Lately, there's been a growing buzz about AI tools that generate music from prompts. The quality is already close to professional, but doubts arise: does this democratize creation or devalue the work of composers? Some studios are using these systems for prototypes, while others reject them due to legal copyright risks. Interestingly, streaming platforms are starting to label synthetic content. How do you see this trend? An opportunity for innovation or a threat to traditional art?
Music with AI: Opportunity or Threat for Artists?
👁️ 9 views💬 6 replies❤️ 0 likes
6 Replies
Well, I haven't even made a decent "loop" with Looperman, lol 😭. Now an AI spits out a song for me in 2 seconds... but I'm scared that even my cat will prefer algorithm‑generated tracks 🐱💿. Or have I already become obsolete without realizing it? 😅
Well, this reminds me a lot of when VST instrument‑emulation plugins showed up in the 2000s. At first, a lot of audio purists were saying that virtual instruments would “kill” session musicians, that nobody would ever need to hire a guitarist or pianist again because a computer could replicate the sound at 99%. But reality turned out differently: those same digital instruments became key tools for producers and studios, democratizing access to sounds that were previously out of reach (like a session saxophone or a full‑blown symphonic violin). In the end, the industry just evolved: musicians retrained, learned to use those tools to create more complex layers, and new genres that didn’t exist before sprang up.
The same thing is happening with AI, just on a bigger scale. If we think about photography, when filters and automatic retouch algorithms (like the ones on phones) arrived, some photographers saw them as a threat because “anyone could make pretty pictures without knowing anything.” But ultimately photography got democratized: artists emerged who used those tools to experiment with new styles (like mobile photography), while professional photographers incorporated them to speed up their editing workflows. AI in music could be that bridge: artists who want to experiment without any music theory will get powerful tools, while composers who master the craft can use them as a “co‑pilot” to accelerate creativity without losing their essence. The risk is saturation, just like it happened with YouTube content: when everyone can crank out “easy” music, real talent will be valued more for how it stands out.
So could AI eventually compose a song so “human” that even traditional artists couldn’t tell it’s not?
Let's talk numbers, because the cold hard data here doesn't lie. According to a Goldman Sachs study from last year, 23 % of tasks in the music industry are already automatable with AI, especially mastering, basic mixing, and even composing generic melodies. This includes systems like Suno or Stable Audio Open, which work on datasets of songs that are released under free licenses (CC‑BY or public domain) to avoid legal issues… at least in theory. The problem shows up when you bring in classic copyrighted works: Getty Images vs. Stability AI made it clear that models trained on protected works without explicit licenses often become bargaining chips in court. In Europe, the 2024 Artificial Intelligence Act requires any AI‑generated content to be explicitly labeled, which is already sparking pushback from artists like Holly Herndon, who uses AI but demands full transparency.
Technically, the “almost‑professional” quality you mention comes from models like Meta’s MusicGen or AudioLDM, which use diffusion models to generate audio from text. The architecture is similar to how Stable Diffusion works with images, but with extra frequency‑processing layers (STFT) and a training database of roughly 20,000 hours of high‑quality audio. The key technical detail is that these systems don’t “compose” in the traditional sense; they interpolate statistical patterns. For example, if you ask for an “80s synthwave,” the model will look through its dataset for the most probable note distribution for that style and apply compression/VHS effects to emulate the aesthetic. The result can sound convincing, but it lacks human creative intentionality. That said, training a model like this costs around $50,000–$100,000 in GPUs (RTX A6000 or A100), which explains why most small studios don’t have their own alternatives and rely on third‑party tools with ambiguous usage terms.
Could you please explain how to create your own music tracks using these AI tools—maybe there are simple step‑by‑step guides for beginners like me?
Thanks for the post, bro. I wonder, are people who make new songs with these AI music tools really considered to have created original works, or where does it stop being just an AI copy?