Suno, Dope ChatGPT for Music, Is Out But Here’s Why It Might Be Problematic
It’s not that the world needs even more AI text-to-music tools, but they keep emerging anyway — some are better, some are worse. Suno is probably among the most popular “ChatGPTs” for music at the moment. And honestly, among the best ones. But if there are so many, why has Suno gained that much attention in the last couple of weeks?
We won’t review the tool in this article, nor will we showcase how good it is. It is good, and that’s the problem. Instead, let’s try to figure out if Suno might be a threat for human artists.
What is Suno?
As reported by Rolling Stone, Suno was founded nearly two years ago by Mikey Shulman, Keenan Freyberg, Georg Kucsko, and Martin Camacho, machine-learning experts who worked together until 2022 at a Cambridge company, Kensho Technologies, which focused on finding AI solutions to complex business problems.
Suno is a web-based AI-powered music generation tool. It’s different from other AI text-to-music solutions, though, because it can create an entire song with a single prompt. Yes, a two-minute song, using large language models and diffusion models to generate original music, lyrics, and vocals based on user inputs, using the same general approach as ChatGPT. You can simply enter specific details about the desired genre, mood, and topic of the song, and Suno will attempt to generate something matching those parameters. However, Suno’s music is still limited in its ability to be edited or customised, similar to how Stable Diffusion works for images.
With a free account, users can generate up to ten songs per day, while paid plans offer more credits for creating more songs. The free version doesn’t allow commercial use of the generated songs, but paid plans do.
“We are not here to make more Fake Drakes”
What makes this tool stand out is that there’s unironic excitement around Suno’s potential to disrupt the music industry, with some predicting it could lead to the “end of the digital artist and rise of live performers.”
In mid-March, Suno released v3, its first model “capable of producing radio-quality music.” And it does truly make high-quality songs, some of which sound almost indistinguishable from mediocre human-made tracks.
According to Suno’s blog post, v3 is designed “for creating original music, and [their] models don’t recognize references to other artists.”
“We are not here to make more Fake Drakes. To further protect against misuse, we have developed proprietary, inaudible watermarking technology that can detect whether a song was created using Suno,” the team shares.
It’s interesting that they pointed that out because many generative AI companies use copyright-protected data to train their models, without openly disclosing that, of course. OpenAI’s Sora AI text-to-video tool, which is also a top player in its field now, recently said that they used “publicly available” videos to train their model but didn’t go into specific details on what kind of videos that would be.
As of writing, Suno doesn’t reveal its data sources, either. Maybe they use royalty-free music for training, which is unlikely considering that the quality of music it generates is really, really good. Besides, as Rolling Stonereports, one of their investors, Antonio Rodriguez, has said that “the risk [they] had to underwrite when [they] invested in the company, because [they]’re the fat wallet that will get sued right behind these guys. “Honestly, if we had deals with labels when this company got started, I probably wouldn’t have invested in it. I think that they needed to make this product without the constraints,” Rodriguez shares with Rolling Stone.