IMPF and IMPEL, which both represent independent music publishers, have published a framework that it wants the industry to adopt when licensing generative AI platforms. It leads with one big “non-negotiable” principle: “the song must be properly valued and - absent other salient factors - at least equally with the recording”.
Which means when companies secure licences to use existing music in their AI training and products, licence payments for the two copyrights contained in each track used - the recording (or master) right and the song (or composition) right - should be more or less equal. Which is different than with streaming, where the recording has always received significantly more than the song.
In a statement, indie publisher trade group IMPF and digital licensing body IMPEL say, “after years of addressing unequal revenue splits in streaming, the independent publishing community sees generative AI as an opportunity to establish a fairer framework, not another market in which outdated assumptions about the value of masters and compositions are carried forward”.
“Whilst AI generation can take place in various forms that must be treated on their merits”, they add, any AI licensing model that begins by treating the song “as a minor share of value” while allocating “the overwhelming majority of revenues to master rights” is simply “unacceptable”.
With generative AI, they continue, “songs, lyrics, melodies and compositions are central to the training, prompting, generation, output and commercialisation of AI music products”. And “generative AI licensing must reflect the distinct and essential contribution of songwriters and publishers”.
Music publishers are often one-step removed from the licensing of digital platforms, including AI products and services, because - with a lot of the industry’s songs repertoire - it’s collecting societies rather than publishers that negotiate the licences. Once a deal is done, the society administers the licence and passes through the money it generates to writers and publishers, minus any admin fees.
Even where publishers are directly involved in digital licensing, which is common with Anglo-American repertoire, technology companies have a long history of sorting out deals with record companies and the majors first, with those deals often dictating how each digital platform’s business model works, and how money will flow into the music industry.
The majors are involved in both recordings and publishing, of course, and many of their AI deals have involved both sides of the business. Which in theory means both recording rights and song rights have been factored in as the majors and AI companies discuss new business models.
However, it’s long been argued that the majors have a vested interest in more money flowing through their labels than their publishers, because with publishing deals most of the money has to be paid to the writer, whereas with conventional record deals artists usually receive a minority share.
The major publishers have always strongly denied that that impacts on what share of the money publishing receives, but not everyone in the indie sector is convinced.
Both IMPF and organisations representing songwriters have been very vocal for some time now on the need to ensure songs and recordings are valued equally in AI deals, though with new AI deals now being announced on a regular basis, the need to ensure that happens is increasingly urgent.
IMFP and IMPEL add that “the independent music publishing community is ready to engage constructively with AI companies, technology providers and licensing partners, but emerging AI business models must not repeat or deepen the structural imbalances that have shaped music publishing in the digital market”.
There are various complexities to the music industry’s deals with the AI companies, especially as the majors are pressuring their AI partners to prioritise remix and mash up tools that use existing tracks in the output, and which possibly separately exploit an artist’s voice.
Which means there will be at least two income streams - one for the use of music in training an underlying AI model and one for the use of existing music in the output. And the way the money is shared may be different on each of those income streams, especially where an artist’s voice is being exploited alongside existing recordings and songs.
Then there’s the question of how basic training income is allocated to individual works - is it split equally across the catalogue or will data be used to identify what each work is due? And if so what data?
And will there be additional admin costs in processing the money? Plus some of the initial AI deals will also include legal settlements where AI companies have already trained models with large quantities of music without getting permission from the music industry. How will that money be distributed?
With that in mind, the IMPF/IMPEL framework also talks about the importance of transparency. It says, “rightsholders must understand what is being licensed, how works are being used, how attribution will operate, what deductions are being applied, and how revenues will be calculated and distributed”.
Getting more specific, it adds that “payments for past use, training, outputs and future exploitation must be clearly distinguished and fairly treated", and “any deductions, costs or revenue calculations must be transparent, justified and applied fairly”.
Commenting on the new framework, IMPF President Annette Barrett says, “IMPF exists to give independent publishers and the songwriters we represent a collective voice, a mission that has never mattered more, as our industry navigates disruptive change and new rules are being written. The song needs protecting as the keystone of it all”.