stems is an instrument that writes musical parts for the track you’re making. Using it — through the plug-in, the stems app, an API key, or this site — means you agree to the following. Written to be read, not to hide things.

stems 1.0: on your mac

From version 1.0 the plug-in writes on your mac. The audio it hears, the words you type, and the parts it writes never leave the mac — we don’t receive them, store them, or train on them. What reaches us is only your account and the model download (see below). Everything after this section about uploads and generations applies to the cloud that older plug-ins (before 1.0) and the API still use.

what we collect

• Audio you send (older plug-ins, the API): the context clips uploaded when you generate in the cloud.
• What comes back: the generated stems and mixes.
• Your prompts: captions, lyrics, and generation settings (steps, seed, length).
• Account & usage: your email, hashed API keys, per-day cloud generation counts, and that your key fetched the model (which files, when).
• Technical logs: timestamps, request metadata, and a hashed (never raw) credential identifier tying requests together.

what we do with it

• Run the service: route your audio to the model and bill your daily allowance.
• Debug and improve quality — including listening to inputs/outputs internally.
• Train our models. Uploaded audio, prompts, and generated stems may be used as training data for future stems models. This is the deal that keeps the free tier free. If you are not comfortable with your playing being training data, do not use stems yet — an opt-out tier may come later.

what we don't do

• No selling your audio or your data to third parties.
• No publishing your uploads or takes without asking you first.
• No storing raw credentials — keys and tokens are hashed at rest in our systems.

who touches the data

Infrastructure the service runs on: Cloudflare (routing, storage, the model download), Supabase (accounts, database), and Hugging Face (cloud inference for older plug-ins, and the model download when you set stems up without a key). Cloud audio transits and may be stored on these providers under their own security terms.

your rights

• Yours stays yours: you keep all rights to the audio you upload, and we claim no ownership over stems generated from your material.
• Deletion: email colo28197@gmail.com from your account address and we will delete your account, keys, and stored audio within 30 days. Models already trained are not un-trained.
• We may update these terms; the version and date above change when we do.

credits

Powered by Stability AI. stems 1.0 runs Stable Audio 3 Medium by Stability AI, unmodified, under the Stability AI Community License (and the Gemma terms for its text encoder). The model’s license, not ours, asks organisations with more than $1M a year in revenue to get an enterprise license from Stability AI. The parts you write are yours.