On Wednesday (Sept. 9), Suno announced its new line of AI music models, which the company says are its first “in partnership with the music industry.” Called v6, the models are said to be the fastest models released by the AI music start-up to date — and they are also the first to provide compensation to Suno’s music industry partners in the process.
Related
Suno Launches First AI Music Models ‘in Partnership With the Music Industry’
Billboard to Add Smart Media Albums As Chart-Eligible Sales Formats
Duane 'Keffe D' Davis Gives First Interview After Tupac Shakur Murder Conviction
The news comes after a string of announcements from Suno over the last year, touting new agreements with Warner Music Group, BMG and — as of Tuesday (Sept. 8) — Believe. Still, while these music companies are coming to the table to build products in conjunction with Suno, the AI music company faces ongoing litigation from several others, including Universal Music Group, Sony Music, SOCAN and Koda. It also recently lost a copyright infringement lawsuit in Germany filed by local collection society GEMA over illegal training of its model on copyrighted works.
As part of the v6 roll out, Suno also announced that it has retired all past models and is working towards future products that allow for users to take participating songs from their favorite artists and remix, sample and otherwise make derivatives of them. While it’s unclear when these next products will debut, they would mark Suno’s entry into the crowded AI music remixing race, which companies like Spotify, Udio, KLAY, Hook and more are already pursuing in various ways.
So, is the music business ready to work with Suno? And is Suno actually ready to do the same? Here, Billboard reporters break down readers’ most frequently asked questions about Suno’s latest move.
Related
Suno CEO Mikey Shulman Breaks Down V6 & Warner/BMG Partnership | On the Record
What does it mean that Suno “partnered” with music industry companies?
Bill Donahue: It depends on who you ask.
If you ask Warner Music Group, it means that Suno’s new model is “fully licensed.” In an internal memo Wednesday, CEO Robert Kyncl said the AI company will be shutting down its previous model that was trained on millions of songs without permission. Money from the new licensed model will flow to the participating industry partners and, ultimately, their artists and songwriters, he said.
That would be a huge win for the music industry — and for human creators writ large. The legality of using vast quantities of unlicensed copyrighted works as “inputs” to create AI models is the trillion-dollar, existential fight between AI firms and creative industries. If Warner’s interpretation of the settlement is true, Suno has largely caved in that fight.
But Suno never uses the word “license.” The new model was developed “in partnership with the music industry,” the company said, but it’s left entirely unclear exactly what each partner provided to the union. And in an interview with Billboard, Suno chief product officer Jack Brody explicitly said that the money flowing to the music industry would not be in exchange for training. That makes sense, since Suno is still litigating against Universal Music Group and Sony Music over this exact issue.
Then what is Suno paying Warner, BMG and Believe for? Is it merely an agreement that those companies not sue them? The right to eventually remix specific recordings? The ability to use the names and likeness of particular artists who opt into that kind of arrangement? Those questions remain unanswered for now.
Rachel Scharf: As Bill aptly pointed out, Suno’s reticence towards the word “license” may have to do with the fact that it’s still litigating against Universal and Sony. In that ongoing copyright lawsuit, Suno is continuing to argue that the fair use principle of copyright law allows it to train on unlicensed music without paying for it. Experts have told Billboard that doing exactly the opposite — paying companies like Warner and BMG to license music for training — could come back to bite Suno in the litigation. Because of that, Suno is likely being very careful about its public-facing word choices; a press release might one day be evidence in a courtroom.
Dan Rys: One of the key elements of the announcement of the new model was unrelated to the v6 models themselves, but about “launch[ing] new products that [will] open up revenue opportunities across the music ecosystem” in the future, including models that can remix and sample from popular songs. That’s something they’ll need labels, publishers and artists on board for, though it’s unclear how they will generate revenue opportunities moving forward. So these deals with Warner, BMG and Believe technically satisfy that word of partnership — though they’ll need explicit artist opt-in to do much more in the future anyway.
Related
Warner Music Group CEO Pens Memo About Suno Partnership: ‘This Is a Smart Win’
What songs exactly from WMG, BMG and Believe will be included in the training of Suno’s v6 models?
Kristin Robinson: Industry sources tell Billboard that WMG has licensed a subset of its catalog for the training of v6 and that it is up to Suno to clear any additional rights they need for those works. The sources also note that Suno is contractually obligated, according to its deal with Warner, to ensure its new models are fully licensed and trained on fully licensed content, unless a work is in the public domain.
A source close to the matter confirms that BMG will be asking their artists and songwriters for their opt-in where Suno needs rights, whether that’s in the AI training process, or with other permissions needed in future products from Suno.
Then, Believe said in its press release about the Suno deal on Tuesday (Sept. 8) that “participating Believe and TuneCore artists and labels will be included in the new music models that Suno is launching,” meaning there appears to be some level of choice for the thousands of artists and labels that run through Believe and TuneCore.
How are the v6 models trained?
Kristin Robinson: Suno CEO Mikey Shulman told Billboard On the Record it is a mix of “partner data” and “user interactions.” When asked about what “partner data” refers to, Shulman said “for the initial release it is Warner,” though a Suno representative later clarified that this also includes BMG and Believe works from the initial release. (Shulman’s interview was conducted before the Believe deal was announced.) Shulman also explained “user interactions” as “when users make a song, they will implicitly tell us whether it is good or bad — or more accurately, will give users two songs, and they will implicitly tell us one of those is better than the other.” He added, “We are able to take that signal and learn from it what makes a song particularly catchy, particularly good, and use that information about these differences between songs to steer our models toward making music better or worse, hopefully better.”
Shulman stressed that “there is no unlicensed partner data” that trained the v6 models. When asked directly to clarify that “there [are] no Universal songs, or Kobalt songs or something [else] that are training the v6 models,” Shulman replied, “Correct. No Universal songs. No Kobalt songs. No Sony songs, etc.”
Dan Rys: The key question is whether the model was trained on licensed works, or on only licensed works — and that is where Shulman stresses that both there is “no unlicensed partner data,” but also there are “user interactions,” a definition which leaves open the possibility that it was trained on past user behavior that itself was based on unlicensed works in previous models. The language matters, and it is not exactly crystal clear.
Related
Suno Inks ‘Strategic Partnership’ With Believe
How are BMG, WMG and Believe being compensated for this partnership with Suno?
Kristin Robinson: This will be a revenue share. Jack Brody, chief product officer at Suno, told Billboard that “the revenue share is going to the rights holders, and the rights holders decide how this gets paid out.” When asked to clarify if rightsholders meant Warner and BMG (Believe’s deal had not been announced at the time of this interview, either), Brody replied, “Yes.”
So, if it is up to Warner, BMG and Believe to divide up money from Suno, how much will really flow to the artists and songwriters that created the catalog, and will this new revenue stream be significant enough to be worth it to the average artist? Given these are closed-door commercial partnerships, that is currently unclear.
Rachel Scharf: Some musician advocates have made clear that they, too, would like to know what kind of upside they’ll see from Suno’s partners. The American Federation of Musicians, the union for session musicians, is suing Warner claiming its members are being kept in the dark about the flow of compensation from its Suno agreement. Warner says this will all get worked out at the collective bargaining table — but, as Kristin said, only time will tell.
Suno and Warner’s announcement said it would require artists to “opt in.” Did that mean artists got to choose whether their music was used for training Suno’s new model?
Kristin Robinson: If you look back at the WMG and Suno partnership announcement, Warner CEO Kyncl specifically promised that Warner and Suno would be “providing artists and songwriters with an opt-in for the use of their name, image, likeness, voice and compositions in new AI songs.” This never promises that artists or songwriters get the choice to opt in to AI training. Instead, it vows to provide an opt-in when an artist’s NILV rights are used — which, as Dan explained previously, is usually a right type that music companies don’t have anyway — specifically in new AI songs. WMG also says artists and songwriters can opt in for the use of their compositions (which is another word used to describe “publishing” or “songs”) in new AI songs.
Rachel Scharf: According to Billboard’s reporting, some music companies have taken the position that, legally, they do not need permission from artists and songwriters to train AI on their music due to the broad licensing rights granted in many existing record and publishing deals. In other cases, musicians signing new contracts are being presented with clauses that hand over those AI training rights from the jump. It’s still possible that Warner, BMG and Believe could have sought opt-in approval from their rosters regardless of all that, and some artists and songwriters with savvy lawyers could negotiate individual approval rights for AI training. But I’m skeptical that every song or recording in the new Suno dataset has been greenlit for that specific use by each of its human authors.
Bill Donahue: Not to belabor the point, but this entire situation is once again rooted in a fundamental disconnect between how AI companies and human artists view AI training. AI labs believe deeply that using millions of copyrighted works to create a for-profit AI model does not require permission or approval from the millions of humans who created those works. Full stop. Most artists unsurprisingly take the opposite view, and they could be forgiven for thinking that a legal settlement in a lawsuit over training would have required their sign-off to train new models. But from Suno’s perspective, it all makes perfect sense: If they don’t need approval to train in the first place, why would artists need to opt in?
Related
Is Suno the Music Industry’s Biggest Nightmare — or Greatest Hope?
Later on, Suno plans to release models that can remix, sample or otherwise create derivatives of pre-existing music. Do artists get to decide whether or not they “opt in” to this part?
Dan Rys: Yes — because artists own their rights in that regard, which would fall under Name, Image & Likeness (NIL) rights, which most people have probably heard of in the context of college sports over the past few years. Most — the key word here is most — record contracts do not include broad NIL rights, sources say, which are generally retained by artists and individuals. So, even if a record label wanted to, they probably couldn’t just opt all their artists into something like this. (Most record labels that have weighed in on this have explicitly said they would not automatically opt artists in, even if they could.) There are exceptions to every rule — all contracts are different, as label execs and lawyers like to stress — but standard deals at the moment do not contain those rights, meaning individual artists would need to agree for their names and personal characteristics to be used in future products like these.
Moreover, Suno seems to understand that they will need individual artists to opt in to build some of these products they’re talking about in the future. Again, what they want to build is unclear, but one could imagine a bespoke portal, like a video game, within which users could alter and remix an artist’s music, voice, style and more, all with the artist’s participation and consent (and payment). How many artists might want to participate in that is up in the air, but that does seem to be what Suno and its industry partners are envisioning — though a lot more rights will be needed to pull things like that off. And then, would those creations be able to be downloaded off the platform, as Suno currently allows? Or would they be restricted to a “walled garden,” as Udio’s platform will now be?
When will those future AI music remixing models launch?
Kristin Robinson: It’s unclear. Shulman told Billboard, “We will release something, at some point, in the coming months. It will evolve.”
Suno has, so far, made its business out of new AI music generation. But a lot of AI music companies have been pursuing a similar idea of AI music remixing of licensed, existing songs, which seems to be the music industry establishment’s current preferred vision of the future of AI music. That’s likely because companies like Suno, creating new songs with generative AI, are basically creating music at a scale human creators can’t keep up with and that adds a lot of competition for listeners’ time. Still, despite Spotify, Udio, KLAY, Hook — and as of this week Suno and ElevenLabs — all announcing plans to pursue some form of AI music remixing, only a few have actually launched their products. It will be worth watching how long it takes to get to these products, how differentiated each of these companies will end up being, and then asking the multi-million dollar — if not multi-billion dollar — question: do users want this type of product?