Music Addendum

This addendum applies when you use nal-music at music.nal.digital, the ACE-Step audio generation pipeline, or any NAL Service that produces audio output. It is part of the Terms of Service and incorporates the AUP, Privacy Policy, and Content License.

Effective date: 2026-05-15

1. What nal-music Is

nal-music generates audio (songs, instrumentals, sound effects, voice-over) from text prompts and optional reference inputs. It uses the ACE-Step pipeline (currently 1.5-XL-turbo + DF11 weights) running on NAL infrastructure and community GPUs via NAL Net, with Runware as a fallback provider.

2. Pricing

2.1. Generation is priced per second of output, per model. The price applicable is shown before you submit a job.

2.2. Quality vs. price. Higher-quality models (longer denoising steps, higher sample rates, vocal models with cloning support) cost more. The model picker shows the per-second price for each.

2.3. Fallback to Runware. If NAL's primary ACE-Step pipeline is unavailable or times out, jobs may be transparently re-routed to Runware. The per-second price quoted at submission is the price you pay regardless of which provider serves the job; if Runware's price is higher, NAL absorbs the difference.

2.4. Failed jobs are refunded in full (see Payments and Refunds Policy).

3. Rights in Generated Audio

3.1. Commercial use is intended. nal-music is built for commercial output — advertising, monetised video, paid streaming, soundtrack work, client deliverables. This is the default for almost every model.

3.2. Ownership and the AI-copyright gray zone. Subject to Sections 3.3–3.6 below, you own what nal-music generates for you. Section 5 of the Content License applies in full, including the gray-zone disclaimer: copyright law on AI-generated audio is unsettled — in some jurisdictions purely AI-generated output is not eligible for copyright protection. You can use the output, but you may not be able to legally stop someone else from reusing it. We will not arbitrate disputes over copyright in AI-generated audio between users.

3.3. Underlying model licence overrides. Some models have research-only or non-commercial licences. Selecting one pins that restriction to the output regardless of your intent. The model picker surfaces these restrictions.

3.4. Vocal cloning and voice likeness. Generating recognisable vocals of a real person (singer, actor, public figure) requires that person's documented consent. Cloning a public-domain or licensed training voice that NAL provides is allowed. See AUP Section 4.4.

3.5. Style mimicry. Generating audio in the recognisable style of a living artist who has not consented is restricted. nal-music filters certain style prompts. Bypassing the filter is a violation of the AUP.

3.6. Training-data attribution. ACE-Step is trained on a wide corpus. Output may resemble training data. NAL does not warrant that output is original or non-infringing. You are responsible for verifying before commercial release.

4. Lyrics

4.1. You-provided lyrics. Where you supply lyrics, you represent you have the right to use them. NAL stores lyrics with the job for reproducibility.

4.2. Lyric prohibitions. Lyrics that would violate the AUP (sexual content depicting minors, incitement to violence, terror-promoting, etc.) are blocked at submit time. Filter evasion violates the AUP.

5. Generated Tracks — Storage and Sharing

5.1. Generated tracks are stored in NAL's S3-compatible storage (Hetzner / MinIO) under your account. You may:

• download the track at any time from the music library;
• share a public link (with optional expiry);
• post the track to Guild (subject to the Guild Addendum and the Guild post fee);
• delete the track, after which it is purged from storage within 24 hours and from backups within 30 days.

5.2. Streaming bandwidth. NAL's hosted streaming for shared links is included in the generation price. Extreme outbound bandwidth (non-personal redistribution at scale) may be rate-limited.

6. Real-Time Visualisation

6.1. nal-music provides a real-time visualiser (Vibe Cells, particle visualiser, fullscreen mode). The visualiser runs in your browser. NAL does not stream visualiser output.

7. Reference Tracks

7.1. Where you upload a reference track (for style transfer, melody conditioning), you represent that you have the right to use the reference for that purpose. Reference uploads are not stored beyond the active job unless you save them explicitly.

8. NAL Net Routing — Third-Party Visibility

⚠️ Your job is sent to a third party. Music generation jobs default-route to community-operated GPU workers through NAL Net. The worker is an independent third party running a worker on their own hardware. They see your prompt — including lyrics, style prompts, and any reference audio — and the generated track in plaintext. Pick "hosted endpoint only" in the model picker if your material is sensitive (unreleased commercial work, NDA'd content, personal voice recordings).

When music jobs route through community GPUs, the privacy and routing terms in Section 4 of the Studio Addendum and the NAL Net Addendum apply. Workers are contractually forbidden from retaining, copying, training on, commercially using, or republishing material that passes through their worker — but NAL cannot physically prevent exfiltration by a determined bad actor.