Summary
Create one auditable licensing, attribution, and download-acknowledgement flow for every third-party model and runtime asset used by LSDJ. LSDJ is fully open-source and non-commercial, but the model weights and related assets remain governed by their own licenses and terms.
This issue centralizes the user-facing compliance work so backend and platform issues can focus on engineering. It is not a request for legal advice and should avoid putting sensitive business information in a public issue.
Relationship to other work
Scope
1. Build a revision-specific asset inventory
For every shipped, downloaded, or officially discoverable asset, record:
- upstream project and canonical URL;
- exact package/tag/commit/model revision;
- code license;
- weights/model license and applicable terms;
- required copyright and attribution notices;
- acceptable-use, privacy, or gated-access links that must be shown;
- whether LSDJ redistributes the asset or downloads it from upstream;
- whether credentials or upstream account acceptance are required.
At minimum, cover:
- Magenta RealTime 2 code and weights;
- Apolinario's
magenta-realtime-2 PyTorch package/port and its relationship to the underlying model;
- Stable Audio 3 code and weights;
- T5Gemma/Gemma components and terms;
- each built-in or officially downloadable LoRA independently;
- LSDJ's own code license and bundled third-party notices.
2. Record the applicable project-use path
- Project owners should record that the current product is open-source and non-commercial and confirm the applicable upstream license path for each restricted asset.
- Do not put revenue, account, contract, credential, or other sensitive details in the public issue.
- Treat a future change in distribution or commercial status as a release gate requiring re-review.
3. Define distribution and access rules
- Do not bundle restricted model weights in application installers unless redistribution has been explicitly confirmed for that exact asset/revision.
- Prefer pinned downloads from the canonical upstream source with verified technical provenance.
- Support gated Hugging Face or equivalent access when the upstream requires a user token or prior acceptance of terms.
- An optimized or anonymously downloadable artifact does not remove its underlying license obligations.
- Never log access tokens or store them in plaintext application data; use an appropriate OS credential store or an intentionally ephemeral flow.
4. Add user-facing disclosure and acknowledgement
- Add an About/Licenses surface listing model/runtime names, revisions, upstream links, licenses, notices, and acknowledgements.
- Before a first download, show the applicable model terms and required links when upstream terms require acknowledgement.
- Record acknowledgement against the relevant license/model revision and timestamp so changed terms can be presented again.
- Make clear that LSDJ's license does not relicense third-party weights.
- Keep acknowledgement separate from telemetry or marketing consent.
5. Cover LoRA provenance
- Show provenance and license metadata for LSDJ-provided downloads.
- Show a clear responsibility/provenance notice when users import their own LoRA files.
- Do not mirror or redistribute a LoRA without confirmed permission for that specific artifact.
6. Documentation and tests
- Document where notices live in the repository, application, packaged artifacts, and release notes.
- Add tests for acknowledgement versioning, gated-download blocking, offline/error behavior, and credential redaction.
- Add a release checklist that verifies the inventory matches the revisions actually shipped or downloaded.
Acceptance criteria
- Project owners have recorded the applicable use/license path for every supported model family without publishing sensitive information.
- A complete revision-specific inventory exists for MRT2, the upstream PyTorch port, Stable Audio 3, T5Gemma/Gemma, and every official LoRA.
- Required notices and links appear in the app and packaged third-party notices.
- Downloads that require acknowledgement cannot begin until the current terms are acknowledged.
- Changed license/model revisions trigger a fresh acknowledgement when required.
- Credentials are neither logged nor stored in plaintext application data.
- Installers contain no model weights whose redistribution has not been confirmed.
- Release documentation explains the separation between LSDJ's code license and third-party asset licenses.
Out of scope
- Providing legal advice.
- Enterprise or commercial license procurement unless the project's status changes.
- Reviewing the contents or licensing of users' private imported files.
Summary
Create one auditable licensing, attribution, and download-acknowledgement flow for every third-party model and runtime asset used by LSDJ. LSDJ is fully open-source and non-commercial, but the model weights and related assets remain governed by their own licenses and terms.
This issue centralizes the user-facing compliance work so backend and platform issues can focus on engineering. It is not a request for legal advice and should avoid putting sensitive business information in a public issue.
Relationship to other work
Scope
1. Build a revision-specific asset inventory
For every shipped, downloaded, or officially discoverable asset, record:
At minimum, cover:
magenta-realtime-2PyTorch package/port and its relationship to the underlying model;2. Record the applicable project-use path
3. Define distribution and access rules
4. Add user-facing disclosure and acknowledgement
5. Cover LoRA provenance
6. Documentation and tests
Acceptance criteria
Out of scope