Short version: we do not sell, license, or otherwise provide our audio catalog for generative AI training. Every track in our library is composed and produced by a human being. That is not going to change.
What happened
In July 2026, two companies we worked with, including our database provider SourceAudio, offered us a combined $100,000 deal to let generative AI audio companies train their models on our catalog.
$100,000 is a significant amount for a label our size, close to twice what our Patreon generates in a year. We’re the largest TTRPG audio label, but TTRPG music is a niche within a niche. Turning it down wasn’t a hard decision though.
We said no. We ended two ongoing contracts with the other company involved, and closed our SourceAudio database entirely. We spent the following weeks building our own database and licensing tool from scratch instead, now fully connected to Patreon, something SourceAudio was never able to do.
We don’t know of another TTRPG audio creator who’s received an offer like this. We doubt we’re the only one.
We first talked about that topic on our Patreon. Read the whole story here.
Why a “small” catalog was worth six figures
Our library is about 6,000 tracks. That might sound too small to be worth real money to an AI company, most training datasets run into the millions of hours. But a few things make our catalog specifically valuable, not audio catalogs in general.
We have real multitrack stems, not just finished mixes. A stem is one isolated layer of a track (percussion, drone, melody, foley) before it gets mixed into the final file. Genuine multitrack stems are hard for AI companies to get their hands on legally: exposing the individual layers behind a mix means exposing actual recording and mixing techniques, and rights to that material are usually held tightly by the original studio or composer. Most of what’s floating around online for AI training is AI-approximated separation, an algorithm’s best guess at un-mixing a track after the fact, which is a lossy, imperfect substitute for the real thing. We have the real thing, for most of our catalog.
Every track is tagged consistently: genre, era, mood, all of it, across the entire library. Call it a Swiss thing. We can only tolerate the most complete kind of order, and a folder structure held together by inconsistent naming and missing metadata was never going to survive contact with me for long. Most audio libraries are exactly that kind of mess. Ours isn’t.
The rights are clean and centralized. One point of contact, direct contracts with every composer, no fragmented ownership scattered across dozens of unreachable rights holders.
The quality is independently validated. We’ve won ENNIE Awards recognition and we’re Paizo’s official audio partner. For a buyer trying to judge whether a dataset is actually good and not just big, that kind of third-party validation matters.
Put together: a small, clean, fully-tagged, legally simple, quality-verified dataset with real stems is worth more per track than a massive pile of scraped, unlabeled audio. Scale isn’t what’s scarce anymore. Being small and well-organized doesn’t protect you from these offers. If anything, it makes you a more attractive target.
What our community said
We posted the story publicly on Patreon, and the response told us we’d made the right call. One comment stuck with us: “$100k once. And then what? You sold your livelihood for a pittance and will soon be competing against slop-you.” Our own reply summed up how it felt: turning down that offer was like refusing to sell family land to an oil company.
A few patrons told us directly they joined or upgraded specifically because of this decision. That means more to us than the offer itself ever could.
Are we perfect on this? No, and we’d rather tell you that ourselves
For a while, videos on our YouTube channel used AI-generated artwork. That wasn’t someone else’s call made behind our back, it was a deliberate decision we made ourselves, and looking back, we regret it.
Here’s the math that led there: the average video on our channel makes about $0.50 a month. A single commissioned illustration from one of our artists starts around $300. At $0.50 a month, that’s roughly 50 years to break even on one piece of art, for a channel that used to publish three times a week. That math simply never worked.
We started using AI specifically for ambiances so original and niche that no existing artwork could represent them; some tracks sat unpublished for months because we genuinely couldn’t figure out how to illustrate them. That narrow use case made sense to us at the time, and it somewhat still does. But it crept. We ended up leaning on AI even in cases where we didn’t strictly have to, and that’s the part we regret.
We’ve stopped entirely. We’re working through roughly 150 affected videos, replacing that artwork with human-made art, the same way we’ve credited real artists across 900+ videos.
Where YouTube stands for us now
Honestly, YouTube in 2026 is close to a lost cause for a channel built the way ours is, which is part of why we’ve pulled our time investment there back. Channels like Bardify repost stock audio, pulled straight from libraries like Epidemic Sounds, behind a pretty thumbnail, with essentially no original work involved. Marketplaces like Fiverr openly sell entire “ambience channel” packages for a few dollars, mass-produced and built to be forgotten. We genuinely believe our audio is far better than what’s flooding that space. But the YouTube of 2026 doesn’t reward that the way the YouTube of 2022 did. So we’d rather put our energy where the work still gets valued: our Patreon, and Opus.
Our commitment: no generative AI training, ever
- We will not license our catalog, in whole or in part, for generative AI training, regardless of the offer.
- Every composer we work with signs a contract with a simple clause: using AI to generate the sounds they deliver to us is grounds for immediate termination.
- Every track we release is composed, performed, and produced by a human composer or sound designer, credited and compensated under standard commission terms.
- If we ever receive another offer like this, we’ll say no again, and we’ll say so publicly.
If you want to support this
The best way to support labels who’d rather say no to offers like this is simple: keep supporting the people making the work.
Our Patreon gives you access to a growing, entirely human-made library.
Our album shop gives you access to our 60+ albums and FoundryVTT modules.
Our license shop lets you license tracks for your own projects, TTRPG or otherwise, all cleared, all human-made.
—
Last updated: August 2026
