TL;DR
- Browser-based AI music DAW built on open-weight Stable Audio 3
- One-click local install via Pinokio, CPU (Small) or 8GB NVIDIA (Medium)
- MIT-licensed, instrumental and SFX only, 44.1kHz stereo, up to 380s tracks
System Requirements
| RAM | 16GB |
| GPU | RTX 3060 (8GB VRAM) |
| VRAM | 8GB |
✓ Apple Silicon
Update, 2026-07-25: StableDAW has been archived and continues as theDAW, now with ten workspaces instead of seven. The Pinokio launcher below has not been updated and still installs the archived June build. This article remains an accurate account of that build. For the current release and working install paths, read part two: StableDAW is now theDAW.
StableDAW just landed in the Pinokio catalog, and it is the first AI music tool that feels like an actual studio instead of a text box with a download button. It is a browser-based digital audio workstation built around Stability AI's open-weight Stable Audio 3 models, it installs in one click, and the small model runs on a plain CPU. No account, no API key, no cloud credits. You generate, edit, mix, and perform entirely on your own machine.
The catch up front, so nobody wastes a download: Stable Audio 3 makes instrumental music and sound effects. It does not sing. If you need vocals, this is not your tool. If you want a local, MIT-licensed beat factory that you fully control, read on.
StableDAW, branded "theDAW" in the app, was built by GANTASMO at a Music Hackspace hackathon.
What StableDAW actually is
StableDAW (the app calls itself "theDAW") was built by GANTASMO during a Music Hackspace music-technology hackathon at Berklee College of Music. The code is MIT-licensed on GitHub, and cocktailpeanut wrote the Pinokio launcher that turns the whole stack into a one-click install.
It is not a one-trick prompt-to-clip generator. It is a full workstation with seven workspaces, laid out as a left-to-right workflow across the top tab bar:
- MAKE generates audio from text, from your own recordings, or both. Text-to-audio, audio-to-audio, inpainting, continuation, and a "Chimera" blend that fuses multiple clips.
- EDIT is a multitrack timeline with clip arrangement, trimming, fades, and a snap grid.
- MIX hosts 24 FFmpeg-powered effects, a mastering chain, a drum sequencer, and a piano roll, and exports to FLAC, MP3, AAC, or Opus.
- DJ is a two-deck console with sync and FX, and it maps to MIDI controllers.
- VJ drives a 3D audio-reactive visualizer for live shows.
- TRAIN fine-tunes LoRA adapters on your own audio (this one needs an NVIDIA GPU).
- LEARN draws an interactive genealogy graph that traces how every clip descended from its parents.
Those seven are the June build documented here. theDAW now runs ten: TRAIN was renamed Underfit, and Perform, Foundry and Audimate joined it. Part two covers all four.
That last one is the tell. StableDAW keeps a persistent, searchable library and remembers the full lineage of every render: which prompt made it, which clip it was inpainted from, which stems it was split into. For anyone who has lost a good take to a closed cloud tool, that local history is the selling point.
The engine: Stable Audio 3, open weights, no vocals
Under the hood is Stable Audio 3, a diffusion transformer paired with Stability AI's SAME autoencoder. It outputs 44.1 kHz stereo, which is CD quality. StableDAW ships two checkpoints, and the difference decides what hardware you need.
| Model |
Size |
Hardware |
Max length |
Use it for |
| Small (433M) |
~3.5 GB |
CPU or any GPU |
120 sec |
Default. Loops, sketches, SFX, laptops without a GPU. |
| Medium (1.4B) |
~10.4 GB |
NVIDIA, ~8 GB VRAM |
380 sec |
Full-length tracks, higher musicality. |
The launcher pulls both from ungated community mirrors (cocktailpeanut/stable-audio-3-* on Hugging Face), so there is no license click-through and no token. The Small model downloads automatically; you grab Medium from a button when you want it. A separate Medium-RF checkpoint is the base for LoRA training in the TRAIN tab.
On licensing, the app code is MIT. The model weights sit under the Stability AI Community License, which lets organizations under 1 million dollars in annual revenue use the generated audio commercially. Past that threshold you need an enterprise agreement. For an indie producer or a small studio, the practical answer is: the music you make is yours.
Install it in one click with Pinokio
No longer current. The launcher below took its last commit on 2026-06-12 and still clones the archived gantasmo/stabledaw repo. The install succeeds, so there is no error to warn you, but you get the June build. See part two for install paths that land on theDAW.
If you have not used Pinokio before, it is a desktop launcher that installs and runs AI apps inside their own isolated environments, no terminal required. Install Pinokio first, then add StableDAW from its repository page.
# 1. Install Pinokio from pinokio.computer (Windows, Linux, or macOS)
# 2. In Pinokio, open the StableDAW launcher repo:
# github.com/cocktailpeanut/stabledaw.pinokio
# 3. Click Install, then Start. The launcher handles the rest.
Behind that one click, the launcher clones the repo, builds a Python 3.10 environment with the right torch for your platform via uv sync, runs npm ci for the React frontend, and downloads the ~3.5 GB Small model. Then it boots a FastAPI backend on port 8600 and a Vite frontend on port 5173, and opens the web UI.
Two things to budget for. You need roughly 10 to 15 GB of free disk for the environment plus the Small model, and another ~10.4 GB if you add Medium. And both ports (8600 and 5173) must be free, because Vite runs with strictPort and will not shop around for another one.
Platform notes: Windows installs CUDA 12.8 and Flash Attention automatically, Linux x86_64 wants CUDA 12.6, and macOS runs the Small model on CPU. That includes Apple Silicon, which works fine without any NVIDIA hardware, just slower than a CUDA box.
Hands-on: making your first track
The MAKE tab is where you live at the start. Drop a description into the prompt box, set the controls on the left (model, length, sampler steps, CFG, seed, batch), and hit CREATE in the bottom right. Be specific. Genre, tempo, instrumentation, and mood all move the output.
The MAKE workspace: prompt boxes up top, sampler controls on the left, the Chimera blend stage in the center, output format on the right.
A prompt that works well as a starting point:
Lo-fi boom bap with dusty vinyl crackle, warm Rhodes chords,
mellow upright bass, soft swung drums, 84 BPM
Generation speed depends entirely on the model and the silicon. On a recent NVIDIA card the Small model returns a two-minute clip in a few seconds. On CPU, the same clip takes longer, and the Medium model on CPU is measured in minutes, not seconds, so save it for the GPU. These are community-reported figures; we cover the disclosure at the end.
Once you have a clip you like, the generative tools stack. Paint a region and inpaint just that bar. Extend a loop with continuation. Or drag several clips into Chimera and let it beat-align them into one new generation. Every result drops into the library with its full prompt and parameters, so a good seed is never lost.
This is where StableDAW earns the "DAW" in its name. The EDIT tab gives you a real multitrack timeline. You arrange clips, trim them, add fades, and commit edits, all in the browser.
The EDIT tab: a standard multitrack timeline. Drop clips from the library and arrange them.
The MIX tab is a surprise for a hackathon project. It carries 24 FFmpeg-powered effects grouped into a library (mastering chain, vocal processing, lo-fi and vinyl, stereo widener, reverb and delay, club EQ, compressor, loudness normalization, and more), a QUICK MASTER section with punch, air, drive, and ceiling knobs, plus a drum sequencer and a piano roll. You chain effects left to right and hit process.
The MIX tab: an FFmpeg effects library, a quick-master section, and an effects chain you build left to right.
Then it keeps going. The DJ tab is a two-deck console with sync, per-deck FX, and MIDI mapping. The VJ tab is a 3D audio-reactive visualizer for live performance. And the TRAIN tab lets you fine-tune LoRA adapters on a folder of your own audio, then stack them at generation time for a custom sound, provided you have the Medium-RF model and an NVIDIA GPU.
Limitations and gotchas
It is a hackathon build, and a few rough edges show:
- No vocals, period. Stable Audio 3 is instrumental and SFX only. Pair it with a separate vocal tool if you need singing.
- CPU is slow. The Small model is usable on CPU; Medium really is not. Treat a GPU as the requirement for full-length work.
- First boot lags. The initial PyTorch import takes about a minute. That is normal, not a hang.
- The Reset button is a real reset. It wipes your generated-audio library. Export anything you care about first.
- The original repo is archived. Active development moved to gantasmo/theDAW. The Pinokio launcher tracks the working build, so the one-click path still works, but watch the new repo for updates.
Who should use it, and what to do in the next 10 minutes
StableDAW is for producers, indie hackers, and anyone who wants AI music generation they actually own: local files, open weights, full edit history, and a clear commercial path under the Community License. It trades the polished vocals of closed tools for control, transparency, and a studio that never phones home.
The 10-minute version: install Pinokio, add the StableDAW launcher, click Start, and let the Small model download. While it pulls, free up ports 8600 and 5173. When the UI opens, paste the lo-fi prompt above into MAKE and hit CREATE. You will have a track playing before your coffee is cold, and it will be sitting in a library on your own disk, not someone else's server.
If you like running models locally, our tools section has more open-source workstations worth your disk space.
Sources and further reading
Tested on: install walkthrough verified against the Pinokio launcher and the StableDAW GitHub README. Generation timings are community-reported, not benchmarked on our own hardware. Model version: Stable Audio 3 Small (433M) and Medium (1.4B) via the StableDAW launcher.
Date tested: 2026-06-15
Frequently asked questions
Is StableDAW still maintained?
No. StableDAW was archived and development continues as theDAW. The Pinokio launcher referenced in this article still installs the archived build, so use one of the current install paths instead.
Does StableDAW generate vocals?
No. Stable Audio 3 produces instrumental music and sound effects. If you need vocals, this is not the tool.
Can I use the music commercially?
The app code is MIT, and the model weights sit under the Stability AI Community License, which permits commercial use for organisations under one million dollars in annual revenue. Past that threshold you need an enterprise agreement.
