License the music behind Collection One.
Collection One is a licensed music dataset: 500 commercially released, ISRC-verified, rights-cleared tracks with full metadata and rights flags, free on Hugging Face for research and evaluation. This page is for what comes after the research: commercial AI training, production deployment and the larger catalog behind the corpus.
{ "name": "collection-one",
"tracks": 500,
"source": "commercially_released",
"isrc_verified": true,
"generated": false,
"license": "research_and_evaluation",
"commercial_use": "license required" }The corpus is the sample, not the catalog.
Collection One is drawn from 110,000+ globally cleared songs from 3,500+ independent artists. All of them can be licensed for traditional, non-generative AI, such as search, tagging, classification and recommendation. 35,000+ are also cleared for generative AI today under signed GenAI Addenda, opt-in and artist by artist, and that number is growing.
Recording and composition are cleared together. AI permissions are set track by track. The generative AI license includes downstream commercial rights. Nothing in the catalog was scraped, and nothing was generated.
If the 500 tracks earned your trust, the catalog is the same thing at scale.
Research is free. Revenue needs a license.
The rule is simple: the commercial license begins where deployment begins. Internal evaluation stays free.
| Use | Research and Evaluation License (free) | Commercial license |
|---|---|---|
| Research, education and benchmarks | Yes. Papers, benchmarks and research code are welcome. | Not needed |
| Model evaluation and experimental training | Yes, internal only | Not needed |
| Training or fine-tuning a model you will ship | No | Yes |
| Serving a model built on the corpus, by API or hosted endpoint | No | Yes |
| Using tracks in content you publish or monetize | No | Yes |
| Going beyond the 500 tracks | No | Yes: 110,000+ for non-generative AI, 35,000+ for generative AI |
Built for the procurement review, not around it.
From first email to licensed data, with every use on record.
Artists share in every license.
Artists share in the revenue from every generative AI license we sign, on an ongoing basis. No flat-fee buyouts.
That is the point of the company.
Collection One FAQ
Straight answers for researchers, AI teams and artists.
Is Collection One really free?
Yes, for research, education, benchmarking and model evaluation, under the Research and Evaluation License you accept on Hugging Face. Commercial use needs a separate written license from The Rights Foundry.
Is the music AI-generated?
No. Every track is real, human-made, commercially released music, professionally composed, performed and produced. Nothing in the corpus was generated, and nothing was scraped.
Where does the music come from?
The corpus comes from Audiosocket, The Rights Foundry's first catalog provider, a rights-cleared commercial catalog built over seventeen years. The Rights Foundry publishes and licenses it.
Can I train a commercial model on this licensed music dataset?
Yes, with a commercial license. The free license covers experimental training only. AI training for production is exactly what our commercial licenses are for. Email licensing@therightsfoundry.com.
Can I license more than 500 tracks?
Yes. All 110,000+ globally cleared songs can be licensed for traditional, non-generative AI, such as search, tagging, classification and recommendation. 35,000+ are also cleared for generative AI, including training and generative output, and that number is growing.
Do artists get paid?
Yes. Artists share in the ongoing revenue of every generative AI license. No artist rights are ever bought out with a flat fee.
Ready to go further?
If your model or product needs music it can actually ship, tell us what you are building.
Built with love. Cleared by default.