What's a render?

One generation. Free covers 5 a day. Pro covers 300 a month. Each render produces a WAV master at 24-bit / 44.1 kHz and an MP3 320 preview.

How long does a song take?

About 8 seconds for the v1 mock take. The v1.5 latent diffusion track targets 12 seconds end-to-end for a 90-second song. We post a status note when latency moves.

Can I edit lyrics after generation?

Yes. Open the song page and pick Remix this. The composer prefills with the original prompt and lyrics, and the persona stays attached unless you change it.

What goes in the textarea?

Sentences, comma-separated tags, or a quoted lyric phrase. The model reads all three. A short BPM / key suffix (96 BPM, F minor) anchors the take when you have one in mind.

What do the three modes do?

Simple takes a sentence and infers the rest. Custom adds explicit BPM, key, and persona slots. Lyrics expects a verse and chorus block. We render the take from the words first.

How does top-k sampling affect output?

We sample at k=40 by default. Lower k tightens the take toward the persona's mean voice; higher k widens it. The setting is exposed in Studio for fine control.

Are personas real people?

No. Each persona is a curated synthetic voice anchored to a 768-dim WavLM embedding pulled from consented studio takes. The voice tag maps to a stable model checkpoint.

Can I add my own voice?

Studio ships a custom voice slot in v1.5. You record 8 minutes of clean speech and one short sung phrase, we fine-tune the embedding, and you keep the persona on your account.

Can I move a persona between accounts?

Studio personas are seat-bound for v1. Team-share is on the v2 schedule once SSO and seat billing land.

Are stems available now?

Not in v1. Demucs v4 (htdemucs_ft) is wired to the queue and opens to Studio in Q3 2026, Pro in Q4 2026. The /stems page lists the spec we are committing to.

Will stems be phase-aligned?

Yes. Vocals, drums, bass, and other sum back to the master within 0.05 dB RMS. We test that bound on every release before promoting the build.

Can I split a track I did not render?

Closed beta accepts uploads up to 64 MB WAV. You retain rights to the upload and to the resulting stems. We do not retain the source file after the split job clears.

Where is my library stored?

Drafts stay on this device until you sign in. After sign-in, songs persist to your account and sync across devices over the same /api/graphql gateway the studio uses.

How long do drafts last on this device?

Indefinitely, until you clear browser site data. Drafts are stored in localStorage under sonichaos.library. Sign in to keep them after a browser reset.

Can I export everything?

Yes. Account settings > Export builds a ZIP of every WAV, every MP3 preview, and a JSON metadata file with the prompt, persona, BPM, and timestamps.

How do I cancel?

Account settings > Plan > Cancel. Billing stops at the end of the current period. Annual plans keep their current term until the renewal date.

Do I get a refund?

Seven days from the first paid charge on Pro or Studio. Email billing@sonichaos.ai with the order ID. Refunds do not delete the renders you made during the window.

Can I switch monthly to annual mid-cycle?

Yes. The unused portion of the monthly term credits against the annual invoice. Downgrade back to monthly at the next renewal.

Who owns the songs?

You do, under the license tied to your tier on the date of render. Free is personal use only. Pro covers paid client work up to 1 million plays. Studio is unlimited commercial, broadcast, and sync.

Can I publish to Spotify?

Pro and Studio cover commercial DSP publishing. Free tier renders may not be published to commercial platforms, including Spotify, Apple Music, YouTube Music, and TikTok Sounds.

What happens if I cancel?

Renders you made stay yours. The license follows the tier active when the song was generated, regardless of your current subscription state.