How to use Stable Audio

OVERVIEW

A tape machine, not a render button

Most music generators hand you a finished file. If one part is wrong, your only option is to try the whole thing again and hope. Stable Audio works the way a studio does: the music lands on a deck and stays there, so you can layer parts onto it, re-record what is wrong, cut it up and mix it — instead of starting over.

A session is one piece of music. It holds up to 4 active tracks — drums, bass, keys, whatever the music called for — along with everything you do to them. Sessions save as you work, so you can leave and come back.

GETTING GOING

Starting a session

Type what you want to hear and press START. Be specific about the things that matter to you — genre, instruments, tempo, mood, era. “Slow dub techno, deep sub bass, tape hiss, 118 BPM” will get you closer than “electronic music”.

Two settings sit under the prompt.

FULL MIX: Your prompt goes to the model exactly as you wrote it, and comes back as one finished mixed track. Nothing rewrites or expands what you typed. This is the mode for direct control — what you ask for is what gets recorded. Start here.

MULTI-TRACK: Your prompt goes to an agent first, which works out the arrangement — which parts the music needs and how they fit together — and then records each part onto its own track. You get separate tracks to mix, mute, solo and re-record individually. The trade is that the agent makes musical decisions you did not ask for, so the result can drift from a literal reading of your prompt. Experimental.

The length picker sets how long the piece runs: AUTO lets the model decide, or choose 0:30 for a loop, 1:00 for a short, 1:30 for a song, or 3:00 for an extended piece. Longer pieces take longer to generate.

You can also match an existing track’s tempo and key. Attach a file under TEMPO + KEY and the arrangement is locked to the tempo and key detected in it. Works with MULTI-TRACK only. The audio itself is not used in the generation — nothing you attach is heard in the result, so it will not sound like your file, it will share its tempo and key.

THE MIX

The deck

When the session lands, every track gets a row: a waveform, a volume fader, pan, and M / S for mute and solo. Solo lights blue, so you can tell at a glance which track you have in isolation. A FULL MIX starts as a single row; MULTI-TRACK starts with one row per part. Either way you can add more.

Press play to hear the mix. Faders, pan, mute and solo are free and instant — nothing there generates new audio, so move them as much as you like. These four record new audio, and cost credits:

ADD TRACK: Records one more part into the session — a counter-melody, a percussion layer. It listens to what is already there and plays along with it.

REGENERATE: Re-records a single track from scratch, leaving every other track untouched. Use it when one part is wrong and the rest is right.

REPLACE SECTION: Select a time range on a track and re-record just that stretch. Good for fixing a bad four bars without losing a take you like.

EXTEND: Continues a track past its current end, keeping the same feel.

A session holds 4 active tracks. When all 4 are full and you want another, there are two ways to make room. Erase a track you do not want — X on its channel strip, twice to confirm — and it moves to a strip under the deck you can restore it from. Or consolidate two or more into one lane — see below — which frees the slots without losing the sound.

CHAT

The producer

The panel beside the deck is a producer who can hear your session. Ask for what you want in plain language — “the drums are too busy”, “add something warmer underneath”, “make the second half darker” — and it will either answer, or go and do the work.

When it decides an action is needed it starts it for you: adding a track, regenerating one, or composing a new arrangement. The operation appears on the deck exactly as it would if you had pressed the button yourself, and costs the same credits. You can also just ask questions — not every message triggers a generation.

TAPE

Tape edits

Tape edits rearrange audio you already have rather than generating anything new. They are free and they apply straight away.

REVERSE: Plays a section backwards.

STUTTER: Repeats a short slice in place — the classic edit-point retrigger.

FREEZE: Holds a moment, sustaining it across the section.

Each edit applies either to TRACK — one track only — or ALL, which hits every track at once so the whole arrangement turns together. Whole-mix edits are how you get transitions and drops.

Underneath those there is a full splice surface: CUT, COPY and PASTEworking through a reel, with the paste landing as INSERT, OVERWRITE or REPLACE. It is closer to editing tape than to a prompt, and it is the right tool when you know exactly which bar is wrong. Everything there is free too, and the panel has a help toggle that labels each control.

PROCESSING

Effects

Every track has its own effects chain, and the master bus has one of its own for the finished mix. Effects are applied live during playback, so you hear changes as you make them, and they cost nothing.

They are also non-destructive: the original recording sits untouched underneath, and removing an effect returns the track to where it was. Effects are printed into the audio when you bounce or export.

FINISHING

Bouncing and export

Consolidating commits several tracks down to one — the old tape trick of freeing up heads by mixing finished parts together. It prints levels, pan and effects exactly as you have them. This is how you get past the 4-track limit with everything still in the mix: consolidate what is done, keep working in the space that opens up.

You get one hour to change your mind. UNDO BOUNCE puts the original tracks back, and the window is an hour from the bounce. It is a single step, not a history — and it goes away if you switch to another take or reload the page, so if you want it, take it while you are still looking at the result. After that the bounce stands.

When the piece is finished you can export the mix, or share a link to the session so someone else can listen

BILLING

What costs credits

Credits are spent only when the model records new audio. Everything you do to audio you already have — mixing, editing, effects, bouncing, exporting — is free.

Starting a session in MULTI-TRACK is one flat charge however many tracks come back, so a four-track arrangement costs less than building the same thing a track at a time. Free, and unlimited:

• Playback, looping and scrubbing
• Volume, pan, mute and solo
• Effects — adding, tweaking, removing
• Tape edits — reverse, stutter, freeze
• Bouncing tracks down
• Talking to the producer
• Exporting and sharing

Your balance is shown in the studio. A generation that fails is not charged. Credits are shared across Stable Audio — the studio and the plugin draw on the same balance.

HELP

If something goes wrong

Generation failed: Failed generations are not charged. Try again; if one particular prompt keeps failing, rephrase it.

Out of credits: Anything that records new audio is unavailable until your balance tops up. Playback, mixing, editing and export keep working, so you can still finish what you have.

A track sounds wrong: Regenerate that one track rather than starting over — the rest of the session is preserved.

Nothing is playing: Check the track is not muted, and that no other track is soloed. A lit S anywhere on the deck silences everything else.

No free slots: Bounce two finished tracks together to open one. See Bouncing and export.