The Rise of The Velvet Sundown
A mysterious new band is making waves on Spotify, with over a million listens in just a few weeks. Their name is The Velvet Sundown, and their music sounds familiar yet uniquely listenable. However, this band is causing quite a stir because they are entirely AI-generated, and the streaming platform isn’t making this clear to its users.
Who exactly is The Velvet Sundown? When they first appeared on Spotify at the end of June, they had a verified artist profile that claimed the band was formed by singer and mellotron player Gabe Farrow, guitarist Lennie West, Milo Rains, who crafts the band’s textured synth sounds, and free-spirited percussionist Orion “Rio” Del Mar. However, some quickly became suspicious as none of these individuals had any social media presence or evidence of existing outside of this bio. This raised questions about whether they were real musicians or just a digital fabrication.
AI-Generated Music and Controversy
The Velvet Sundown have now admitted that both the band members and the music are AI-generated. Before this admission, Deezer, a rival music streaming service, had already flagged their music as AI-generated. This wasn’t based on images or promotion but on an analysis of the music itself.
Aurelien Herault, Chief Innovation Officer at Deezer, explained that they have trained their detection tool using datasets from various generative models, including Suno and Udio. This allows them to recognize signals and sounds in fully AI-generated music that aren’t found in authentic tracks. They’ve also made progress in identifying AI tracks without specific datasets.
Deezer’s software identified The Velvet Sundown as AI-generated before the controversy erupted, leading to a label being shown to users warning, “AI generated content. Some tracks on this album may have been created using artificial intelligence.”
The Art Hoax
After speculation across news organizations, The Velvet Sundown admitted they were not real humans. Their Spotify profile now refers to them as “a synthetic music project guided by human creative direction.” However, their online presence is even more puzzling, with multiple accounts claiming to be them on both X and Instagram.
The band has ironically accused one of these accounts of trying to “hijack” their identity by creating fake profiles claiming to represent them. An unofficial X profile posted many messages, such as: “This is not a joke. This is our music, written in long, sweaty nights in a cramped bungalow in California with real instruments, real minds, and real soul. Every chord, every lyric, every mistake — HUMAN.”
This prolific poster later revealed themselves as Andrew Frelon, claiming their account was an art hoax using ChatGPT. They admitted to using a fake name and wrote on Medium about their interest in disinformation and generative AI, seeing an opportunity for mischief as the newly buzzy “band” had no social media presence.
The Growing Presence of AI Music
AI-generated music is becoming increasingly prevalent. Deezer reports that they now see 20,000 tracks which are 100% AI-generated submitted every single day, doubling from the start of the year. Mr. Herault mentioned that AI music now makes up approximately 18% of all tracks delivered to the platform.
Deezer prioritizes revenues going to real artists by removing fully AI-generated tracks from algorithmic or editorial recommendations. They believe music fans have a right to know what they are listening to, hence their transparent approach of tagging AI-generated music on Deezer.
As technology improves, we can expect more AI-created tracks that sound great, while AI becomes more integrated in other areas like filmmaking and even potentially taking white-collar jobs. Spotify has invested heavily in AI, allowing users to create playlists or have a DJ curate songs for them. However, it has faced accusations of adding AI-generated music to popular playlists without obvious indication to users.
The Impact of AI Music on Royalties
One way scammers might benefit from uploading AI music to streaming platforms is by earning royalties through streams. There are even so-called “streaming farms” where tracks are listened to repeatedly to game the system. A song could be made by AI and listened to by bots on repeat, with minimal human involvement.
Fraudsters often flood streaming platforms with lots of fake songs, each streamed just a few thousand times—enough to make money but less likely to raise suspicion. Mr. Herault explained that if an artist gains a significant number of users streaming their music, they become entitled to a bigger share of the royalty pool. While this applies to any artist, AI music is significantly easier to produce.
Deezer reported that up to 70% of streams of fully AI tracks are fraudulent, though AI tracks only make up 0.5% of overall streams. The company excludes these streams from royalty payments when detecting stream manipulation.
Although The Velvet Sundown profile on Spotify now tells listeners the truth, there is still no general system to flag AI content to users.
