Honest take:
Those demos have soul. The AI versions don't.
They succeed in some ways at capturing the human emotion in the demo, and fall flat in other ways.
- The lyrics capture
- The melody captures
- The production is too polished
- The voice doesn't have the same emotion at all
Depending on how you evaluate, it's either amazing it gets this far, or not impressive at all. :-)
I'd add the demos to my playlist, I wouldn't add the others. (I'm not against AIs in general.)
I would love for someone to make this type of post but switch which songs were AI and which are human created. I would bet there would be people saying the humnan-created songs were soulless and the AI songs had that "je ne sais quoi".
I bet an experiment like that could fool us.
What I'd like to see is a "tormented artist" filter.
I'm sure the emotion of the lyrics can be synthesized.
If humans can fake their own emotions, surely computers can.
Right now all AI songs I've heard have this dead emotion to it that makes me cringe a lot. This crispness of the vocals also don't match the song at all, I'm surprised the author didn't mention this.
Yeah, but here’s the question: If/when A.I. can nail those things, will that be good or bad?
The way I see it is, AI cannot nail these things until you can imbue it with a comprehensive set of human experiences, knowledge and opinions, and the knowledge of every step of the artistic creation process, AND how to weave these two different things together to create something that conveys inherent meaning to other humans by virtue of being the thing that it is.
Basically until we can actually model a human brain in real time, AI generated media will always be kinda shit. Not exceptionalism - just the truth.
Of course it will be good at some point. If AI can match the range of vocals, emotions, pronunciations, why not? Whatever intent behind the art can be attached to generated music after the fact. What boggles me is that some people find the current level of AI acceptable for anything else than playing around or find inspiration.
I'm wondering if we'll ever reach a point where the differences you listed truly become indistinguishable.
Eh, I felt the soul more in Suno’s version, especially the second song.
I’ve got a friend who sends me happy birthday wishes using Suno. I hate it. I love the friend though.
It feels a bit like watching an old movie where the scene drags for no reason and the background’s kinda off, but you still enjoy it. Then you go watch Netflix and everything is smooth, predictable, perfect, and somehow super boring.
I honestly like your version more, with all its imperfections. Feels more alive.
> I hate it. I love the friend though.
Have you mentioned it to them? Presumably you can be honest about it with such a friend. I’m curious as to how the conversation went and what you both learned.
Yeah, he knows. We just laughed about how cringe we are. For his last birthday I downloaded our entire chat history, ran it through Claude, and sent him a visual of our most used words. I also added a few of the funniest lines we’ve had (It wasn’t actually that funny by normal standards, just to be clear)
Honestly, I have no idea how that’s birthday-related but we both enjoy it
This is really cool, the original demos were great to start with too.
The AI vocals still feel too "easy listening" and smooth in general though.
It works great for mastering the instruments though, I wonder if it'd work better using it just for that with the original vocals.
I like experimenting with suno. This is the only way for me to create music. I do not have time to learn fruity loops, or something.
Most of the lyrics are generated using chatgpt, but I often force it to have some parts of chorus, and there are many 'themes' I want the song to have.
So I play with lyrics, play with tools, and sometimes I am surprised by the final effect.
https://www.youtube.com/shorts/Q2FiqrppN4g
I created also a concept album. Folk metal about technology and fantasy. That is how "PC goblin", or "Trolls in the motherboard" were created. Some parts of songs are better than other, but in reality I see not so many other songs exploring that niche tech-fantasy area.
The other time I created power metal fantasy album, about AI and end of times. It was really a fun time for me.
I upload these videos to youtube, expecting no views. Unexpectedly shorts get views. I am not sure why. Probably youtube forces any shorts it can find to users.
Suno is incredibly fun. I've realized there's ton of lyrical and topical niches that nobody is exploring in the music space, so AI generated tools are really good at helping you scratch your own itch. The iterative loop still feels like it has a lot of room for tightening up, especially when you want to start refining minor details to match more closely with the idea you have in your head.
Personally, I had some experience learning to compose music from a college course, and I've been trying to relearn some of what I'd forgotten by trying out this textbook: The Complete Musician, but progress is slow. And even if I'm able to put together a musical piece, I wouldn't be able to afford hiring the people needed to bring it to life.
I wonder how far we are from AI tools where you can just start describing and humming your idea out, and it'll give you samples that can be mixed in real time. Sort of like being in a studio with a full support staff of professionals with a lot of improvisational experience. Vibe mixing and composing in real time.
"This is the only way for me to create music."
You're not creating music.
It feels incongruous to read that you don't have time to learn a DAW, or an instrument, but you have time to upload to YT, in multiple formats, and compare outcomes between shorts/vids?
Small pockets of free time can go a long ways: putting in 15-30 minutes a day towards learning something new will produce tangible results.
Learning to create music is such a multifaceted pleasure: exploring tooling and mechanics, integrating somatic processes (feeling it!), investigating theory and history, increasing execution incrementally, exploring and expressing emotions+ideas, and maybe, eventually, conveying those emotions and ideas to listeners? What richness!
The Suno->Youtube pipeline feels like an attempt to experience a watered-down version of the conveyance bit, at the expense of the rest.
Thanks for feedback.
Though I must correct you in places. Uploading a video to youtube takes me 1 minute, if not less. I really think that learning instruments takes a lot more time. I don't upload into youtube in multiple formats. I only use one format (as is from suno).
I can play a little bit on guitar and ukulele. I know that this is definitely not a route for me. I use my small pockets of time to lead my hobby projects at github. In life you have to choose about how you invest your time. I have chosen. It is still fun to "create" music through suno. So please do not suggest I should be playing 15 minute a day, since this is my time and I will invest it the way I like.
I understand that "creating" music through AI might be a lesser form of creating music, but for somebody who does not have that time for music this might be a fun thing. I hope you can understand that.
I think that I am creating music, since parts of lyrics are written by me, the rest I accept as they are. I remember xbcd cartoon about how programmers argue about how "real" programming is done.
I also remember that Jean Michelle Jarre had troubles with being recognized because he created "electronic" music, which was treated as something of a lesser form. I remember Andy Serkis not being recognized for his work in motion capture.
I think that you can "do" music any way you like, and play with different tools and things. Sure some musicians will laugh at you because play from tabs, not from notes, but hell I despise elitism.
> This is the only way for me to create music. I do not have time to learn fruity loops, or something.
Maybe try physical instruments, I've got a few synths, an acoustic guitar, a few weird microtonal flutes, I don't know any music theory or how to properly play them (I don't even use the fretboard on the guitar, just the strings themselves) but I know what sounds good to me and can get some really interesting stuff just mucking around with them. No doubt you can achieve something similar with software instruments too, I've just never found any that feel as immediate as physical ones.
I'm a bit biased against AI music, but I still believe even if you enjoy it that actually making something yourself that sounds good is more satisfying than prompting an agent to do it. Maybe that's because you do learn something each time even if you don't intend to, or maybe it is just my bias about AI music, but might be worth a try
You keep using the word “create”, but did you really create any of it, or did you commission someone else (which happens to be a machine) to generate it as they saw fit, with some minor guidance?
Let’s say LLMs and generative AI did not exist and your chat sessions were with a human. Could they not have delivered the same (albeit taking a bit longer)? Would you also have claimed you had “created” the music then? Basically the difference here is that with AI you can take the credit without the other party complaining.
I’m glad you had fun. Nothing wrong with playing with a toy and entertaining yourself. But to claim you created something while simultaneously brushing off taking the time to learn the skill doesn’t quite feel right. Like saying “I made this table. I don’t have the time to learn carpentry or something, so I asked a guy to do it for me. I did specify it needed four legs and a back, though, so I did it”.
I think that people often project their own assumptions. How do you know how much of it (lyric-wise) was AI generated?
If prodigy created track based on samples from other songs they cannot use word "create"?
If I create a "collage" art, am I not creating something?
I think that some people are fixated on creation part. "You didn't create this picture, because you used Adobe filter on you digital art", "You didn't create this picture because you are digital artist, and was not using real brushes", "You didn't create Hellblade Senua game, because you just used Unreal Engine, and it did the heavy lifting for your, so you didn't really create the game".
> I think that people often project their own assumptions.
I agree. It is unfortunately also a fact that people often don’t accurately read what was written and don’t respond to the argument.
> How do you know how much of it (lyric-wise) was AI generated?
You said it:
> Most of the lyrics are generated using chatgpt
Most.
> If prodigy created track based on samples from other songs they cannot use word "create"?
> If I create a "collage" art, am I not creating something?
> I think that some people are fixated on creation part. "You didn't create this picture, because you used Adobe filter on you digital art", "You didn't create this picture because you are digital artist, and was not using real brushes", "You didn't create Hellblade Senua game, because you just used Unreal Engine, and it did the heavy lifting for your, so you didn't really create the game".
None of those are the same thing. Your examples all require skill to do well. You, by your own admission, have eschewed learning a skill. That is a gargantuan difference.
It feels very nitpicky to police the use of the word create. If GP hadn't willed it, the song wouldn't exist at all, it was only created due to their action.
But I guess it depends on how much creative input is given. If I write the lyrics for a song, I feel like that's a significant contribution in creating a musical piece.
If I made a detailed sketch for a table design and I hired someone to help correct any oversights and to build it, I think you can take some credit for the creation of the table, because your design was a key input.
> It feels very nitpicky to police the use of the word create.
Conversely, I feel the word “police” is overused in these contexts. I’m not stopping (or attempting to stop) their use of the word. Nor could I, I have no power to change their word usage. But words have meanings, and their overuse dilutes and makes communication more difficult. If someone uses a word multiple times in one short comment as part of their main thesis, I think it’s fair to reflect on if it’s being used correctly. That is, of course, a matter of interpretation and open to disagreement. That’s what I’m on HN for, to bounce ideas and develop my thinking.
> If GP hadn't willed it, the song wouldn't exist at all, it was only created due to their action.
Which is no different from a commission, which is why I used that word (again, meaning matters).
> If I write the lyrics for a song, I feel like that's a significant contribution in creating a musical piece.
The OP said “most of the lyrics are generated using chatgpt”. So most of the lyrics are created by an AI tool and most (if not all) of the music is created by another AI tool. Where is the person’s “significant contribution” in this scenario?
> If I made a detailed sketch for a table design and I hired someone to help correct any oversights and to build it
Except that doesn’t look to be at all what happened here. It looks more analogous to giving some vague verbal instructions at different stages of the process. Again, no different from a client commissioning a work from an artist.
Crucially, your example does require learning a skill, which the OP admittedly didn’t. Making a detailed sketch for a table design requires that you learn how it’s done. In this case, it would be somewhat analogous to the OP having written musical notation to generate the song, which is also a skill.
But here, no skill was used or pursued.
Your hypotheticals are valid in general. But I’m not discussing hypotheticals in the aggregate, every analogy I made was about this specific case.
To be fair, after watching a bunch of Rick Beato's videos on YouTube, I'm realizing this is how the music industry actually works, regardless of whether or not AI is involved. Maybe 'create' isn't such a bad word for it!
Like, I didn't know the name Jack Antonoff before I watched those videos, but I sure do now. Gareth Emery and Ashley Wallbridge's song CVNT5 also talks about this.
What you're really complaining about is the ever downward price pressure on creating music. AI is just a useful way to facilitate that, but the process of making music more cheaply has been going on for decades. It's the reason that record labels don't really sign rock bands anymore, because they're much more expensive than single person acts.
He is a producer :)
One thing I’ve always wondered - what happens if/when fans want to see these AI bands live? Or will this be the «sorry we don’t play live» era? Or maybe if something goes viral, one has to hire in musicians that can play/sound close enough?
Anyway, I liked the sound of the instruments better in the demo/first versions. Sounded less mushy.
How do you think it works currently when artists who make highly edited music perform live? And I guess if Gorillaz and Hatsune Miku can do concerts, this isn't really a problem.
I’d say it is highly dependent on the style of music. I mainly go to metal and jazz festivals. I just can’t imagine someone on stage lip syncing to music, some holo/avatar, or whatever.
Holograms is one solution that is already being used, see also:
https://en.wikipedia.org/wiki/Virtual_concert
Example:
It’s pretty standard for electronic / sample based music acts to have live bands assembled for touring after the completely sequenced studio album takes off. It’s a hit or miss depending on how well the musicians can reinterpret the source material.
It is also pretty standard for a DJ to move knobs and sliders as performance art even when the knobs/sliders are doing nothing. It adds to the performance much more than just standing there.
We’re probably already at the point where we can create covers of your favourite song by your favourite band but from a different genre!
Really intriguing use of Suno.ai, generating AI-backed demo tracks to accompany your album feels like a reverse collaboration.
What stands out is how this method reframes demos—not just rough pre-production but fully formed creative statements in themselves.
Curious: for anyone working with AI in music, does this shift in workflow change how you iterate creatively? And how do you balance AI-generated backbone with your own unique touches?
lol
This is actually subtle, funny, and insightful commentary. To me it says, "just as AI-written text is characteristic, annoying, and hard to miss, so too is AI-generated music." Not sure I necessarily agree, but I get it.
You're way better than you give yourself credit for. Those demos trounce the AI.
Makes me wonder how often A.I. will be used by creative people to write themselves out of the equation voluntarily because they are self-critical or have imposter syndrome
> Not a single song is quite ready to publish.
Even if they were, what would be the point? If AI can copy any music, why would your copy be better than any of the infinitely many others?
dude, the demos are better.