The EleonORe Song

As everybody knows, a song comprises three essential elements: the lyrics, the music, and the vocals. While many songwriters traditionally begin by establishing the chord progression or melody before crafting lyrics and vocal melodies around it, I opted for a different approach. Drawing inspiration from Elton John’s early catalogue, where lyrics were penned first—perhaps a sign for the success of my song—I decided to start directly with the lyrics using ChatGPT. Here’s the prompt I used:

The prompt itself is nothing special, merely giving the general direction with some specific mentions to better describe the idea of the envisioned song. Here is a sample of the lyrics from the chorus and the second verse: 

It’s evident from these eight lines that ChatGPT understood the task in its entirety, crafting lyrics that are easy to digest with a smooth flow and transitions and a simple rhyme scheme — exactly what a pop song needs to be successful.

Moving on to the musical composition, Suno took care of everything, including the melody, harmony, rhythm, and instrumentation. I chose not to provide a detailed prompt, simply asking for an energetic disco/pop banger. The result speaks for itself… I created the greatest hit of summer 2024!

Jokes aside, the situation is quite alarming. A song of this quality took less than 10 minutes to create, whereas a traditional songwriting process might take weeks or even months. This rapid turnaround raises several concerns, from potential copyright infringement arising from the creation and use of third-party works to issues of copyrightability, authorship, and even image rights infringement.

Every creative process begins with learning, and learning to a large extent involves imitation. As Aristotle noted in Poetics, humans are naturally imitative creatures, making their first steps through imitation. But does something that holds true for humans apply to machines as well?

Indeed, for an AI system to generate this kind of outputs, it must be trained on vast amounts of input data—in this case, audio data. The success of algorithmic music generation relies heavily on audio mining of existing repertoires and catalogues. By exploring Suno’s list of generated tracks, or even just from this one alone, it’s clear that the system is trained on contemporary music genres (Vivaldi’s work likely holds limited value for the target audience). This hasn’t gone unnoticed: RIAA, the Recording Industry Association of America, has already moved against Suno and Udio, alleging the mass infringement of copyrighted sound recordings.

Despite the fact that text and data mining exceptions have been introduced in multiple jurisdictions across the globe to enhance innovation, the current situation shows that those exceptions alone are surrounded by ambiguities, including having regard to their scope and their subsequent impact on the copyright infrastructure when applied for generative purposes.

In the EU, the adoption of the AI Act (Article 53(1)(c)) has seemingly clarified the applicability of the text and data mining exception of Article 4 of the DSM Directive to generative AI. Of course, this doesn’t mean AI developers have a free pass to feed an entire internet’s worth of music to their systems without considering reservation of rights and also the three-step test. (For more on that, check Eleonora’s article on The three-step test and this Kat’s article on AI covers) 

Innovation, in this context, needs to be understood lato sensu, encompassing a wide range of creative endeavours that go beyond mere technical advancements. Balancing this type of innovation with creativity is essential to ensure that innovation doesn’t trample over the rights of authors and rightsholders, but rather works in harmony with them, creating a fertile ground where creativity and progress are not mutually exclusive but mutually reinforcing.

While “The Eleonore song” is an instant hit that will echo through every beach party along the Mediterranean and beyond, it also elicits reflections on legal issues well beyond the input/training phase.

For example: Are prompts sufficient to make one an author of the AI output? Is a generic description of the idea enough for me to be considered the creator? Who is making the creative choices in the end?

What about image rights? Since voice is not directly protected under copyright and the current image rights regime may not be exhaustive, does this imply that legal remedies for the mining of voice data to generate synthetic voices are lacking?

Equally important, how does algorithmic music impact the consumption of music when anyone can create personalized songs or entire playlists on the spot? Is it a real threat to the music industry or just a new business model?

Stay tuned as I release more tracks from my first EP, “CringeFest”, aiming at conquering the pop music scene while voicing my concerns about these issues.