Tools like Suno are fundamentally enabling. I'm about 40 years old and never "had the music" - not for lack of trying (music lessons at a young age)... but could never carry a tune or keep rhythm. I suppose its what being dyslexic feels like. If I were educated in a culture where music was fundamentally as important as reading or math, I suppose would have spent enough hours on it to eventually be passable... but I got frustrated, the music lessons stopped. But that doesn't mean I stopped appreciating or wanting to make music!
And then comes Suno (and OpenAI's jukebox before that), and it felt like my brain exploded... like the classic scene in a superhero movie when the power was given to me. Is my music good? No - but I spent years writing and fashioning poetry and all of a sudden can put that to music... hard to explain how awesome that feels. and i love using the tools and it's getting better and it's been fundamentally empowering. I know it's easy to say generative art is generative swill... but "learning Suno" is no different than "learning guitar".
Suno is moving toward becoming a browser-based DAW that happens to use AI. There are already more capable and established DAWs, and I see no reason they can't implement AI into their workflows-- in a more precise manner, where it's actually useful, instead of wholesale as a gimmick. Many are already doing this. So I don't understand where Suno is going with any of this.
It either needs to be:
1. So easy anyone can press a button and magically get exactly what they want with perfect accuracy and quality.
2. So robust and powerful it enables new kinds of music production and super-charges human producers.
This is neither. And I don't buy Suno's argument that they're solving a real problem here. Creative people don't hate the process of creating art-- it's the process itself and the personal expression that make it worthwhile. And listeners/consumers can tell the difference between art created with intent and soul, and a pale imitation of that.
"Creative people don't hate the process of creating art"
Yep. I was a professional music producer before the pandemics, and I couldn't agree more.
Honestly, I'm glad we are destroying every way possible to earn money with music, so we find another profession for that purpose and then we can make music for fun and love again.
> It either needs to be: 1. So easy anyone can press a button and magically get exactly what they want with perfect accuracy and quality. 2. So robust and powerful it enables new kinds of music production and super-charges human producers.
Don't forget the secret third option - facilitate a tidal wave of empty-calorie content which saturates every avenue for discovery and "wins" purely by drowning everything else out through sheer volume. We're at the point where some genAI companies are all but admitting that's their goal.
That seems to be the purpose. It doesn’t have to sound that good to the listener. It’s just made to extract dollars from Spotify when you flood the platform with so much slop that some of it starts getting played by users who just let the machine pick the next song.
Yes I'm already doing this manually with Reason. I'll compose something that's quite bare bones, export the audio and run it through Suno, asking it to cover and improvise with a specific style, then when I have something I like, I split that into stems, import some or all of these to Reason and then reconstruct and enhance the sound using instruments in Reason, mostly by replaying the parts I like on keyboard and tweaking it in the piano roll. Often I get additional inspiration just by doing that. Eventually I delete all the tracks that came from Suno stems when I've finished this process.
That way I get new musical ideas from Suno but without any trace of Suno in the final output. Suno's output, even with the v5 model, is never quite what I want anyway so this way makes most sense to me. Also it means there's no Suno audio watermarking in the final product.
This is similar to what I do. There are all kinds of useful ways to incorporate AI into the music production process. It should be treated like a collaborative partner, or any other tool/plugin.
It shouldn't be a magic button that does everything for you, removing the human element. A human consciously making decisions with intent, informed by life experience, to share a particular perspective, is what makes art art.
But then you look at image gen. The established one, namely Adobe, are surprisingly not winning the AI race.
Then you look at code gen. The established IDEs are doing even worse.
I don't rule out the possibility of music being truly special, but the idea of "established tools can just easily integrate AI right" isn't universally true.
Agreed. The problem with being an incumbent in this era is that much of the existing UI/UX assumptions are based on the idea of manual manipulation. We're so early that foundational assumptions are still up for debate, and for large companies like Adobe, there's just no way they'd be able to move at the required pace to keep up. Heck I'm at a company that's less than 2 years old, with less than 20 people, solely devoted to AI, and it's still hard for us to keep up.
What Adobe and others ought to be doing is setting up internal labs that have free reign to explore whatever ideas they want, with no barriers or formality. I doubt any of them will do that.
The innovator's dilemma is real. IMO none of the big DAWs are well-positioned to capitalize on AI, but that doesn't mean they couldn't.
I'd argue music generation is different from image or code generation. It's closer to being purely art. Take image generation for example. Most of the disruption is coming from competition with graphic design, marketing, creative/production processes, etc. The art world isn't up in arms about AI "art" competing with human art.
It does mean. The switch from writing “applicable” software to creating cutting edge AI is almost impossible. The parent comment makes great examples, we can add to that list JetBrains (amazing IDEs, zero ability to catch up with ML), for example. It’s a very different fast-paced scientific driven domain.
> Creative people don't hate the process of creating art
I mean, I hate when it's difficult to get the medium to express my vision... not that AI especially would help with that when I'm actually attached to that vision in detail....
Meanwhile I've mixed a song of mine down on a Tascam 688 8-track tape recorder. I have a big smile on my face because I find this very enjoyable. The haptics, the sound and my hand-crafted product. A piece of art made by a human. No AI will replace this for me.
> Then tried to generate interesting music, failed spectacularly.
You can upload music and let suno arrange it in different styles. I'm a musician myself and am also interested in "interesting" music. I made experiments with my own music and was positively surprised by the "musicality" and creativity of the generated arrangements (https://rochus-keller.ch/?p=1350).
What's special about Suno 5 is that the songs are actually good to listen to in place of professionally produced songs. For example, my favorite genre is new jack swing and there is a very limited number of this genre as it was briefly popular during the 90s. Now I have an endless supply of it and you can't tell that its AI produced anymore. Sure to an expert they might be able to detect it but for consumers its just as good as spotify playlist.
This is the first time I'm actually paying for generated AI content because the value I get is immense. I really think we are headed towards and over supply of content where there will be more stuff to read, watch, listen with very real value in all of them.
This spells out the inevitable change in the labor market for content creators. There will always be value for human created content and some will make more money but it will always have the AI generated content generation competing with it to the point where it will be hard to stay ahead and eventually people will stop caring.
Case in point, I see some comments being snarkish towards Suno but for as a consumer I could care less if you put your soul and years into producing art vs the one I can get a lot of today and now especially when there is virtually no difference in quality.
Truly an amazing accomplishment from Suno team, and probably the first time I've subbed to a music service after decades of downloading mp3s, hunting down new songs to listen to on Youtube. Suno 'steamified" this process and while I will use youtube to discover new genre, I am spending now most of my time in Suno, listening to endless amount of the exact sound I am looking for.
> I really think we are headed towards and over supply of content where there will be more stuff to read, watch, listen with very real value in all of them.
> as a consumer I could care less if you put your soul and years into producing art vs the one I can get a lot of today and now especially when there is virtually no difference in quality
I've had some fun with Suno 5, but the songs absolutely don't replace well-made human music. They're much more formulaic and over-produced. They're usually forgettable. People I play them for can always tell they're AI produced.
True music enthusiasts will holdout for a while but I think AI music will easily replace most Pop currently on the radio and streaming for your average Joe. That stuff has been "fake" as early as the mid 2000's by being quantized straight to the grid, pitched, with programmed drums, guitar, even vocals and then churned out like widgets on a conveyor belt.
Currently for me, in the type of music I have enjoyed from v4.5+. V5 of their model is a regression.
I was very impressed with v4.5+ that I could get quite good songs evocative of Devo, Yeah Yeah Yeah's, Metric, etc.
Version 5 is currently harder (or I haven't figured out a way) to generate this kind of chopped/produced sound. It doesn't follow complex style definitions and tends to generate songs that are too slow and "smoothed" over.
I find it depends how you generate it. Asking Suno to make covers of uploaded recordings tends to give much, much better results than asking it to cook a song from scratch. There are still quite a few tells that it's AI-made but it's not bad at all, at least in my experience so far.
I simultaneously feel repulsed by AI music and "art" and yet am totally open to being captivated by AI music if I really feel something is musically better than almost any human-made music I've heard.
I just haven't heard anything that isn't "slopful" yet. If I do, I will still feel weird about it, but I'm a big believer in the value of "aesthetic objects in themselves", so I am eager to find something I do actually like.
Even just knowing something was drawn or composed by an AI will negatively taint my opinion from the start, but I'm still open.
I don't totally discount the position that the human "soul" is what makes art art and all that, but I still do think something can be very enjoyable and good without being created by a sentient entity, in theory.
Couldn't agree more. Instead of seeking out people making that art, we are now leaving "art" or human expressive emotion to random noise and paying for it.
Rather than call it a lie, I think it would be more fair to say in this instance it's a matter of taste/perspective. Personally, I enjoy listening to Suno songs.
I consider myself an amateur musician and pulled a decent side hustle teaching piano back in university. I also worked the occasional gig as a cocktail pianist.
I've actually had a lot of fun using tools like Suno/Udio as a means of sonic exploration to see how some of my older compositions would sound in different mediums.
When I composed this piece of classical music practically a decade ago, it was intended for strings but at the time I only played piano so that's where it stayed. By increasing the "Audio Influence Slider", Suno arranged it in a chamber quartet style but stayed nearly 1:1 faithful with the original in terms of melody / structure.
Well there's now a polka about diarrhea after eating spicy food. And a rock ballad duet about macaroni and cheese with a female singer and a screamo male part.
I've been working on an AI detector for the last few months. Updated it to handle Suno V5 on Wednesday - looks like it's very similar to V4.5. Am curious to see how this Studio version impacts the model I've trained.
After login it's free. But my site has been targeted by a lot of spam/abuse over the last decade, and login is something I've needed to set up to avoid that :(
I wasn't expecting to, but I got chills listening to some Suno creations from artists who are clearly very talented at using this new medium.
Much like those of us hammering away at LLMs who eventually get incredible results through persistence, people are doing the same with these other AI tools, creating in an entirely new way.
I'm sure Suno are working hard on this and these AI tools can only come together as fast as we can figure out the UX for all this stuff, but I'm holding out for when I can guide the music with specific melodies using voice or midi.
For "conventional" musicians, we (or at least I) would love to have that level of control. Often we know exactly what it should sound like, but might not have session musicians or expensive VSTs (or patience) on hand to get exactly the sound we want. Currently we make do with what we have - but this tech could allow many to take their existing productions to the next level.
Not OP, but on the off chance you haven't seen this, I found the suno explorer thing quite nice. Hitting random a few times, I'll usually stumble onto something interesting. This was the first demo I heard where some of the AI tunes gave me goosebumps close to what human music does.
I'm not the person you responded to, but these are some examples from someone I know that had accompanying music videos (actual video production) made for them:
They probably won't. And if they do probably everyone will ridicule the songs. But maybe they will link the songs and maybe at least half the repliers will say the songs actually are good. I like rooting for the underdog.
I listened to it when you posted before. Better than most of the others I have listened wich were all much more "cold".
The visual stuff also helps to make it more powerfull and cohesive.
The bad part is that it wanders a lot to get nowhere and it does not create a climax that bridges with the second part. The same sounds and ambient with a producer behind that creates an arragment for it would be much more powerful.
How is the stem support these days? In particular I would like to create some songs with vocals (my lyrics), then be able to remove the vocal track and replace it with my own.
I'm really curious how the traditional DAW features stack up against the incumbents. A good DAW has a ton of features. Developing a whole DAW from scratch just to add a "generate part" button sounds like a lot.
Okay, but why would you? What's the advantage of that? VSTs are resource hogs (when you have like 10+ running at once), not DAWs.
I somehow doubt a full-blown browser connected to more than a couple of VSTs would be less of a resource hog than doing the same in a DAW. On your computer. That you own. In your house. Without like additional 50ms of latency for the data to travel to the server and back.
I split pretty evenly between Bitwig, Logic X, and MuseScore and I can tell you the first big reservation I have with something like this is that it's an online DAW.
What does that mean? It means that your compositions (outside of bouncing them down to audio stems) exists within a highly proprietary SaaS format and that the moment you stop paying, you've got NOTHING.
99% of major DAWs (Ableton, Logic, FL Studio, Bitwig, Studio One, etc.) are a perpetual license.
The video seems to be trying to convince me that this is totally targeted at actual musicians. But I guess maybe that's how their target users imagine themselves?
As an actual musician who doesn't have any troubles creating my own melodies and timbres (both generated and 'manually' created), its very obviously not targeted at someone like me.
It seems similar to a Garage Band type of software, aiming to entice people with little audio production experience and give them an interesting sounding snippet they can play back to friends.
For example, the only actual audio editing they displayed was slicing and re-pitching (you can't even choose the time-stretch algorithm), which is conceptually very simple to understand.
There's no ability to actually edit dynamics or do very accurate frequency adjustments that I can see from the demos, so it's basically useless for anything I would want to do.
Before AI there was a general consensus that creative areas (eg. Cities) were becoming a homogenized experience. A Starbuckization if you will. I can’t help but wonder what gets lost when using tools like this.
It's unclear to me whether it will result in more homogeneity, as a result of prompts being a coarse medium that results in the AI choosing what it's seen to fill in the rest, or less homogeneity, as a result of more people with non-mainstream tastes being able to create music aligned with their niche that otherwise wouldn't exist due time/money restrictions. I think the latter seems a bit more likely, but time will tell.
There's not really any need to speculate when this has already played out in other mediums - would you say that the proliferation of LLMs has led to an explosion of novel and interesting works of fiction, or just an explosion of cookie-cutter slop ebooks?
I would say too soon to tell. There has been an uptick in ebook slop, but I'm not sure if it's impacted the homogeneity of literature, because I don't think anyone is reading ai ebooks. It's not enough for it to exist to impact culture, it has to be being consumed.
Music is a uniquely interesting case, since music has a much lower barrier of entry to consume.
My thought to who you replied to exactly. Am I going to invest several days to read an AI slop novel? No. But I will take several minutes to read a blog post and likely have read many that were AI generated or assisted.
Since you get exactly the kind of music you want, I think it leads to extremely small bubbles, which is pretty much the opposite of homogeneity.
For example, I had never heard epic power metal about birds, but with Suno I got exactly what I wanted. Sure, the sound quality (I only used v3.5) could be better and the songs could be longer, but I don’t care, I now have epic songs about my Bourke’s parakeet. However, I’m not pretentious enough to think those songs are interesting to anyone other than my wife and me, hence the smallness of the bubble.
This is an excellent example of where AI-assisted pro music production needs to go. Five months ago there was an HN post on Suno's prompt-based music authoring and I asked for exactly this.
> "I have little interest in typing a text prompt and getting a complete song as an output. However, I will gladly pay serious money for a tool that interactively collaborates with me in a granular, iterative process of generating, adjusting and mixing individual instruments and sections toward a finished multi-track song project." https://news.ycombinator.com/item?id=43873781
It's clearly early beta and some key elements still aren't there - like optionally getting dry tracks (no effects) and MIDI output for generated solo instrument tracks - but Suno is absolutely headed in the right direction. So... congrats to the team. I'm excited! Although I recently retired and no longer have a studio, big budget or session players - I also no longer have the client, the client's "wife with great ideas" and a hard deadline. But I've still got passion, experience and a good ear along with more time to explore creatively interesting directions. Vibe coding is fun but extrapolating forward from the Suno Studio demo video, it's clear Vibe Producing has the potential to be amazing.
ASKS: I'd also love to see the ability to collaboratively evolve one aspect of a generated instrument track while holding most other aspects constant. This is how we'd work in the studio with session players. Stuff like "That's almost it... but give me a few variations with a trill at the end like this clip but played with the sultry sound of the trumpet in this clip."
CONCERNS: My concerns are similar to the gaps experienced devs report with coding AIs being like "a junior developer with a bad memory and a parakeet's focus." If Suno Studio's AI is granular enough to collaboratively evolve its generated tracks over several iterations without losing sight of what's good about them, then it'll be productively useful. The only other significant concern is Suno Studio being it's own web-based DAW. A real DAW like Cubase or Bitwig is Hard. A browser-based tool is going to quickly run into thorny challenges around sample-accurate, jitter-free audio I/O, multi-track disk buffering, real-time effects, sub-mixes, ASIO plugins, etc. And none of that intractable mess of heavy-lifting is required for Suno to deliver their core value prop.
STRATEGY: If I was having a beer with Suno Studio's PM Henry Phipps, I'd be encouraging him to either partner with or buy a mature, cross-platform DAW that's already solved those problems. The "buy" strategy has obvious benefits in quickly side stepping all the challenges of browser-based pro audio tools as well as securing an installed base of experienced producers used to paying real money for real tools. The partnering strategy seems cheaper and easier but, in my experience, doing it well is harder. I'd suggest working with one leading DAW but developing an open API to enable seamless workflows between traditional DAWs and AI services. Such an API would let Suno focus on what Suno does best while letting DAW makers avoid having to become competitively competent in neural nets, licensing tagged training data and running cloud GPU services profitably. No DAW developer has the capital (or access to capital) to seriously compete in those domains. Whether through acquisition or partnership, the long-term silver-lining for Suno may be gaining deep experience in digital media production tooling, workflows and customer mindsets.
While I like using AI for assisting with repetitive programming, I can’t help but feel sorry for my producer and illustrator friends who are now having to compete with generated tools.
Is it snobby of me to look down upon art that is created using these tools as lesser because the human did not make every tiny decision going into a peice? That a persons taste and talent is no longer fully used to produce something and for someone reason to me what is what makes the art impressive and meaningful?
Something about art with imperfections still feels exciting, maybe even more so than if I see something that is perfect but if I see an AI gen picture with 6 fingers, I just write it all off as slop.
I am happy to allow my generated code to come from “training data” but I see the use of AI in art, writing and music as using stolen artists hard work.
I feel like as time goes on, I feel even more conflicted about it all.
Applying your logic, did you feel bad for seamstresses when industrial revolution took off? did you feel bad for hardware manufacturers in America when they were outsourced to China? Art is also a form of labor and whoever can produce quality at quantity wins. Idealizing art in some sort of religious idolation is just plain silly. We haven't had the Picassos or Mozarts or Oscar Peterson for quite some time now yet the world is just fine. People play playlists in front of millions of live crowd and get accolade for it vs real instruments. Times change, technology change and art changes.
You either adapt or go hungry just like everybody else and art shouldn't be exempt from the mechanics of supply and demand.
I almost agree with you that this is about quality, but I still feel that the context in which art comes from influences how I perceive it.
Take, for example, a track by Fontaines D.C., a band from Ireland that writes extensively about the lived social and political experience.
Knowing where they are from and the general themes of their work makes their tracks feel authentic, and you can appreciate the worldview they have and the time spent producing the art, even if it does not align with your own tastes.
Trying to create something of the same themes and quality from a prompt of “make me an Irish pop rock track about growing up in the country” suddenly misses any authenticity.
Maybe this is what I am trying to get at, but like I said, I feel some conflict about this, as I personally value these tools for productivity
You as a human chose to write this very common opinion and even include writing errors like the following
> That a persons taste and talent is no longer fully used to produce something and for someone reason to me what is what makes the art impressive and meaningful?
Human output isn't sacred. yes this is snobbery, a useless feeling of superiority.
I feel the same, including code. I cannot justify it. I can easily counter my own arguments. Still, the further we automate human thought and creativity the worse it makes me feel. I am disappointed that so many are content with mediocre imitation.
As someone who's both a software engineer and music producer its been really interesting to watch the parallel progression of the AI advancements in both areas, as well as my own sentiments towards those advancements.
I've always been pretty bullish about using LLMs to help with coding. My comment history here should verify that. I just thought it was insanley cool tech, and I found I was rediscovering the joy of programming when I could delegate out all the tedious crap I could never get myself to do. Side projects that had languished for years were finally getting somewhere near "done". I also found in many cases it was actually smarter than me, finding cleaner and more elegant solutions to problems than I ever would have. And that was great!
The first time I played with Suno though, it was all so different. I felt deflated. What was the point in making music if a robot could do it better? Yes, the first models were crap, but the writing was on the wall. Music is all about pattern recognition and repetition after all, why wouldn't the robots be great at it? I suddenly knew how all those visual artists felt once the image models started rapidly improving.
So it's been interesting try to disect why I thought coding models were great, and creative models... well, depressing.
Part of it in my case was because being that guy who made cool music was always part of my identity, a part of me I was proud of. Coding, not so much. Don't get me wrong, I've had some pretty big career achievements that I'm very proud of, and I love coding. But at the end of day, the lines of code written were just a means to an end to making something cool.
Which brings me to the other reason I think I was so much more positive about LLMs: the application I was bringing to life with the help of an LLM still very much felt like "mine". Yes, maybe the tedium of writing a bunch of boilerplate was being delegated, but the idea, the architecture, the UX, were still all mine. So I was building something I could still feel proud of.
But typing a prompt into a box and getting a song back? Nah, that's not really mine. It's no different than shouting an idea to a musical improv musician and getting a song back. Maybe you gave them a cool idea, but the song isn't yours.
Which brings me full circle back to this new Suno DAW announcement: this is absolutely incredible. I've only skimmed the announcement so far, but I feel like this brings AI song generation firmly back into the court of how I'm using LLMs to code: letting AI take care of the boring shit, and letting me focus on the composition.
I've only really ever produced electronic music of various varieties, and I have so many uncompleted songs whithering on the vine because I wasted a week of time flipping endlessly through patches, tweaking them, trying to find "that sound" in my head that I could never bring to life, then eventually getting sick of the song and saying fuck it.
And in my experiments with Suno, I found it was actually crazy good at matching a "sound" I described, I just wished I had a way to compose my own song based on the sounds it generated. And now here it is.
It's human nature to want to feel like we've accomplished something. AI generaters like Suno, where all you have to do is type in a prompt and you get the final result, take that sense of accomplishment away from us.
However, if we start working on a project where we're assisted by AI, for example, we're making a game where the sprites are generated by AI or the background music is generated by AI, but the overall game is still directed by humans, that sense of accomplishment stays.
But at some point we're going to reach the stage where the entire game can be generated in high quality, at the same level as humans. What then?
Tools like Suno are fundamentally enabling. I'm about 40 years old and never "had the music" - not for lack of trying (music lessons at a young age)... but could never carry a tune or keep rhythm. I suppose its what being dyslexic feels like. If I were educated in a culture where music was fundamentally as important as reading or math, I suppose would have spent enough hours on it to eventually be passable... but I got frustrated, the music lessons stopped. But that doesn't mean I stopped appreciating or wanting to make music!
And then comes Suno (and OpenAI's jukebox before that), and it felt like my brain exploded... like the classic scene in a superhero movie when the power was given to me. Is my music good? No - but I spent years writing and fashioning poetry and all of a sudden can put that to music... hard to explain how awesome that feels. and i love using the tools and it's getting better and it's been fundamentally empowering. I know it's easy to say generative art is generative swill... but "learning Suno" is no different than "learning guitar".
https://open.spotify.com/album/45CY60A8GCHxBQb7DCJsIl
https://songxytr.substack.com/p/what-is-a-songxytr
Suno is moving toward becoming a browser-based DAW that happens to use AI. There are already more capable and established DAWs, and I see no reason they can't implement AI into their workflows-- in a more precise manner, where it's actually useful, instead of wholesale as a gimmick. Many are already doing this. So I don't understand where Suno is going with any of this.
It either needs to be: 1. So easy anyone can press a button and magically get exactly what they want with perfect accuracy and quality. 2. So robust and powerful it enables new kinds of music production and super-charges human producers.
This is neither. And I don't buy Suno's argument that they're solving a real problem here. Creative people don't hate the process of creating art-- it's the process itself and the personal expression that make it worthwhile. And listeners/consumers can tell the difference between art created with intent and soul, and a pale imitation of that.
"Creative people don't hate the process of creating art"
Yep. I was a professional music producer before the pandemics, and I couldn't agree more.
Honestly, I'm glad we are destroying every way possible to earn money with music, so we find another profession for that purpose and then we can make music for fun and love again.
> It either needs to be: 1. So easy anyone can press a button and magically get exactly what they want with perfect accuracy and quality. 2. So robust and powerful it enables new kinds of music production and super-charges human producers.
Don't forget the secret third option - facilitate a tidal wave of empty-calorie content which saturates every avenue for discovery and "wins" purely by drowning everything else out through sheer volume. We're at the point where some genAI companies are all but admitting that's their goal.
https://www.hollywoodreporter.com/business/digital/ai-podcas...
That seems to be the purpose. It doesn’t have to sound that good to the listener. It’s just made to extract dollars from Spotify when you flood the platform with so much slop that some of it starts getting played by users who just let the machine pick the next song.
Yes I'm already doing this manually with Reason. I'll compose something that's quite bare bones, export the audio and run it through Suno, asking it to cover and improvise with a specific style, then when I have something I like, I split that into stems, import some or all of these to Reason and then reconstruct and enhance the sound using instruments in Reason, mostly by replaying the parts I like on keyboard and tweaking it in the piano roll. Often I get additional inspiration just by doing that. Eventually I delete all the tracks that came from Suno stems when I've finished this process.
That way I get new musical ideas from Suno but without any trace of Suno in the final output. Suno's output, even with the v5 model, is never quite what I want anyway so this way makes most sense to me. Also it means there's no Suno audio watermarking in the final product.
This is similar to what I do. There are all kinds of useful ways to incorporate AI into the music production process. It should be treated like a collaborative partner, or any other tool/plugin.
It shouldn't be a magic button that does everything for you, removing the human element. A human consciously making decisions with intent, informed by life experience, to share a particular perspective, is what makes art art.
That's the same process as AI-assisted coding. Or AI-assisted writing. Or AI-assisted anything.
All you said is very reasonable.
But then you look at image gen. The established one, namely Adobe, are surprisingly not winning the AI race.
Then you look at code gen. The established IDEs are doing even worse.
I don't rule out the possibility of music being truly special, but the idea of "established tools can just easily integrate AI right" isn't universally true.
Agreed. The problem with being an incumbent in this era is that much of the existing UI/UX assumptions are based on the idea of manual manipulation. We're so early that foundational assumptions are still up for debate, and for large companies like Adobe, there's just no way they'd be able to move at the required pace to keep up. Heck I'm at a company that's less than 2 years old, with less than 20 people, solely devoted to AI, and it's still hard for us to keep up.
What Adobe and others ought to be doing is setting up internal labs that have free reign to explore whatever ideas they want, with no barriers or formality. I doubt any of them will do that.
The innovator's dilemma is real. IMO none of the big DAWs are well-positioned to capitalize on AI, but that doesn't mean they couldn't.
I'd argue music generation is different from image or code generation. It's closer to being purely art. Take image generation for example. Most of the disruption is coming from competition with graphic design, marketing, creative/production processes, etc. The art world isn't up in arms about AI "art" competing with human art.
It does mean. The switch from writing “applicable” software to creating cutting edge AI is almost impossible. The parent comment makes great examples, we can add to that list JetBrains (amazing IDEs, zero ability to catch up with ML), for example. It’s a very different fast-paced scientific driven domain.
> Creative people don't hate the process of creating art
I mean, I hate when it's difficult to get the medium to express my vision... not that AI especially would help with that when I'm actually attached to that vision in detail....
Meanwhile I've mixed a song of mine down on a Tascam 688 8-track tape recorder. I have a big smile on my face because I find this very enjoyable. The haptics, the sound and my hand-crafted product. A piece of art made by a human. No AI will replace this for me.
I'm pretty sure your creation will be more interesting than any AI-generated "art".
Suno can create catchy songs and succeed in matching genre expectations / cliches.
I've been in phases where I had output I generated with it playing in my head constantly (due to repeated listening).
The output was catchy.
Then tried to generate interesting music, failed spectacularly.
And I, among other stuff, enjoy a lot of music that people consider formulaic, abstract or straight-up boring.
What's missing in AI "art" is intent and well... creativity.
I think it will have a disrupting influence on commercial pop culture, no question.
I also wouldn't claim to be able to classify correctly whether something is AI output.
But art is something entirely different.
> Then tried to generate interesting music, failed spectacularly.
You can upload music and let suno arrange it in different styles. I'm a musician myself and am also interested in "interesting" music. I made experiments with my own music and was positively surprised by the "musicality" and creativity of the generated arrangements (https://rochus-keller.ch/?p=1350).
What's special about Suno 5 is that the songs are actually good to listen to in place of professionally produced songs. For example, my favorite genre is new jack swing and there is a very limited number of this genre as it was briefly popular during the 90s. Now I have an endless supply of it and you can't tell that its AI produced anymore. Sure to an expert they might be able to detect it but for consumers its just as good as spotify playlist.
This is the first time I'm actually paying for generated AI content because the value I get is immense. I really think we are headed towards and over supply of content where there will be more stuff to read, watch, listen with very real value in all of them.
This spells out the inevitable change in the labor market for content creators. There will always be value for human created content and some will make more money but it will always have the AI generated content generation competing with it to the point where it will be hard to stay ahead and eventually people will stop caring.
Case in point, I see some comments being snarkish towards Suno but for as a consumer I could care less if you put your soul and years into producing art vs the one I can get a lot of today and now especially when there is virtually no difference in quality.
Truly an amazing accomplishment from Suno team, and probably the first time I've subbed to a music service after decades of downloading mp3s, hunting down new songs to listen to on Youtube. Suno 'steamified" this process and while I will use youtube to discover new genre, I am spending now most of my time in Suno, listening to endless amount of the exact sound I am looking for.
> I really think we are headed towards and over supply of content where there will be more stuff to read, watch, listen with very real value in all of them.
We're already here with human created content.
> as a consumer I could care less if you put your soul and years into producing art vs the one I can get a lot of today
a quantity over quality argument with regard to art is wild.
people just start liking things if there is a lot of it
> as a consumer I could care less if you put your soul and years into producing art vs the one I can get a lot of today and now especially when there is virtually no difference in quality
as a fellow consumer I care a lot actually
I've had some fun with Suno 5, but the songs absolutely don't replace well-made human music. They're much more formulaic and over-produced. They're usually forgettable. People I play them for can always tell they're AI produced.
True music enthusiasts will holdout for a while but I think AI music will easily replace most Pop currently on the radio and streaming for your average Joe. That stuff has been "fake" as early as the mid 2000's by being quantized straight to the grid, pitched, with programmed drums, guitar, even vocals and then churned out like widgets on a conveyor belt.
Currently for me, in the type of music I have enjoyed from v4.5+. V5 of their model is a regression.
I was very impressed with v4.5+ that I could get quite good songs evocative of Devo, Yeah Yeah Yeah's, Metric, etc.
Version 5 is currently harder (or I haven't figured out a way) to generate this kind of chopped/produced sound. It doesn't follow complex style definitions and tends to generate songs that are too slow and "smoothed" over.
I find it depends how you generate it. Asking Suno to make covers of uploaded recordings tends to give much, much better results than asking it to cook a song from scratch. There are still quite a few tells that it's AI-made but it's not bad at all, at least in my experience so far.
What a dystopian and depressing outlook on the valuation and enjoyment of art. Truly hope you're an anomaly.
I simultaneously feel repulsed by AI music and "art" and yet am totally open to being captivated by AI music if I really feel something is musically better than almost any human-made music I've heard.
I just haven't heard anything that isn't "slopful" yet. If I do, I will still feel weird about it, but I'm a big believer in the value of "aesthetic objects in themselves", so I am eager to find something I do actually like.
Even just knowing something was drawn or composed by an AI will negatively taint my opinion from the start, but I'm still open.
I do love generative music. I don't care if you get your notes from a markov chain a shif register a LLM or your brain.
The problem with AI music is that is just sounds like shit.
Right. That's pretty much my stance.
I don't totally discount the position that the human "soul" is what makes art art and all that, but I still do think something can be very enjoyable and good without being created by a sentient entity, in theory.
Couldn't agree more. Instead of seeking out people making that art, we are now leaving "art" or human expressive emotion to random noise and paying for it.
What's special about Suno 5 is that the songs are actually good to listen to in place of professionally produced songs.
This is a big big lie.
Rather than call it a lie, I think it would be more fair to say in this instance it's a matter of taste/perspective. Personally, I enjoy listening to Suno songs.
Personally I enjoy listening and smelling my own farts.
I consider myself an amateur musician and pulled a decent side hustle teaching piano back in university. I also worked the occasional gig as a cocktail pianist.
I've actually had a lot of fun using tools like Suno/Udio as a means of sonic exploration to see how some of my older compositions would sound in different mediums.
When I composed this piece of classical music practically a decade ago, it was intended for strings but at the time I only played piano so that's where it stayed. By increasing the "Audio Influence Slider", Suno arranged it in a chamber quartet style but stayed nearly 1:1 faithful with the original in terms of melody / structure.
Comparison blog piece
https://mordenstar.com/blog/screwdriver-sonata
DAW stands for Digital Audio Workstation (to save anyone else who didn't know from looking it up). https://en.wikipedia.org/wiki/Digital_audio_workstation
Hard pass on a generative AI company trying to reinvent a DAW. Make VST/VSTis please.
Well there's now a polka about diarrhea after eating spicy food. And a rock ballad duet about macaroni and cheese with a female singer and a screamo male part.
Yup totally won't mess with their algorithms.
I've been working on an AI detector for the last few months. Updated it to handle Suno V5 on Wednesday - looks like it's very similar to V4.5. Am curious to see how this Studio version impacts the model I've trained.
If you want to test it, here's the link: https://www.submithub.com/ai-song-checker
If only I could use it before logging in.
After login it's free. But my site has been targeted by a lot of spam/abuse over the last decade, and login is something I've needed to set up to avoid that :(
We just can't have nice things anymore.
I wasn't expecting to, but I got chills listening to some Suno creations from artists who are clearly very talented at using this new medium.
Much like those of us hammering away at LLMs who eventually get incredible results through persistence, people are doing the same with these other AI tools, creating in an entirely new way.
I'm sure Suno are working hard on this and these AI tools can only come together as fast as we can figure out the UX for all this stuff, but I'm holding out for when I can guide the music with specific melodies using voice or midi.
For "conventional" musicians, we (or at least I) would love to have that level of control. Often we know exactly what it should sound like, but might not have session musicians or expensive VSTs (or patience) on hand to get exactly the sound we want. Currently we make do with what we have - but this tech could allow many to take their existing productions to the next level.
That sounds really interesting. Would you mind sharing some examples of such creations?
Not OP, but on the off chance you haven't seen this, I found the suno explorer thing quite nice. Hitting random a few times, I'll usually stumble onto something interesting. This was the first demo I heard where some of the AI tunes gave me goosebumps close to what human music does.
https://suno.com/explore/
I'll share something! I really enjoy this artist @kant, it's rap which isn't everyone's cup of tea but here's one of his more approachable songs.
https://suno.com/song/07ab552c-1a76-43f4-a619-21a74e774dbd
I enjoy this AI generated album https://open.spotify.com/album/6C6PJzxkHctvk1ibKM2zMx?si=YgL...
I'm not the person you responded to, but these are some examples from someone I know that had accompanying music videos (actual video production) made for them:
Sunscreen: https://youtu.be/VBaWtOHPTZw
Purple Sunset Over Lake 2: https://youtu.be/lD7rSxPncs4
These are terrible
Thanks, I'll let them know.
They will not.
They probably won't. And if they do probably everyone will ridicule the songs. But maybe they will link the songs and maybe at least half the repliers will say the songs actually are good. I like rooting for the underdog.
Suno with its $125 million VC funding is everything but an underdog my friend.
By underdog I meant the above HN commenter wanting to prove to us that the AI-created songs he likes are actually considered good by others.
I actually have listen to all the links people posted. They are not that bad.
But when they say it can replace Pop music I can only laugh. It is the most boring early 2000 RnB ever created and it souns thin.
Any Aphex Twin model out there?
https://www.youtube.com/watch?v=lD7rSxPncs4
I listened to it when you posted before. Better than most of the others I have listened wich were all much more "cold".
The visual stuff also helps to make it more powerfull and cohesive.
The bad part is that it wanders a lot to get nowhere and it does not create a climax that bridges with the second part. The same sounds and ambient with a producer behind that creates an arragment for it would be much more powerful.
How is the stem support these days? In particular I would like to create some songs with vocals (my lyrics), then be able to remove the vocal track and replace it with my own.
I don't know if Suno does it, but it should be doable.
For voice removal I use Ultimate Vocal Track Remover, is on Github.
Nice, I commented looking for this last year: https://news.ycombinator.com/item?id=40588038
Looks like the "covers" need some better instrument isolation, but this is really huge for the music industry.
I'm really curious how the traditional DAW features stack up against the incumbents. A good DAW has a ton of features. Developing a whole DAW from scratch just to add a "generate part" button sounds like a lot.
I can tell you exactly how any professional is gonna evaluate this DAW: press Ctrl+F, type "VST", see 0 results, close the page.
If they do add VST support I could see them becoming a legitimate player. Without it it's definitely just a toy.
They're never gonna let you upload and run your own executable files to their infrastructure.
That + latency with MIDI devices is why every DAW-in-a-browser is just a toy.
It's a pipe dream of course, but WebAssembly would be an ideal target for C++ VSTs for its portability and sandboxing
Didn't realize this was in a browser. That tells me all I need to know.
Coming soon: Suno Desktop (Electron version). "Look ma, no browser!"
You can get a vst to run locally and still work with the browser.
Okay, but why would you? What's the advantage of that? VSTs are resource hogs (when you have like 10+ running at once), not DAWs.
I somehow doubt a full-blown browser connected to more than a couple of VSTs would be less of a resource hog than doing the same in a DAW. On your computer. That you own. In your house. Without like additional 50ms of latency for the data to travel to the server and back.
I just assumed all the audio data would be local even in a "browser DAW", so VST calls wouldn't go through a network.
... VSTs are executable code?
I split pretty evenly between Bitwig, Logic X, and MuseScore and I can tell you the first big reservation I have with something like this is that it's an online DAW.
What does that mean? It means that your compositions (outside of bouncing them down to audio stems) exists within a highly proprietary SaaS format and that the moment you stop paying, you've got NOTHING.
99% of major DAWs (Ableton, Logic, FL Studio, Bitwig, Studio One, etc.) are a perpetual license.
The people using Sunno AI don't care about writing or mixing music.
They don't need the same feature list.
The video seems to be trying to convince me that this is totally targeted at actual musicians. But I guess maybe that's how their target users imagine themselves?
As an actual musician who doesn't have any troubles creating my own melodies and timbres (both generated and 'manually' created), its very obviously not targeted at someone like me.
It seems similar to a Garage Band type of software, aiming to entice people with little audio production experience and give them an interesting sounding snippet they can play back to friends.
For example, the only actual audio editing they displayed was slicing and re-pitching (you can't even choose the time-stretch algorithm), which is conceptually very simple to understand.
There's no ability to actually edit dynamics or do very accurate frequency adjustments that I can see from the demos, so it's basically useless for anything I would want to do.
As much as I think making AI music is the pinacle of lazynes and instant gratification, in my opinion they are actual musicians.
What I mean is that in a DAW you have a lots of tools that don't make sense in AI context.
Like for example, people who use agentic workflows don't need a Visual Studio license.
Well, music is art. Some of it is done for hire or commission. Most code is written for an employer or contract. Code written as art is an edge case.
For advertising jingles it makes some sense. For artistic expression, like... what's the point?
If the act of creating this AI music provides joy to them... I think they are artist
I used to release music that people listend to, but not anymore, now the only joy comes from making it for myself. Am I still an artist?
In my book, you are definitely an artist. Perhaps the purest kind, not that it's a competition.
If the act of creating this AI music provides joy to some, they should do it. I just have a hard time understanding that.
Before AI there was a general consensus that creative areas (eg. Cities) were becoming a homogenized experience. A Starbuckization if you will. I can’t help but wonder what gets lost when using tools like this.
It's unclear to me whether it will result in more homogeneity, as a result of prompts being a coarse medium that results in the AI choosing what it's seen to fill in the rest, or less homogeneity, as a result of more people with non-mainstream tastes being able to create music aligned with their niche that otherwise wouldn't exist due time/money restrictions. I think the latter seems a bit more likely, but time will tell.
There's not really any need to speculate when this has already played out in other mediums - would you say that the proliferation of LLMs has led to an explosion of novel and interesting works of fiction, or just an explosion of cookie-cutter slop ebooks?
I would say too soon to tell. There has been an uptick in ebook slop, but I'm not sure if it's impacted the homogeneity of literature, because I don't think anyone is reading ai ebooks. It's not enough for it to exist to impact culture, it has to be being consumed.
Music is a uniquely interesting case, since music has a much lower barrier of entry to consume.
My thought to who you replied to exactly. Am I going to invest several days to read an AI slop novel? No. But I will take several minutes to read a blog post and likely have read many that were AI generated or assisted.
Culture
Since you get exactly the kind of music you want, I think it leads to extremely small bubbles, which is pretty much the opposite of homogeneity.
For example, I had never heard epic power metal about birds, but with Suno I got exactly what I wanted. Sure, the sound quality (I only used v3.5) could be better and the songs could be longer, but I don’t care, I now have epic songs about my Bourke’s parakeet. However, I’m not pretentious enough to think those songs are interesting to anyone other than my wife and me, hence the smallness of the bubble.
This is an excellent example of where AI-assisted pro music production needs to go. Five months ago there was an HN post on Suno's prompt-based music authoring and I asked for exactly this.
> "I have little interest in typing a text prompt and getting a complete song as an output. However, I will gladly pay serious money for a tool that interactively collaborates with me in a granular, iterative process of generating, adjusting and mixing individual instruments and sections toward a finished multi-track song project." https://news.ycombinator.com/item?id=43873781
It's clearly early beta and some key elements still aren't there - like optionally getting dry tracks (no effects) and MIDI output for generated solo instrument tracks - but Suno is absolutely headed in the right direction. So... congrats to the team. I'm excited! Although I recently retired and no longer have a studio, big budget or session players - I also no longer have the client, the client's "wife with great ideas" and a hard deadline. But I've still got passion, experience and a good ear along with more time to explore creatively interesting directions. Vibe coding is fun but extrapolating forward from the Suno Studio demo video, it's clear Vibe Producing has the potential to be amazing.
ASKS: I'd also love to see the ability to collaboratively evolve one aspect of a generated instrument track while holding most other aspects constant. This is how we'd work in the studio with session players. Stuff like "That's almost it... but give me a few variations with a trill at the end like this clip but played with the sultry sound of the trumpet in this clip."
CONCERNS: My concerns are similar to the gaps experienced devs report with coding AIs being like "a junior developer with a bad memory and a parakeet's focus." If Suno Studio's AI is granular enough to collaboratively evolve its generated tracks over several iterations without losing sight of what's good about them, then it'll be productively useful. The only other significant concern is Suno Studio being it's own web-based DAW. A real DAW like Cubase or Bitwig is Hard. A browser-based tool is going to quickly run into thorny challenges around sample-accurate, jitter-free audio I/O, multi-track disk buffering, real-time effects, sub-mixes, ASIO plugins, etc. And none of that intractable mess of heavy-lifting is required for Suno to deliver their core value prop.
STRATEGY: If I was having a beer with Suno Studio's PM Henry Phipps, I'd be encouraging him to either partner with or buy a mature, cross-platform DAW that's already solved those problems. The "buy" strategy has obvious benefits in quickly side stepping all the challenges of browser-based pro audio tools as well as securing an installed base of experienced producers used to paying real money for real tools. The partnering strategy seems cheaper and easier but, in my experience, doing it well is harder. I'd suggest working with one leading DAW but developing an open API to enable seamless workflows between traditional DAWs and AI services. Such an API would let Suno focus on what Suno does best while letting DAW makers avoid having to become competitively competent in neural nets, licensing tagged training data and running cloud GPU services profitably. No DAW developer has the capital (or access to capital) to seriously compete in those domains. Whether through acquisition or partnership, the long-term silver-lining for Suno may be gaining deep experience in digital media production tooling, workflows and customer mindsets.
> Commercial use rights for songs made while subscribed
LOL oh hell no! Why would anyone use this if a perpetual subscription is required to maintain the rights? Absurd.
i wonder what https://www.renoise.com/ would be like like this
Well it has lua scripting so it just needs someone determined
What is the current SOTA for open source or open weights music generation model?
While I like using AI for assisting with repetitive programming, I can’t help but feel sorry for my producer and illustrator friends who are now having to compete with generated tools.
Is it snobby of me to look down upon art that is created using these tools as lesser because the human did not make every tiny decision going into a peice? That a persons taste and talent is no longer fully used to produce something and for someone reason to me what is what makes the art impressive and meaningful?
Something about art with imperfections still feels exciting, maybe even more so than if I see something that is perfect but if I see an AI gen picture with 6 fingers, I just write it all off as slop.
I am happy to allow my generated code to come from “training data” but I see the use of AI in art, writing and music as using stolen artists hard work.
I feel like as time goes on, I feel even more conflicted about it all.
Applying your logic, did you feel bad for seamstresses when industrial revolution took off? did you feel bad for hardware manufacturers in America when they were outsourced to China? Art is also a form of labor and whoever can produce quality at quantity wins. Idealizing art in some sort of religious idolation is just plain silly. We haven't had the Picassos or Mozarts or Oscar Peterson for quite some time now yet the world is just fine. People play playlists in front of millions of live crowd and get accolade for it vs real instruments. Times change, technology change and art changes.
You either adapt or go hungry just like everybody else and art shouldn't be exempt from the mechanics of supply and demand.
I almost agree with you that this is about quality, but I still feel that the context in which art comes from influences how I perceive it.
Take, for example, a track by Fontaines D.C., a band from Ireland that writes extensively about the lived social and political experience. Knowing where they are from and the general themes of their work makes their tracks feel authentic, and you can appreciate the worldview they have and the time spent producing the art, even if it does not align with your own tastes.
Trying to create something of the same themes and quality from a prompt of “make me an Irish pop rock track about growing up in the country” suddenly misses any authenticity.
Maybe this is what I am trying to get at, but like I said, I feel some conflict about this, as I personally value these tools for productivity
Saying that, maybe a DAW experience makes what can be created more personal
I hear this but this is not the industrial revolution buddy.
You as a human chose to write this very common opinion and even include writing errors like the following
> That a persons taste and talent is no longer fully used to produce something and for someone reason to me what is what makes the art impressive and meaningful?
Human output isn't sacred. yes this is snobbery, a useless feeling of superiority.
I feel the same, including code. I cannot justify it. I can easily counter my own arguments. Still, the further we automate human thought and creativity the worse it makes me feel. I am disappointed that so many are content with mediocre imitation.
> Is it snobby of me
Yes. But aesthetic taste and snobbery usually go hand in hand.
v5 is really good. I can't believe how much progress Suno has made in such a short time.
I don't understand why their vocals are still so bad, though. They always have this tinny, synthy vibe that's very noticeable.
it doesn't interfere with my enjoyment of Suno 5's output and enough for me to pay for it now.
Suno 6 should solve those issues.
As someone who's both a software engineer and music producer its been really interesting to watch the parallel progression of the AI advancements in both areas, as well as my own sentiments towards those advancements.
I've always been pretty bullish about using LLMs to help with coding. My comment history here should verify that. I just thought it was insanley cool tech, and I found I was rediscovering the joy of programming when I could delegate out all the tedious crap I could never get myself to do. Side projects that had languished for years were finally getting somewhere near "done". I also found in many cases it was actually smarter than me, finding cleaner and more elegant solutions to problems than I ever would have. And that was great!
The first time I played with Suno though, it was all so different. I felt deflated. What was the point in making music if a robot could do it better? Yes, the first models were crap, but the writing was on the wall. Music is all about pattern recognition and repetition after all, why wouldn't the robots be great at it? I suddenly knew how all those visual artists felt once the image models started rapidly improving.
So it's been interesting try to disect why I thought coding models were great, and creative models... well, depressing.
Part of it in my case was because being that guy who made cool music was always part of my identity, a part of me I was proud of. Coding, not so much. Don't get me wrong, I've had some pretty big career achievements that I'm very proud of, and I love coding. But at the end of day, the lines of code written were just a means to an end to making something cool.
Which brings me to the other reason I think I was so much more positive about LLMs: the application I was bringing to life with the help of an LLM still very much felt like "mine". Yes, maybe the tedium of writing a bunch of boilerplate was being delegated, but the idea, the architecture, the UX, were still all mine. So I was building something I could still feel proud of.
But typing a prompt into a box and getting a song back? Nah, that's not really mine. It's no different than shouting an idea to a musical improv musician and getting a song back. Maybe you gave them a cool idea, but the song isn't yours.
Which brings me full circle back to this new Suno DAW announcement: this is absolutely incredible. I've only skimmed the announcement so far, but I feel like this brings AI song generation firmly back into the court of how I'm using LLMs to code: letting AI take care of the boring shit, and letting me focus on the composition.
I've only really ever produced electronic music of various varieties, and I have so many uncompleted songs whithering on the vine because I wasted a week of time flipping endlessly through patches, tweaking them, trying to find "that sound" in my head that I could never bring to life, then eventually getting sick of the song and saying fuck it.
And in my experiments with Suno, I found it was actually crazy good at matching a "sound" I described, I just wished I had a way to compose my own song based on the sounds it generated. And now here it is.
It's human nature to want to feel like we've accomplished something. AI generaters like Suno, where all you have to do is type in a prompt and you get the final result, take that sense of accomplishment away from us.
However, if we start working on a project where we're assisted by AI, for example, we're making a game where the sprites are generated by AI or the background music is generated by AI, but the overall game is still directed by humans, that sense of accomplishment stays.
But at some point we're going to reach the stage where the entire game can be generated in high quality, at the same level as humans. What then?
Degenerative art in full swing.