Stable Audio Plugin: Generation on a Live Instrument Track
Stability AI's Stable Audio plugin puts 3.0 generation on a Mac AU/VST3 instrument track. Early beta: six-minute takes, BPM sync, a rebuilt web mixer.
Stability AI just put Stable Audio 3.0 on an instrument track. The plugin generates audio inside the session, at the project tempo, then leaves a clip you can arrange like any other audio. The same day, StableAudio.com became a session: prompt a take, give it notes, split it into tracks, keep working.
Both are early beta. Some features are experimental. This is a first look from official Stability AI pages, not a session report. We have not run the installer.
May's Stable Audio 3 review covered open weights, ComfyUI, and a six-minute instrumental model. This is the product layer: a Mac AU/VST3 plus a browser mixer.
Stable Audio plugin at a glance
- What it is: a Stable Audio 3.0 generator that loads as an instrument
- Formats: macOS AU and VST3, Apple Silicon and Intel
- Windows: Coming Soon on the official download page
- Named DAWs: Logic Pro and Ableton Live. The product page also lists Pro Tools and "most major DAWs"
- Load: instrument slot, not an audio insert. Ableton: PlugIns on an Instrument track. Logic: Instrument slot
- Length: a short section through a full six-minute song
- Tempo: syncs to session BPM
- Takes: multiple generations in a playlist
- Models: hosted Large, plus optional on-device Small and Medium music/SFX packs
- Web companion: iterative direction, audio-to-audio, multi-track mix, bounce and stem export
- Status: early beta. No hands-on yet
- Price: unpublished on the announcement and the plugin page
- Rights: commercially safe models; you own outputs, per Stability AI
Download: stableaudio.com/plugin. Announcement: Sharing a new way to work with Stable Audio (August 18, 2026).
What it is
Two surfaces. One model family.
The plugin is the session tool. You stay in Logic or Live, prompt, set a length, print a take, drag it onto the timeline. Official download-page line: "Prompt it, print it, drag it to the timeline."
The web app is the iteration tool. May was generate-then-export, or a ComfyUI graph. The new browser deck is a four-track session with free tape edits and paid generations when you add, replace, or extend audio. Same-day guide: Make your first mix with Stable Audio.
Neither is 1.0. Stability AI calls this "an early step toward a larger vision."
The plugin: generation as an instrument
Stable Audio loads as an instrument. That is not a wording quirk. In Ableton it sits under PlugIns on an Instrument track. In Logic it sits in the Instrument slot. You instantiate a source, then print audio into the arrangement.
Three controls define the first pass:
Tempo. Syncs to session BPM so the clip is meant to fit the project you already have, instead of arriving at 120 and forcing a warp.
Length. You pick the duration. Official copy: a short section through a full six-minute song. That matches the 3.0 family's variable-length design. May's Medium figure was 6:20.
Takes. Generations land in a playlist. Keep more than one. Decide later.
No published spec for latency, sample rate, or how a print behaves under freeze.
Two model options
Official copy: "One plug-in, two model options. Switch between models depending on what you're creating and where."
| Model | Where it runs | Official pitch |
|---|---|---|
| Stable Audio 3.0 Large | Hosted | "The most advanced musicality in the Stable Audio 3.0 model family, with composition up to more than six minutes." |
| Stable Audio 3.0 Small & Medium | On-device music and SFX packs | "Full music composition and SFX generation on-device. Useful for when your studio is offline or you'd rather keep a session off the network." |
On-device packs are optional downloads. File sizes are fetched live. No fixed GB figure is published.
This is not a vocal writer. The first-mix guide: "Stick to instrumental prompts: Stable Audio 3.0 models are not designed for vocal output." For sung demos, use the vocal-first tools in best AI music generators in 2026.
OS and DAWs, as published
- macOS AU and VST3, Apple Silicon and Intel — announcement
- Download for Mac — live on the plugin page
- Download for Windows (Coming Soon) — same page
- Logic Pro and Ableton Live — named on the announcement
- Pro Tools and "most major DAWs" — named on stableaudio.com
- AAX: not listed. Announcement only names AU and VST3
Windows users wait. Pro Tools users should confirm the current installer. The product page says yes; the announcement does not mention AAX.
The web app: a session, not a one-shot
StableAudio.com now starts from FULL MIX or MULTI-TRACK, then a length preset: AUTO, 0:30 LOOP, 1:00 SHORT, 1:30 SONG, 3:00 EXT. The homepage also says one second to six minutes. MULTI-TRACK is experimental.
You land in a four-track deck. Official metaphor: analog tape. When slots fill, bounce tracks together. Bounce can be undone for one hour.
Edits that do not cost credits (no new audio):
- Tape: Reverse, Stutter, Freeze
- Splice: cut, copy, paste, then PRINT to TRACK or ALL
- Mix moves, bounce, export
Actions that do cost credits:
| Action | What it does |
|---|---|
| Add Track | Records one more part that listens to the session |
| Regenerate | Re-records a single track; leaves the rest |
| Replace Section | Re-records a time range on one track |
| Extend | Continues a track to a real ending, not a loop |
Failed generations are not charged.
The chat box — Stability AI calls it "the producer" — takes plain-language direction: less low end, a later build, colder, more clinical. Audio-to-audio is the other path: start from audio you already have and push genre, mood, or instrumentation.
Deck controls: level, pan, mute, solo, plus per-track and master effects. Export is MIXDOWN (one WAV) or STEMS (a ZIP). Stem generation is still marked experimental. SHARE copies a session link.
Four tracks and a chat box are a sketchbook. They are not a DAW.
Specs
| Official detail | |
|---|---|
| Product | Stable Audio plugin + enhanced StableAudio.com |
| Status | Early beta; some features experimental |
| Plugin type | Instrument |
| Formats | macOS AU, VST3 |
| CPU | Apple Silicon and Intel |
| Windows | Coming Soon |
| Named DAWs | Logic Pro, Ableton Live; product page adds Pro Tools |
| AAX | Not published |
| Length | Short section to a full six-minute song |
| Tempo | Session BPM |
| Takes | Playlist |
| Models | Large (hosted); Small & Medium on-device |
| Web modes | FULL MIX, MULTI-TRACK (experimental) |
| Web length | AUTO / 0:30 / 1:00 / 1:30 / 3:00; homepage also says 1s–6 min |
| Web tracks | Four slots; bounce to free space |
| Vocals | Not designed for vocal output |
| Licensing | Fully licensed training data; you own outputs |
| Plugin price | Unpublished |
| Official launch video | None on the Aug 18 post |
How it compares
May's Stable Audio 3 release was a model family plus ComfyUI. Suno Studio is the other "keep working" surface — a browser DAW, not a plugin.
| Stable Audio plugin + web (Aug 2026) | Stable Audio 3 (May 2026) | Suno Studio | |
|---|---|---|---|
| Surface | AU/VST3 instrument; browser deck | Open weights, API, ComfyUI, original web app | In-browser generative workstation |
| Where audio lands | On a DAW track, at session tempo | Files you import, or a graph | A browser timeline you later export |
| Length | Up to a six-minute song | Small ~2:00; Medium/Large past six minutes (Medium 6:20) | Full songs inside Suno's editor |
| Vocals / lyrics | Not designed for vocals | Instrumental and SFX | Songs with vocals are the point |
| Iteration | Playlist takes; web direction, audio-to-audio, add/replace/extend | Prompt, inpaint, continuation, LoRA, ComfyUI | Timeline edits, stems, browser tools |
| Offline | Optional on-device Small/Medium packs | Yes, if you run the weights | No |
| OS | macOS now; Windows Coming Soon | Whatever runs the weights or a browser | Browser |
| Status | Early beta | Released May 20, 2026 | Separate product; still a browser DAW |
| Price | Unpublished for the plugin | Open weights free to run; API/enterprise separate | Suno's own subscription |
Deeper model notes: Stable Audio 3 review. Wider field: Best AI music generators in 2026.
Tuesday-session difference: May generates somewhere else. The plugin generates on the track. Suno Studio stays in the browser.
Who it's for — and who should skip
Use it if you are on a Mac in Logic or Live and want beds, loops, cues, SFX, or instrumental starters at session BPM. On-device Small/Medium packs matter if the session cannot go to the cloud.
Use the web app if you want to talk a mix into shape before you bounce stems. Four tracks and a producer chat are a sketchbook.
Skip it if you need Windows today, vocals, published pricing, or AAX. ComfyUI and open weights did not go away.
Pricing
Unpublished.
The August 18 announcement lists no dollar price, credit pack, or plugin SKU. The plugin download page lists none either. We are not inventing a storefront.
What is official: the web guide treats credits as real. Start, add track, regenerate, replace section, and extend cost credits. Tape, splice, mix, bounce, and export do not. Failed generations are not charged. That is a web-app fact, not a plugin price list.
Rights, on both surfaces: models trained on fully licensed data, commercially safe, you own outputs and can distribute them freely. Organizations above the Community License revenue threshold still need the Enterprise path from the May 3.0 launch. Read the Community License Agreement before you ship a library.
As of August 19, 2026, there is no plugin price we can print from the launch materials. Check stableaudio.com/pricing if a plan page is current.
FAQ
Is this a finished plugin or a beta?
Early beta. Stability AI says some features are experimental and that it will iterate in real time. The first-mix guide repeats the same warning for the web app.
Does it run on Windows?
Not yet. The download page offers Mac and marks Windows Coming Soon. The announcement only specifies macOS AU and VST3.
How long can a generation be?
Official plugin copy: a short section to a full six-minute song. The web app lists 30-second through 3:00 presets, and the homepage says one second to six minutes. May's Medium model was documented at 6:20.
Can I use the audio commercially?
Stability AI says yes for these commercially safe models: you own the outputs and can distribute them freely, under the Community License (Enterprise above the revenue threshold). Read the license.
Is there an official walkthrough video?
No. The August 18 announcement does not embed one. Older channel videos are 2.5-era or general intros, not this plugin.
What we still cannot tell you
We have not installed the AU or VST3. No CPU figures, no print-quality notes, no BPM-sync test. We cannot confirm Pro Tools beyond the product-page line. We cannot quote a price.
The idea is the one May was missing: generation on the instrument track, at session tempo, at a length you choose. The web deck is the other half — keep the take, talk to it, bounce stems, leave.
Download the Mac build from stableaudio.com/plugin if you want into an early beta. Keep a real instrument on the next track.
