio to Prompt and give this model its description, with the tempo and key it measured.
What Stable Audio 3 is best at
Beds cut to the exact length of an edit
Long background music, up to six minutes
Sonic logos and short stings
Instrumentals from a model trained on licensed music
Exact length Set the seconds and the track is made to that length, so it ends on the cut instead of being faded or chopped.
Long-form pieces Up to six minutes in one take, long enough for a webinar, a store, a livestream or a hold line without a loop seam every minute.
Licensed training data Stability says Stable Audio 3 is trained on licensed music, which matters to brands that have to answer for where their assets came from.
One price whatever the length Priced per take, not per second, so a six-minute piece costs the same as a ten-second sting.
Where Stable Audio 3 falls short
Instrumental only. It cannot sing or take lyrics, so jingles and anything with a voice belong on Lyria 3.5.
It is built for beds and background. When a sung hook has to do the selling, Lyria does that job.
Prompts should describe the sound, not name an artist: music here is made from a description of its qualities. Read a reference with Audio to Prompt to get that description.
Cleared for commercial use online and offline, in ads, videos, podcasts and social posts. It is not cleared for film, TV, radio broadcast or games distribution.
Stable Audio 3 specifications
Provider: Stability AI
Model ID: stable-audio-3/medium
Length: 3 seconds to 6 minutes, to the second
Vocals: No, instrumental only
Output: 44.1 kHz stereo MP3
Best at: Beds cut to an exact length
Provider model ID: fal-ai/stable-audio-3/medium/text-to-audio
Routed via: fal.ai
What you can do with Stable Audio 3
Bed cut to your edit Set the exact length of your video and the music ends with it, mixed to sit under a voiceover.
Long background Up to six minutes for a webinar, a store, a livestream or a hold line, with no loop seam every minute.
Sonic logo A three-to-five-second signature to end every video on: a few notes, polished and ownable.
Make it with AI Music Generator, which runs Stable Audio 3 and shows the price before you spend anything. To start from a track, a voiceover or a sound that already works, read it with Audio to Prompt first: it describes the sound without naming anyone and hands that description to the generator.
Prompts that work on Stable Audio 3
Set the length, then describe the ending
A 27-second bed for a product video: steady pulse, warm synth chords, no lead melody, ending on a clean final chord at the last second.
The length is exact, so tell the model how to end. A named final chord or hit lands on your cut instead of trailing off.
Instruments, tempo, key
Light funk instrumental, 104 BPM, in E minor: tight drums, slap bass, clean rhythm guitar, muted brass stabs.
Concrete instruments and a tempo give this model the most to work with. It is an instrumental model, so spend the words on the band, not on vocals.
Carry a reference's measurements
Read a track you like in Audio to Prompt, then use its description with the tempo and key it measured.
Audio to Prompt measures tempo and key from the audio itself, which is exactly the detail an exact-length instrumental model can use.
What Stable Audio 3 costs
Cost on Prompt Reverse, per take: 5 AI Credits, two takes in one run: 10 AI Credits. AI Credits come from a one-time credit pack and never expire, with no subscription and no separate provider account.
Stable Audio 3 vs the other models
Lyria 3.5: Switch for vocals, your own lyrics or a full song with a chorus.
Lyria 3 Clip: Switch for a quick 30-second piece with a sung hook.
ElevenLabs Sound Effects: Switch for a single sound: a whoosh, a chime or an impact rather than music.
Stable Audio 3 is Stability AI's music generation model. On Prompt Reverse it makes instrumental music to an exact length, from 3 seconds to 6 minutes, reached through fal.ai.
Can Stable Audio 3 make music to an exact length?
Yes, and that is its main strength. Set the seconds and the track is made to that length, which is why it is the model for beds under a video of a set duration.
Does Stable Audio 3 do vocals?
No. It is instrumental only. For singing or your own lyrics, use Lyria 3.5 or Lyria 3 Clip.
Is Stable Audio 3 trained on licensed music?
Stability AI says it is. Music you make with it can be used commercially, online and offline, and every file is tagged as AI-generated in its metadata. It is not cleared for film, TV, radio broadcast or games distribution.
Does a longer track cost more?
No. It is priced per take whatever the length, so a six-minute piece costs the same as a short one. The price is shown before you run it.