It’s a fun service and an impressive demo, but to “make music” I need a lot more control than just entering the lyrics and describing the style. That said, there are some awesome examples there already, even with the current limitations, so I’m hopeful. Apparently so are their investors.
So this is just using voice synthesis and some sort of "AI" to put lyrics over a backing track? Can't imagine anything lasting or worthwhile will come out of this.
You describe the music you want in text and it generates the whole music with or without lyrics (your choice) a la stable diffusion. You can optionally supply it with your own lyrics. I don't think it counts as "creating music" yet but with inpainting and better tooling, it can probably get to where text to image generation is today in short order.
It’s great for what it is. Go to their site and give it a listen. I was even able to generate my own stuff that sounded quite good to my hobbyist musician ear. Some tracks are so good that you wouldn’t know they’re generated if you heard them on your favorite streaming service
This is about generating something to consume though, not make? Making music is about so much more than the end product. It's the path not the destination etc. You learn so much about life when you have an artistic practice in general (not just music). I'm perfectly fine embracing AI tools, but clicking some buttons to generate a song isn't making music.
Ostensibly, many folks here philosophically believe either that amalgamating other people's art is tantamount to making art, or that the artistic process is irrelevant. I think many of the former are inexperienced artists that see generative AI as simple art tools. From that vantage point, it's tough to understand how many specific, artistically consequential decisions other people made to create the end product, and have a hard time understanding why people who make those decisions are rankled by people copping them. The latter seem to think tech + art = neat regardless of the context, economics, etc. Either way, I think I've seen exactly zero people budge on either stance. My gut says folks with those perspectives are probably over-represented in SV-focused spaces.
In the free tier, Suno owns all of the Output. I'm wondering when someone will make a song, realize it's pretty good, and attempt to recreate it after creating a paid tier account.
Subject to your compliance with the terms of this Agreement, if you are a user who has subscribed to the paid tier of the Service, Suno hereby assigns to you all of its right, title and interest in and to any Output owned by Suno and generated from Submissions made by you through the Service during the term of your paid-tier subscription. If you are a user of the free tier of the Service then, as between you and Suno, Suno owns all Output generated from Submissions made by you through the Service, and, subject to your compliance with the terms of this Agreement, Suno grants you a license to use such Output solely for your lawful, internal, and non-commercial purposes, provided that you give attribution credit to Suno in each case.
Not make but generate. The time comes when music played by real people, even with mistakes, will be valued more than electronically generated performance.
> The time comes when music played by real people, even with mistakes, will be valued more than electronically generated performance.
Robin Hanson recently said something to the effect that this might be the last era of human-dominated artistic production. As such, that works produced in these days -- these final days of the human arts -- might be especially prestigious and valuable in times to come.
I believe it. Suno is shockingly good already, and it's only going to get better. The curtain on human composers is probably going to drop faster than we think.
Suno et al are trained on existing music created by humans. What happens when "real" musicians will stop publishing new material out of frustration because they will drown in a sea of AI generated garbage?
I'm sure it's possible to guide them to originality. And, in fact, even Suno at this very early stage in its development is capable of creating original music, sometimes even by accident.
What it came up with was not much like what I had intended. It appears to have interpreted "ambient" in the alternative way -- as in ambient café noises and background murmurs. What it produced was a conventional hazy ambient track with what seems, in a dreamlike way that's hard to grasp, like a conversation running in the background. Quite unusual! https://suno.com/song/45b23814-59e2-480e-9a2a-7c7a895b3957
I am sure you can use it in original ways. I got some pretty interesting and strange results myself by giving it contradictory or absurd
prompts.
I am a composer of experimental music and I can totally see how I could incorporate these AI tools into my existing art practice and use it in transformative ways. But that's not how most people - and the bad actors in particular - will use it!
Exactly. This is a negative feedback loop. And it will render people stupid, fragile, overwhelmed and depressed. Only those who persevere this will be able to create something new, but rest of humanity will be melted in auto-generated content with less and less entropy
It blows my mind how well Suno works. It's one of the few products I tried lately that exceeded by expectations by a wide margin. As an amateur songwriter who is a bad singer, hearing its interpretations of my songs was beautiful. The implications of this for the music industry is pretty disruptive.
I would be surprised if Suno is very disruptive to the music industry.
It has extremely impressive output for a music generator, and it is very fun to play with - but people will always want to hear something new, and something real.
I don't think that music fans are going to connect with completely AI-generated music without a real personality and story behind it - and good, original music will always be more impressive to people and likely more successful than an AI-generated amalgamation of what already exists.
Random anecdote: I've created a Suno song as an anniversary gift for my girlfriend. She was absolutely mindblown by it as it's an earworm song with many of our memories.
Always good to be aware of our small tech bubble here and that things we take for granted already, might not even be close to adoption :)
I’ve created a few variants of a song for a singer/songwriter friend of mine using his lyrics. His mind was likewise blown, and it gave him some fresh ideas on how to improve his song
I don't think the Big Three record labels will want to stop this, even if it's massive copyright infringement, because it's not a threat to their business model. Labels create a whole ecosystem around a limited set of artists through marketing and tastemaking, then capture multiple revenue streams (streaming, licensing) for the few artists who people mostly play and pay for. They aggressively persuade musicians to sign away the rights, so the labels control the terms of payment, and they work together with a tiny group of companies in streaming/radio/etc. who have the same self-interest.
Everything outside that structure is an afterthought. The occasional indie hit songs and labels have failed to upend the music industry power structure for a century (they tend to get acquired if they get big enough). Tons of people making songs mostly for themselves will only dilute the power of smaller players.
The labels will probably extract some licensing fees off the stolen copyrighted training data, but they famously don't care about their musicians earning a livelihood.
For greater control, higher quality audio results and less censorship, but with the downside of being a bit fiddly, there's Udio. It's otherwise functionally similar:
I'm really intrigued with both Suno and Udio as to how the music is represented internally.
Does the model build up track by track vertically, which would then lend itself to a more capable product for professionals, an AI powered DAW if you will. Or is it building a linear stream of all the sounds beat by beat e.g. horizontally?
FWIW I got consistently more musically pleasing results from Udio than Suno. Although occasionally Udio would sing AI gibberish.
This kind of mission of "enabling anyone to do X" seems to undermine the value it claims to provide.
So, if anyone can make music, then what's the value of being able to make music?
But, if what it enables still requires some rare talent or significant learning to make good music, then how is that different from today? And, well, anyone can already make bad music.
Or maybe I'm just in my greybeard "get off my lawn" mode today.
>you were sold a lie that only certain people make good music and you need to spend money.
Well, I didn't say any of that, so not quite sure what you're talking about.
>Everyone should make music. Singing to yourself doesn't make it valueless... with family and friends...
Yes. That was my point. Anyone can sing to themselves or with family and friends now.
But, I am guilty of not reading the article and grouped it with the spate of AI tools promising to enable anyone in a manner that gives them some marketplace advantage.
I see now that this is not that. They claim instead to want to help amateurs and first-timers get into music for the fun of it. More power to them.
Does anybody know whether these services prevent a user from copying an instrumentalist's specific style? For example, let's say I input:
> "make a song with Johnny Cash singing about X"
That would result in IP/copyright issues, no? I don't really know the legal specifics here, so grant me some slack if I am not using the correct language. But, assuming that input does create a legal problem, does the same apply to these sort of prompts:
> "with a guitar solo by Zakk Wylde"
> "with drum fills like Thomas Stauch"
etc...
Are there similar legal protections around instruments as the voice?
Suno has some advantages and cons, it can make any music style and can render 2 minute songs in seconds. Its lyric generator is also a tad nicer than others right now. Vocals are horrible, I think about 25% of the vocals sound nice.
Its training on musical styles gives it a lead, its vocals needs much improvement.
Its a great tool for creativity, but its a tad far from music ending up in my playlist due to vocals. For now. Instrumentals, i can totally see using it for background music in videos, themes, trailers, etc.
Having used it a bit, I'd agree - but when it does good vocals, they have typically been very good. When it's failed it often is because it keeps layering on more "takes" in what ends up as a thick mess. Seems to progress worse as it goes if it does it.
It is kind of crazy that "BBL Drizzy" of all things set a precedent. Using AI to create an incomplete song and then a real producer using it as a royalty-free sample to create an actual hit song.
I could see this paradigm becoming insanely popular with indie artists to make high quality / low skill backing tracks for their vocals.
If you want to see how AI music is going to play out, just look at how AI images have played out already. AI tools are used pretty ubiquitously now by digital illustrators, who use it to flesh out or quickly build up a prototype from hand drawn outlines, then add finishing flourishes to make it look "human."
The difference between music and image though is that very few people work professionally with audio. Just consider how many Photoshop/Figma licenses your average company has, and how many Logic/Ableton licenses they own.
Pop musicians that make a lot of money are 80% the result of marketing/branding/image and 20% music.
Major acts like Taylor Swift are a lot more threatened by AI being used to generate pictures of her having sex at a Chiefs game than they are by AI creating Swift-sounding music.
They're gangster but with lawyers. They'll get the algorithm to generate a bunch of stuff that's dubiously similar to stuff they own, then sue Suno for $200M.
Yeah I've heard this story before. What's actually going to happen is that Spotify and every other music streaming service will be filled with 50 million functionally identical AI generated songs made by people looking to scrape off a few cents when people listen to them by accident.
Spotify is already filled with “10 million of functionally identical songs” made by people “looking to scrape off a few cents”. The state of the art, such as it is, could hardly be any worse.
Automating slop will absolutely increase its volume multiple orders of magnitude. At least human made music, however terrible, stands the chance of accidentally being interesting.
When everyone can produce “slop” in seconds the value of slop rapidly approaches zero and you have to do something the machine can’t do, which is produce original, superior material. You also get to use generations as a starting point and improve upon them. I think if anything this will only improve the human SOTA.
The value of slop does approach zero, but as the time and money required to produce slop also approaches zero it takes very little incentive for people to keep churning out infinite amounts of slop. If it costs 2 cents to make an AI generated song and the EV on its Spotify revenue is 3 cents then it's worth doing if you can automate it. Meanwhile everyone making an actual effort, using AI or not, gets buried under an ocean of shit that nobody has the patience to wade through.
This is the principle behind AI-generated SEO blogs, the content is garbage that nobody wants, but they are profitable nonetheless.
I think that's naive - good luck finding the "superior" material when the marketplace is flooded with garbage. Amazon users are facing this exact issue at the moment with chatGPT generated 'books'. Users and site operators cannot scale up to filter out the low quality sludge being generated by AI.
But we already have to do that. The marketplace is already flooded with garbage. People who do original things will still float to the top. People who already produce poorly made crap will have to do something else
Why would they float to the top? Marketing budgets will push garbage to the top like what happens now in music. No real change. Leaving the audio tune era into the AI era isn't going to change anything.
Right, this will probably end up being like how autotune was for singers.
I don't really want "anyone" to make music. I want people who are good at making music to make music. Yeah we should remove barriers for those people, but honestly nothing beats practice and dedication. Tech seems more and more focused on regurgitation instead of creation.
Autotune sounds different than non-autotune regardless of whether it's used by a good or bad singer. As a result, it can be used for creative purposes, even by good singers.
Drum machines sounds different from kits, which doesn't mean that drum machines are an inferior version of drum kits. They are different, and there's room for both.
Maybe you will never like any song with autotune in it. That's fine. But the idea of objectively "good at making music" leads to some nonsense. There's no accounting for taste.
In my opinion, the more music made, the better. There's something for everyone.
The thing is, we have already removed all of the financial/logistic barriers. Anyone who owns a laptop can setup their own little home studio with a mic + audio interface + pair of speakers for under 200 bucks. That's how I started as a kid. But as you say, you still need practice and dedication (and some talent). If you get rid of that, you essentially get rid of art.
That being said, I do remember Magix Music Maker and Acid Music from the early 2000s where you could easily create your own songs by drag&dropping loops on a timeline. Actually, that's what got me into music production as a kid. I think this kind of software can be a great starting point. The problem with Suno and Udio, IMO, is that the interface is too basic and offers very little control over the actual music.
Now, the actual problem is that people will abuse these programs to spam established distribution channels with low effort AI garbage on a massive scale. We already see this happening in other domains (digital visual art, books).
Of course, you can't expect tech companies to consider the broad effects of their products on society...
We were already there a few years ago before the generative AI boom.
I'm starting to wonder if this stuff is actually going to make real art "hand made" by real people more valuable in some ways because it'll stand out from the mountains of auto-generated trash. Problem will be finding it.
I think it's already flooded with spam from non real artists. I had weekly discovery playlists, where I did downvote all of proposed songs... Each new set was still coming with most boring, uninspired, flat and predictable structure (and abstract cover art), which for me is an exact equivalent of those NFT images.
The images are designed to be a set of replaceable elements that have to follow the same "joint structure". Once you see it, all charm is lost and that vague "why does it look so funny" feeling is simply replaced with disappointment.
I haven't watch this 26 minutes long video but I have already seen AI music being used as like theme song of indie games and people do bot realize it's AI, even under a Youtube video that just upload the music.
EDIT: The video only claims that AI would not be able to pass a "Musical Direction Test", but a blind Turing test only on output, yes.
Sure, anyone can make music so long as they pay the Suno tax. While I appreciate the technical achievement, this kind of technology disempowers as much as it empowers. The process of automation inherently devalues the process that is being automated. It is easy to say this doesn't matter if the process being automated is meaningless or unimportant to you.
Making music is not just about whether a song measures up to some objective standard of goodness. It is about the process of connection and sharing between the musician and audience (which I mean in a broad way -- it could be another musician in the band). There are many amazing musical experiences that I have had that are not possible except in a live experience. My concern is that these kinds of tools will dissuade people from participating, in no small part because ai music is better than what most people can produce -- by the standards of recorded music. Why should I even try if I can't even come close to an ai?
In a worst case scenario, and I'm not saying this will happen, ai generated art (not just music) creates a doom loop where people stop making art themselves. Communities formed around participation in art wither away and we lose the ability to make art ourselves. We then become solely reliant upon ai for art, which means that art will primarily be consumed through the human -> ai interface rather than the human -> human interface. I'm not opposed to people experimenting with ai but I am worried about it replacing the human -> human interface and, frankly, the last 20 years of social media give me ample reason for those concerns.
$125M? opens the website, super excited to check what's their plan with the pedagogy of teaching computer music to realize it's AI, again, probably violating copyright
