The Velvet Sundown and the Rise of AI Music: Innovation or Infringement?

In early July 2025, a mysterious new band called The Velvet Sundown burst onto Spotify with dreamy, lo-fi psych-rock vibes and a pair of albums that quickly attracted more than a million monthly listeners. As it turns out, however, The Velvet Sundown isn’t a band at all. It’s an entirely AI-generated music project. No humans performed, sang, or even existed behind the album covers. Everything—from the music and lyrics to the band photos—was created by artificial intelligence. The project has sparked a wave of fascination, frustration, and concern across the music industry and beyond. It also raises a larger question: Is this creative evolution—or a form of digital piracy?

How AI Makes Music Like The Velvet Sundown

Artificial intelligence can now write and produce full-length songs that sound remarkably close to what a human artist might compose. Platforms like Suno, Udio, and others use machine learning to generate instrumental tracks, melodies, lyrics, and even synthesized vocals. These models are often trained on massive libraries of existing music, learning chord progressions, stylistic elements, vocal phrasing, and rhythmic structures from real songs.

In the case of The Velvet Sundown, the creators appear to have used a combination of these tools to simulate an entire band—complete with fictional names, album artwork, and a digital persona convincing enough to fool listeners and Spotify’s algorithms alike. With just a few genre-based prompts and a handful of keywords, users can guide the AI to generate entire songs that mimic the sonic identity of real artists or genres.

And the results? Well, whether you like the music or not, it certainly is believable and not obviously AI generated. While the genre is not for me personally, I found it all very believable. That’s part of what makes the situation so complex.

The Legal Gray Zone: Copyright, Fair Use, and Training Data

While AI-generated music may be an impressive technological feat, it lives in murky legal territory—especially when it comes to copyright.

Most generative music models are trained on copyrighted material. That means the AI absorbs and learns from thousands (or millions) of songs without the consent of the artists or rights holders. While the final outputs may not copy any one track directly, they are built on the creative labor of others.

This raises the question: Is this fair use or theft?

Currently, U.S. copyright law does not recognize works produced entirely by machines as eligible for copyright protection. Only human-authored content qualifies. So, while AI-generated songs might not be protected by copyright themselves, they may still infringe on the copyrights of the materials they were trained on.

There’s no definitive legal framework yet to resolve this. The courts have not clearly ruled on whether using copyrighted content to train AI models without a license constitutes infringement. Some argue it’s transformative and legal under fair use. After all, don’t all musicians essentially “learn” from (or get inspired buy) other copyrighted works, the legal issue being whether it is an actual “copy” of the work? My guess is that if The Velvet Sundown were a band of humans, there would be no copyright discussion around their music—while it is certainly derivative of a genre and similar bands, they do not appear to have directly “copied” any particular artist’s work. If that is the case, the question seems to be whether AI created works (or the AI “learning” process that enables the creation of purely AI works) needs to be held to a different standard than humans for copyright purposes, or be band or restricted for other policy reasons.

The Economic Threat to Human Musicians

One of those policy issues is the growing concern about the economic impact of AI music. Human artists already struggle to make a living in the streaming economy, where payouts are fractions of a cent per play. If AI-generated songs start dominating playlists and siphoning off streams, real musicians could be pushed even further to the margins.

In the case of The Velvet Sundown, millions of streams—and the accompanying royalties—went to a project with no real musicians at all. Estimates suggest AI music could displace more than 20% of musician income within the next three years. And unless streaming platforms begin labeling AI-generated content, most listeners won’t even realize they’re consuming synthetic music.

Should AI Music Be Disclosed?

Right now, major platforms like Spotify and Apple Music do not require creators to disclose whether music was made by AI. This lack of transparency fuels a deeper ethical debate about authenticity and consumer trust.

Deezer, a French online music streaming service is one of the few exceptions, and has begun actively identifying and labeling AI music. But in most cases, there’s no way for listeners to know whether what they’re hearing is the work of a human artist or a digital ghost.

As AI generative tools become more sophisticated—and more widespread—many are calling for greater regulation and disclosure. Should there be a clear label when music is AI-generated? Should streaming platforms limit or separate AI content? Should the artists whose work trains these models be compensated?

The Bigger Picture

The Velvet Sundown isn’t just a viral curiosity—it’s a symptom of a fast-approaching shift in how music is made, distributed, and monetized. It forces us to confront some hard questions:

• What does it mean to be an artist in the age of AI?

• Who owns creative output made by machines?

• Can you copyright something that has no human author?

• Should synthetic art be treated differently from human-created works?

We don’t have the answers yet. But if the case of The Velvet Sundown proves anything, it’s that these conversations can no longer be theoretical. AI is here. It’s making music. And it’s already competing—directly—with human artists.

Whether that’s progress or peril may depend on what we do next.