Beta Tester Quickstart

Welcome to the BananaRat Studio beta. This page tells you what we built, what to try, what to verify, and how to tell us when something breaks.

What is BananaRat Studio?

An AI music platform built for copyright-conscious songwriters. You describe the song you want; the studio writes lyrics, picks chords, generates tablature, and produces two parallel audio renders: a notation-accurate version (built from MIDI + WAV stems via FluidSynth — registration-authoritative for PROs and USCO) and a production soundtrack with vocals (generated by ACE-Step neural audio).

Every song you produce carries an embedded C2PA content credentials manifest — a cryptographic record of what was generated, when, and a tamper-evident hash of your editorial decisions. Look for the ✓ AuthorMark embedded pill below the player after each render.

Known limitations — read before you judge

We're shipping Open Beta with a few things deliberately deferred so we can get you in the door faster. None are defects — they're disclosed so you can form a fair opinion.

  • EU testers join Week 2 (2026-06-01). First wave is US & Canada only while we ship our GDPR consent flow and Article 17 deletion endpoint. If you're in the EU, your invite is coming — sit tight.
  • Cryptographic signatures verify, but our cert isn't in any public trust list yet. Every song carries a C2PA manifest signed by our development Ed25519 key. The signature math is valid; public verifiers (Adobe, Content Authenticity Initiative) will show "signed by an unknown authority" until we swap to the DigiCert Content Authenticity cert at GA (late July). Your AuthorMark is real — just not yet in anyone else's trust list.
  • EU audible disclosure ships June, not today. The EU AI Act Article 50(4) audible-stamp requirement (a spoken "this song includes AI" at the start) activates for EU deliveries in June. For US/CA testers it's irrelevant — you won't hear it either way.
  • Studio Quality re-render is slow. About 5 minutes on the XL SFT model. The progress bar counts honestly; the spinner isn't stuck. Grab a coffee.

Four things to try

  1. Compose a song. Tell the chat what you want — genre, mood, theme. Multi-turn chat ("a sad song about a rainy day in jazz") accumulates more authorship signal than Quick Start single-prompt. Watch the rotating progress labels: Writing lyrics → Arranging chords → Generating tablature → Finalizing. Should take 30-60 seconds.
  2. Regenerate a section. Click Regenerate on any [Verse], [Chorus], etc. in the lyrics panel. Pick "lyrics only" or "everything", optionally type "make it more upbeat", click Generate Alternatives. Pick the one you like, click Apply Selected. The audio re-renders.
  3. Restyle into a different genre. After your song's neural audio is ready, pick a style from the dropdown (Acoustic Folk, Metal, Jazz Ballad, etc.), set the strength slider, click Generate Restyle. Cover Mode keeps your song's structure but applies the new vibe. ~1 minute.
  4. Render in Studio Quality. Below the production soundtrack player, click 🎵 Render in Studio Quality. Routes through ACE-Step XL SFT at 50 inference steps for richer vocals + cleaner production. Takes 3-5 minutes on the GB10 — be patient. The button shows elapsed seconds + % complete.

Three things to verify

  1. Downloads work. Click any Download button. The MP3 / WAV / MIDI / MusicXML / AAF should all download cleanly with the song name as the filename.
  2. Provenance manifest exists. After a render completes, look for the gold ✓ AuthorMark embedded pill below the player card. That confirms your audio file carries a signed C2PA manifest. To verify the manifest itself, drop the downloaded MP3 into the public Adobe verifier at contentcredentials.org/verify — you should see "BananaRat Studio" as the producer with a signed timestamp. (Note: trust state will read "Invalid" until our DigiCert cert lands; the signature itself is cryptographically valid, just self-rooted for beta.)
  3. Share link works. From the My Songs library, click Share on any song. Open the shared URL in a private/incognito window — anonymous visitors should be able to play your neural MP3 + see lyrics. Stems and FLAC remain gated behind login.

Things you might notice

  • Cleanroom edits. If your lyrics mention a real artist by name (band names overlap with common words — "Anthrax", "Nirvana", "Madonna", etc.), the cleanroom layer redacts the name before generating audio. You'll see a notice: "X artist references in your lyrics were replaced with [artist-redacted] before generating audio." This is intentional — keeps you out of style-imitation copyright concerns. Why we do this →
  • Famous lyrics detection on imports. If you upload a lyrics file that matches a known famous song, the import path will warn or block depending on similarity confidence. Three-tier UX: soft-warn at 50-84%, warn-with-override at 85-94%, hard-block at 95%+. Built on a 61-entry verified blocklist + a Haiku 4.5 classifier.
  • One render at a time. Studio Quality and Restyle compete for the same GPU. The studio button greys out while restyle is running and vice versa. You'll see "Audio engine busy — wait for X to finish".
  • "New version available — Refresh" banner. We're shipping fixes during the beta. If a banner appears at the top, click Refresh to load the latest UI.
  • Rendering progress estimates. All progress bars are time-based, not real backend events. If a label says "Almost there" but it's still going, that's normal — XL SFT can take 4-5 minutes on a busy GPU.

How to report bugs

For anything broken, confusing, or surprising:

  1. Email: feedback@bananarat.com with subject "Beta Feedback".
  2. What to include:
    • What you were trying to do
    • What you saw vs. what you expected
    • The composition ID if relevant (visible in your My Songs library)
    • Browser + OS
    • Screenshot if it's a UI thing
  3. Severity hint: "Stuck", "Lost work", "Wrong output", "Cosmetic" — helps us prioritize.

What we're testing for

Honest list of what we don't know yet:

  • Does multi-turn chat intake feel natural?
  • Does the cleanroom redaction notice make sense or feel intrusive?
  • Are Studio Quality / Restyle wait times tolerable?
  • Does the AuthorMark pill mean anything to you, or is it noise?
  • Where do you get stuck looking for an action?
  • What format / use case is missing?

Thank you for testing. Each piece of feedback shapes the platform before public launch. Your songs are yours — we never claim ownership, only platform license. The full legal posture is on our Legal Notice page.