I don't see AI being anything more than using loops to create tracks. It is just quicker as it will work faster than we can with searching and listening to different loops, vst instruments, etc. It is just a way to get ideas started quicker. Caveat: I don't use AI or full music loops but I do use drum programs/sequencers for grooves and fills to create tracks.
30:51 I couldn’t disagree more. Yes it’s true that oftentimes sync music is low in the mix for obvious reasons, but that doesn’t mean that we waste time trying to make it as good as possible. Nor is that a reason to choose AI music that’s not as good, just because it’s low in the mix. 1. There are cheaper cameras. Why not just film your show with a cheaper camera? Why not just hire a cheaper writer? The production is the sum of its parts, and high quality across the board will produce a better product. Whereas short cuts and cost cuttings usually have a detrimental impact on the end product. 2. This is a really obvious one. The music might be under dialogue for a while, but oftentimes it jumps up into the foreground for a while, and then ducks back down below dialogue. Who in the right mind would want a rubbish track playing when it jumps up in volume? Don’t underestimate the value of high quality sound and music, and how that adds the polish to a show. If Paul doesn’t give his 100% on every cue because it’s going to be low in the mix anyway - well speak for yourself ;)
I'm not suggesting at all that you accept poor quality, or ignore every aspect of your production. I'm pointing out the very irony of our music being hard to hear as a reality check for those that say AI music is not good quality from an audio perspective. Some people are saying that AI tracks are not mixed properly, or sound AI-generated. My point is "so what." At the end of the day, whether or not the track sets the right mood for the editor is what matters. AI will only get better, and much faster than most understand. To be clear: ALWAYS deliver your best product... then hope you can get a nice forefront placement. :)
This was an amazingly thoughtful and intelligent conversation. Thank you Jesse and Paul. You’ve subtracted the emotion and fear and presented a very comprehensive set of ideas that could serve as a fine basis for creating the social framework for AI in art and content generation. Thank you both.
Great insights by both of you here. I'm not in the sync music licensing space- I'm a classically-trained musician, and I have played cover music in a lot of different genres for the past 50 years. But these issues about AI-generated music are interesting, and all of us musicians should stay informed about what's going on. Any musician who plays covers, and very few of us don't, is technically plagiarizing others' intellectual works. Most music that we cover is also copyrighted. But cover bands and solo musicians are widely accepted and hired by venues. Their customers mostly want to hear familiar music- things they've heard before. Example: I'm in a Selena tribute band (the Tex-Mex pop star who was killed early in her career). None of the people who come to our shows cares about the fact that we're plagiarizing Selena's music. They just want to re-live the 90s and the feelings that they had listening to her music and watching her shows. Emulating, plagiarizing, copying, learning from copyrighted music is a part of being a musician. Particularly a musician who plays covers. We stand on the shoulders of the musical giants who came before us. I've also heard it said that "Imitation is the sincerest form of flattery." A.I. being trained on existing copyrighted music? I don't really care about it. All of us train ourselves as musicians on copyrighted music. Very few of us only play music that we've composed and recorded. Am I concerned about AI music taking over? Not much- we musicians are probably more concerned with DJs taking over our gigs, because so many club owners are cheap bastards and they don't want to pay bands. Okay, that's not totally fair, because club owners are trying to run a business, and they have to make a cost/benefits decision on whether it's more advantageous for them to hire a DJ or live musicians. But that's a whole 'nother topic.
Personally my biggest gripe with big tech is the industry of mining and scraping data, especially personal data. When you create a free social media, you are allowing them to exploit pretty much everything post, say, upload, your browsing history etc... It seems there are many Paul's out there, that have no issues with this.
This is why opt in is important. Paul can opt in to have his 1,500 cues trained on by AI companies who won’t compensate him And we can all opt out. Win win!
If AI music gets popular or at least ubiquitous in advertising and movies, it will move that music in the direction of what it was trained on - specifically, not your stuff.
For software engineers we were suddenly and swiftly thrown into "retirement" even for those in their mid-20s, I don't think anyone CAN plan on retiring at 27.
I've finished Bansuri Dreams and it really did sound like after 3 hours of manual crafting graft that it's 100% a track that is all Lakha Music Sonoton/Weststar/Royalty Free and Custom Unreleased all in the same track. I thought it was Lakha himself playing Indian not AI. I think the production music artists might have something to put forward a lawful case for a PMA, PRS and Cocatalog licence where the owners of the site don't pay it would be law breaking because not paying is opting out like the TV Licence in the UK where you can't watch any TV unless you pay the BBC to watch all channels.
Love it. This was a wonderful piece of dialogue about a hot topic. I really like Pauls attitude towards creating music and learning new tools to create better. I am definitely in camp Paul as a creator. Ai is a superb tool for me. I can throw ideas at it and get a result back in just seconds. It instantly gives me a way to judge if my idea is good or not and if it gives me material that I can bring into my DAW to work with. It allows me to try things that I would otherwise waste days on before finding out it was a crappy idea. Real timesaver. But. Looking a bit into how the legal contracts of the Ai services it seems to put all responsibility about if the content created is copyrighted material on the end users. I am making music for fun and I don't care if I don't make any money off it, but I sure as hell don't want to get sued for something that I have never heard and had no idea it was not legal to use. The logical solution would be to impose on Ai services a part to the algorithm that makes it impossible for it to generate pieces that is too much alike the original piece to be a legal problem. But listening to Paul it seems like it is not even clear in the law how to judge what is legal and not. Am I right here? Is it impossible to make sure the outputs of Ai stays clear of copyright issues?
great discussion! appreciated the details of how the Google LM trained on music - random 10sec samples, easy to defend legally i'd imagine. Training model protocol (essentially explaining HOW input is selected) / and subject matter should definitely be public information. that would be very educational and interesting.
Really interesting, great conversation... two good knowledgeable experienced talented and wise people banging out their thoughts and advice. so much to sift thru, but bottom line we've no idea what the future holds, but embracing/learning the new way of doing things is at hand, and the sooner the better, because it's the future, and its legalities will work themselves out because they have to. #OdedFriedGaon #OdedMusic #OdedInformation
This generation's heroes are the last generation's hacks. Paul made a great point about the musicians we think are innovative today, who are really taking influence and borrowing from previous musicians. Every sync opportunity starts with, "We need something that sounds like X, Y, or Z." They never ask for something unique. Entire genres are based on the same guitar riff or bass lines. As soon as musicians started making complete songs with loops and library content, this was inevitable. How many times have we heard a song with the AMON break? On the Red Lobster example, I think you are missing the point that it is a parody, and it would be protected the same as any other parody would be today... not a violation. Also, it doesn't take away from the original; it doesn't compete with the sales of the original or turn people off from listening to the original song. People have been making tracks like this for years before AI; now, it's just better at it. Also, AI is used in every aspect of the product process... where would you draw the line. Companies have been using covers of popular songs for decades because it is cheaper.
I'm with Paul on this, and I know it's not a popular viewpoint. Imagine, though, that copyright worked the way most musicians seem to want it to: where training on copyrighted works first requires permission from and payment to the copyright holder. Like Paul pointed out, we all study copyrighted works. We don't get permission and we don't necessarily pay the copyright holder. And most of us take what we've learned from this studying and write new music inspired by what we've heard and if we're professional musicians, we're using that learning for commercial gain (for-profit). If the law works the way most people seem to want it to, no one would be able to listen to any copyrighted music and then write music unless we first get permission from everyone whose music we've ever listened to. It's untenable. AI is doing the same thing. It's studying music (copyrighted or otherwise) and then making new music based on what it's learned. In my view, that doesn't require permission or payment. Jesse, with respect, I don't think you are viewing the 4 elements of fair use properly. Those are for copying. But the AI systems aren't copying. They are making new music that did not exist before. It's not like sampling, for example, where exact copies are being used. In most cases, these systems are outputting stylistically-generic music that's bound to sound a bit like existing music because that's what "country" or "rock" music sounds like. I challenge anyone to make Suno or Udio produce an actual copy of, say, "Bridge Over Troubled Water." That's not what these systems are designed to do, and for the most part, they don't. The "copying" is at the input side--into the training set--not at the output side: the music the AI system generates. Copyright has no concept of "unauthorized studying"; it regulates the output, not the input. And that's what we want because otherwise we are all guilty of "stealing" other people's music. Because all of our music careers are based on hundreds of hours of listening to and studying other people's copyrighted music. But those are inputs to our brains. What copyright cares about is if our OUTPUTs are copies of other people's copyrighted works, not if our inputs are. That said, wouldn't it be nice if we made special rules for use of copyrighted music in AI training sets? I'm for that, if we can figure out a way. Maybe something similar to the compulsory licenses for recording cover songs?
@@SyncMyMusic I look forward to the video! But even if one is able to force the system to generate a copy of a song, that is an outlier case--it's not what the systems are designed to do, and tarring them with the broad brush that they are copyright infringing overall is still wrong in my view. That said, my experience with these tools is that the more specific you need the output to be, the harder it is to get the system to generate it. I would love to see a video of how someone coaxes such specific output from these systems because those techniques could also potentially help me get specifics I need in songs I create with these systems. My use case is a bit different in that I am a training professional and my interest is in creating instructional novelty songs--sort of on the "Schoolhouse Rock" model--to help convey some of the key messages of the training sessions in a fun way.
It’s like if you want to iron out the conversation of charging ai to train you then who and how many gets to charge you from training them? For example we know hip hop pioneers. We know when you aren’t one but you’ve made a living studying training on it. So I believe when he says it’ll be a mathematical mess it’s giving those who fall under if ai is to be charged then you should too a way out if I’m understanding him correctly and it’s definitely a conversation as a whole if anyone wants to continue because what’s the difference between their training for use of for profit company and yours? Speed and connections. The same structure and process is still there and no one expects you to license everything you trained on even if it was a lifetime full of work and a computer chip has multiple lifetimes full of work. This was like a oversight at the beginning of internet when you choose to opt in. Maybe that’s why so many fought on this topic then about what’s still yours on the web. Very interesting topic that shows great courage to at least try to find fairness at least going forward for yourself. Nothing is to hard to talk or figure it out especially when on some level it’s already being done. If we agree that the computer is so powerful then why not rewrite the code so that all training can be licensed even on the micro. That would be great and traceable to pay homage to those being trained on even longer. Let’s do it. That means schools too would owe for teaching those same tracks. My school constantly encourages me in active assignments to download Spotify to learn from something. I’m just not into it because of other work that clouds my priority then listening to music but that’s for a grade at a for profit school so if ever it’s a time to consider changes it’s now. Let these people and their legacies get paid now even for training by all. It could be a major flex to be on file as trained using one of the greats.
I could be very wrong but in my own opinion typing prompts into a computer to generate music especially if the prompts are similar will result in similar music being generated and ownership battles along with copywriting problems will begin.
Udio's terms don't even deny this. They concede their outputs may infringe on copyright but it's the users who are liable. They also state multiple users can generate the same outputs from similar inputs. They seem to be creating a total copyright minefield. It's why comparisons to the visual art industry is misguided. Music IP has pretty cut and dry ownership and people will be wayyy more eagle eyed (eared?) when it comes to AI music. Musicians sue each other all the time for copyright infringement which is almost always accidental. Imagine hundreds of thousands of people generating near infinite tracks with the same generic prompts. It'll be such a mess.
@@SC-ew2fc Thanks for confirming my suspicions I'm definitely not opposed to AI in certain roles in fact I love Izotope and everything they offer. I fully agree this is gonna get messy before it gets sorted out.
It would be incredibly unlikely that this happens. There are exponentially more varaibles involved in creating the outputs... just look at the odds of winning the Lotto. They are infintisemal. Well, the statistical likelihood of AI generating the same track are a fraction of a fraction of a fraction of the chance of winning the Lotto. :)
@@YoPaulieMusic I saw on a forum that someone had generated a track on Udio, but not published it. Then they heard portions of this track in another users generation on the homepage under “staff picks.” They were trying to figure out how that could happen, but it has. It’s probably why Udios terms explicitly state it. It’s probably why you’re also seeing composer contracts with studios forbidding them to use any generative AI in their scores. If they can’t own the music, what’s the point.
@@SC-ew2fc It's not a copyright minefield because Udio effectively claims they have ultimate ownership of all the creations generated in Udio, as per their "Attribution" subsection in the Terms of Service. Attribution is required whenever each piece is used for any reason other than mere personal, private use. Also, claiming that Udio generations can potentially infringe on Copyright works as being a tacit declaration of guilt is not entirely correct. One major reason is because the reality in which we reside has a principle whereby things which exist or were created in one place can also show up or be created in another place, their creation being almost entirely independent of the other. This phenomenon is called "Parallel Development" or "Independent Invention." Parallel development is the process by which similar innovations, technologies, or cultural trends emerge independently in different regions or societies, without direct contact or influence between them. This phenomenon highlights how similar needs, challenges, or conditions can lead to comparable solutions or advancements across diverse contexts. Udio being a massive neural network juggling equations to create music in seconds surely makes this much more likely to trigger that effect, especially when you include the limiting factors or humanity and musical styles as parameters. Although it suggests potential copyright infringement, it alone is not proof of copyright infringement.
Just starting the video but wanted to say, if it wasn't for the AI thing I wouldn't have found either of your channels. Been watching Paul's content for about a week as well. And that's awesome, I love both of your channels!
Paul is conflating human and machine learning. They’re two completely different systems with different processes and outcomes. And because of that, in my opinion, they should be regulated differently.
And yet the output is the exact style of music that you're so afraid of being generated by AI. Clearly, whether a human or machine do the output is hardly as relevant as you are claiming. The real problem is resource flow. Because resource flow is so very unfair in so many cases in general in society, we see AI as the problem. The problem is not the technology--it has never been the technology. Technology is a neutral force. It probably will always be this way, too. The real problem is how we handle money and access to resources, and their distribution in society. Effectively, the problem is Capitalism. In our modern era, although Capitalism does some good things, in other ways it severely handicaps many of us in our ability to not be crippled by advancement in science, technology and general progress. Fairer distribution of wealth and access to resources would significantly ease this problem; and, as a bonus, would allow social power distribution in society to be more equalized, thus, effectively, create a fairer, more peaceful planet for more people. AI is the PERFECT solution for this kind of world, and yet we fear it because we know the system we live in would only use AI to relegate us to lives of destitution or worse. That is the real problem, and not much are actually talking about it. The time to start is now. Enough of the games, they're over.
Live music is "my answer" for artistic expression, not so much for $, though :) There is maybe literally a million times more recorded music being created than 30 years ago, but there's probably no more than twice the live venues.
There are some that see the writing on the wall and some who don't. Midjourney 2021 is a perfect example of where Udio is now. People are laughing, but they won't be laughing for long, because within 3 years Udio will have millions of "subscribers" creating billions of songs that you will not be able to tell if it's AI or not. The only hope there is for human creators is the 4th pillar of Fair Use. They need to trash the databases and start from scratch, which is what the NYTimes is attempting to do. These companies will not be able to tease out specific artists, they'll have to start from scratch with licensed works, in the very same way that Sample Libraries do now.
I know I just said it really can't make production music like wrestling very well. But then today I did another trail and I reckon I made professional grade Indian Production Music in it as an English and with Lakha Music's on Sonoton, Weststar and Royalty Free as well as unreleased custom tracks distinctive older mans voice. It trains off of Lakha Rathnakumer for sure this thing does. This was in custom mode Hindi, 2024 Prompt and in custom starting off with [bansuri solo] and chaining other indian instrument solos into it. But this guy though he also puts select stuff on Soundcloud and Spotify. But then so might other production artists have to to make ends meet getting normal people to spend £1 to keep on their hard drive or stream a track of it to make almost the same as a side hustle as well as live at the proms for instrumentals and live at a festival for vocal ones in the future to make a living on recorded music. But if I being British and just liking to stream production music for free on the websites sometimes for private listening can make my own production music in the Udio then the licence does have to come out. Both PRS For AI and PMA For Ai must both become taxes so we ca have fun in Udio 100% legal I'm liking the sound of this Bansuri Vision track I prompted this is full grade production music and the instruments I'm convinced are played with a lot of soul by a real person as well that track is scary.
What would Star Plus/Utsav Plus say about 1-1 soundalikes of their custom orders from Lakha Music being able to be made on Udio including his older mans voice doing the ah ah ah generating in them as well. I don't think Disney would be happy as they have been using the real man for years and probably have a high percentage of his custom tracks publishing rights.
Hey, one thing, remember all the top sports brands, addidas, nike, etc went crazy when chinese companies copied their stuff identically and sold it for little money. Now these companies changes their view. They profit by these cheap copies because they keep rhe brand in the publics focus. Just a thought
Great discussion. My 50 cents, the idea that an AI music generator that’s been trained on copyrighted music is an innocent tool and all copyright infringement falls onto the user for using certain keywords is a little bit strange. The tool spits out ready made product, instead say a string library where the user decides what to play? Like you guys say an AI music generator is pretty useless without data. And these companies intend to have for profit models while claiming fair use for the training data they use. Interesting times. Keep up the good content!!
Even if a song is a "soundalike" to another popular artist, unless the song is similar, do they really deserve compensation? Is it not just inspiration? At least these models are trained on hundreds of thousands of artists, thereby making the output more varied and drawing from countless sources of inspiration. What about bands like Greta Van Fleet, which essentially sounds like AI modelled on one band only, Led Zeppelin? Do they owe compensation to Robert Plant?
Paul must be a mind reader because everything he said is exactly my opinion on every point. Obviously, I think he’s right about all of it. I also share the experience of going from being totally against AI music to diving in head first after actually using it.
Of course I’ll continue to play and make music. The question is: Is it still à sustainable source of income. That’s the question… Respectfully, maybe Paul doesn’t really need it since it’s just a second source of income for him?
what’s worse is all the unethical people that are going to generate a song with these tools and then go “remake it in their daw" and copyright it. The service should have to publish a publicly available API that can both report any generations they have created as well as compare to anything submitted.
Ai not being copyrightable is not correct, why would Adobe and all these others be putting it in their professional products? It won't standup because the premise is that humans weren't involved, yet there can be a lot of work involved in creating an Ai image or video or music. etc. You can certainly argue that a lot of human input was put into it even if it's purely generated by AI. And this is going to get even more true as more microscopic detailed interactions we're able to perform with the AI. If i hum every part into an Ai that renders all those lines to instruments, then what is that? Is it substantively different to spending a lot time generating a ton of variations prompting lyrics and directions for it's arrangement and so on, to construct the perfect track? The answer is no. The only way the "it cant be copyrighted" works is a situation where it's just generating tracks or images all by itself and no human has done anything other than maybe provide one prompt (that may not even be their own) that you can get infinite variations out of. But don't rely on the idea that udio tracks can't be copyrighted, and if that turns out to be true, it wont last and before that you'll just need to make enough alterations and write over it and so on to make it. I see it similar to construction kits and loop libraries saying you can't use the sounds as-is. Like using the old Percussive Adventures had a full performance percussion bed ,so you cant copyright that and isnt licensed for that. Suno has made some great videos for filmmakers chosen for their Beta, but those are obviously going to be copyrightable because they've done a lot of work on their short little films not just pressed go and that's what they got from it
00:40:25 Is using the AI to create more music is a real strategy though? If TV networks, movie studios etc can use AI to generate whatever they need faster and cheaper, why would they even look for our music?
I have a different view. Imagine when Udio have their AI models to a point that they can serve CBS, Neftlix,etc, these companies will say you can only use Suno because I have an agreement with them and I save tons of money bypassing royalties. So all tv shows will have to use this models because its what the company requires. Even worst, CBS has invested in buying or developing their own AI composer for this manner. The danger I see here is that companies will find a way to save money and if AI is one way then it will grow the interest over time for this to be real.
Honestly, I think Udio will be there by the end of this year and what you're thinking is exactly what will happen. So, it's to a musician's benefit to learn everything you can about these generators because the person who does the prompting for these companies is probably going to be a musician who knows how to manipulate them quickly. Udio actually gives you a lot of control within a prompt by using brackets. I've got it to do a lot of things I didn't think were possibly until I tried. Also, stems are coming and you'll be able to work with clean files in your DAW. What I think people who haven't used the generators don't see is that you actually have to do some work to get a good result. It's faster to search a library at the moment. But that's going to change quickly. Best to learn now.
@@alexadigitalradio I also think video software will have AI music integrated so now you can have music generated for the scen you are working and will fit perfectly on time and emotion. So I do think it will get hard for composers to make money
If art artist couldnt stop Midjourney, Dalle 3, and Stable Difussion. What makes you guys think Musicians can do anything about Udio and Suno? A bunch of the stuff you're saying is what Art Artist said last year. Thet spent all last year trying to fight Ai, don't make the same mistake they did. If you can't beat em, join em. Now if you're main argument is how much one should get paid and focused mainly on money then you guys wont last 1 year with this argument.
Yoo bro I loved your content I loved that you shared about sync music but i am not really able to understand will this work for me i am a hip hop music producer. Please help me 🙏🏽 i will be waiting for your reply bro can i actually generate monthly income??
very interesting conversation, I completely agree with you, thank you guys! and don't be afraid that AI will replace you, the only ones who will use this kind of tools for professional purposes will be people who are already artists and or people who have been thinking about it for a long time but who never had the motivation or the time before that, and people who can't afford to buy "real music" for their works like indie video games developed by a small team for exemple. maybe even that some people will become real artists after generating music and this will awaken the flame within them., even if it is easy to generate music you must have the passion, the equipment, the knowledge, the mastery of professional language, taste, the time etc etc... and the world continues to turn people are not going to change jobs to sell music generated. have a good evening/day
oh nice strawman arguments. I was one of the few arguing against people making similar arguments 20 years ago, that copying mp3s was harmless. I don’t pirate music. I don’t pirate software. It’s precisely because I don’t that I hold we are being robbed. Training should cost a larger multiplier, e.g. every track you train on you need to pay as if you had bought it N number of times. , which should make it cost prohibitive to do this on the cheap. Also, any AI generated song should have to demonstrate lineage in showing which weights contributed the most, and those should be demonstrably mapped to the training data, and all of this is not currently possible which would gate the immediate ability to monetize this tech. Once It is technically possible to credit back the community of contributors in the way that BMI, etc., track and pay, and once these companies have paid upfront for training on copyrighted music, then go crazy.
Paid mode came in today meaning less people are going to make less tracks on Udio now. And one thing I tried it can't even do anyway is the Wrestling Genre with the Production Music sound. Udio can't do production music. In Instrumental I even tried Wrestling as well and below Production grade instrumentals came out of that one. Production Music itself has reused Production Music for years because DAW Stems are production music material and I think Udio does train off of those though with the [Instrument solo] prompt used to generate a stem to use from it. But a Sony Stems or a Sound Ideas Stems and whatever else pack goes under Production Eliments rather than production music and even humans train on them to an extent so Udio can too. As it can't do wrestling I don't think it trains on the real pre built production songs but everything production and mainstream human and now AI trains on the eliment packs. The vocalists are fishy though like they do train the mainstream tracks but in the past packs had generic vocal stems in them as well from non namers. And then you have the people on real deals for vocals like Will I Am confirmed to be a vocalist on a real deal for Udio. They definitely train on melodys with their stem packs, vocalist deals thats for sure but then so do humans as the other guy pointed out and your custom music orders come from prompts as does AI producing. Where does Scurry from Non-Stop turning into What Does Everybody Want Head for Al Snow in WWE and many of the Disney Star Plus tracks being made to sound like an artist from Sonoton go into it those TH-cam dub themes on Imlie although to be fair the real artist of the original Sonoton tracks works on the show himself maybe he replaced his own production track with his own custom seeing as he makes custom music for that show Lakha Music does but then maybe Tapas Relia ripped him off again like he made the midi version of the theme with the other guys vocal stem still in it. Many humans copy production music and with it off with custom versions without paying the custom track single payment lisence off which you are supposted to when they don't want to renew the annuals and want to make their own custom version. What has been happening for years is now happening in AI but then the way I think is waht happening in AI is whats happening in Production Music itself with the custom orders on proper lisence anyway as well as the making their own custom version without paying the lisence at all it's all made using the same stems packs that is in both a DAW and an AI and both the human and AI uses prompts. The problem is the lack of a yearly tax that AI Companies shoud pay to train off of the old melodies using these stem packs. Ok, many humans don't pay it and do their own with the orignal production tracks still in the background or even going as far as reproducing it through a human guuitar playing but there is a human version called the custom music orders lisence these AI companies need their own tax and they must pay it even though the humans often don't get caught not paying their custom orders lisence the humans in charge of AI though must pay to get their melody training rights or in other words opt in with paying or opt out and not being able to train on that library at all. But also as part of opt in opt out on the lisence AI can get a genre on the libraries and only the most skilled prompters get their tracks released by the companies AI range. I do think AI has a future in even ordanary tracks because [instrument solo] is a stem so I can see them being good DAW stem generators in fact there is already audio sound generators to do the stem creating but Udio might be higher quality stems on prompts in the custom mode.
AI is doing the same thing as musicians, except it is faster and more efficient. Any musician can listen to other people's music without having consent to train themselves off of copyrighted music. Humans don't need consent. I know musicians who can hear a piece once and then be able to reproduce it forever. There are protections in place that make it so they can't profit off playing that exact piece, but they can train on it and profit off of their own version. Not much different than AI. The reality is most music, especially popular music is iterations of older music we have heard before. Labels have been following a model for decades to keep reproducing the same music. Labels almost always pass on something that is different, and there are a hundred stories where that "different" band went an independent route to release something that became a major use. Labels are influencing TH-cam to demonetize everything and it really has distorted what is allowed and what isn't. There should be a way to license another artist's voice or style.
But how did those musicians access that copyrighted music in the first place? They bought CD’s/Cassettes, Vinyls, heard them on the radio, streamed them on Spotify, bought concert tickets, purchased sheet music, etc etc. Human beings don’t scrape the entirety of recorded music for free, Eric… Also, objectively, AI doesn’t do what musicians do at all. People can keep saying it does but it doesn’t make it true.
@@SC-ew2fc While it's true that humans don't scrape the entirety of the internet for free, are you really saying that this is the big problem with AI? That Suno and Udio would be perfectly acceptable if they had only first purchased a $20/month TH-cam premium account? I jest, because obviously, that isn't what most people are finding objectionable about AI music systems.
![[UNCUT] The Loyal Pin ปิ่นภักดิ์ EP.15 (1/4)](http://i.ytimg.com/vi/TLBjiHvkKtE/mqdefault.jpg)
I don't see AI being anything more than using loops to create tracks. It is just quicker as it will work faster than we can with searching and listening to different loops, vst instruments, etc. It is just a way to get ideas started quicker. Caveat: I don't use AI or full music loops but I do use drum programs/sequencers for grooves and fills to create tracks.
I think WE should stop giving AI so much attention and start looking as to why a living human being is far superior in every way...
_Sure,_ *1. adaptability.*
30:51 I couldn’t disagree more. Yes it’s true that oftentimes sync music is low in the mix for obvious reasons, but that doesn’t mean that we waste time trying to make it as good as possible. Nor is that a reason to choose AI music that’s not as good, just because it’s low in the mix.
1. There are cheaper cameras. Why not just film your show with a cheaper camera? Why not just hire a cheaper writer? The production is the sum of its parts, and high quality across the board will produce a better product. Whereas short cuts and cost cuttings usually have a detrimental impact on the end product.
2. This is a really obvious one. The music might be under dialogue for a while, but oftentimes it jumps up into the foreground for a while, and then ducks back down below dialogue. Who in the right mind would want a rubbish track playing when it jumps up in volume? Don’t underestimate the value of high quality sound and music, and how that adds the polish to a show. If Paul doesn’t give his 100% on every cue because it’s going to be low in the mix anyway - well speak for yourself ;)
I'm not suggesting at all that you accept poor quality, or ignore every aspect of your production. I'm pointing out the very irony of our music being hard to hear as a reality check for those that say AI music is not good quality from an audio perspective. Some people are saying that AI tracks are not mixed properly, or sound AI-generated. My point is "so what." At the end of the day, whether or not the track sets the right mood for the editor is what matters. AI will only get better, and much faster than most understand. To be clear: ALWAYS deliver your best product... then hope you can get a nice forefront placement. :)
This was an amazingly thoughtful and intelligent conversation. Thank you Jesse and Paul. You’ve subtracted the emotion and fear and presented a very comprehensive set of ideas that could serve as a fine basis for creating the social framework for AI in art and content generation. Thank you both.
Great insights by both of you here. I'm not in the sync music licensing space- I'm a classically-trained musician, and I have played cover music in a lot of different genres for the past 50 years. But these issues about AI-generated music are interesting, and all of us musicians should stay informed about what's going on.
Any musician who plays covers, and very few of us don't, is technically plagiarizing others' intellectual works. Most music that we cover is also copyrighted. But cover bands and solo musicians are widely accepted and hired by venues. Their customers mostly want to hear familiar music- things they've heard before. Example: I'm in a Selena tribute band (the Tex-Mex pop star who was killed early in her career). None of the people who come to our shows cares about the fact that we're plagiarizing Selena's music. They just want to re-live the 90s and the feelings that they had listening to her music and watching her shows.
Emulating, plagiarizing, copying, learning from copyrighted music is a part of being a musician. Particularly a musician who plays covers. We stand on the shoulders of the musical giants who came before us. I've also heard it said that "Imitation is the sincerest form of flattery."
A.I. being trained on existing copyrighted music? I don't really care about it. All of us train ourselves as musicians on copyrighted music. Very few of us only play music that we've composed and recorded. Am I concerned about AI music taking over? Not much- we musicians are probably more concerned with DJs taking over our gigs, because so many club owners are cheap bastards and they don't want to pay bands. Okay, that's not totally fair, because club owners are trying to run a business, and they have to make a cost/benefits decision on whether it's more advantageous for them to hire a DJ or live musicians. But that's a whole 'nother topic.
Great dialogue. Thanks
Paul!! Love Paul’s channel!!
Personally my biggest gripe with big tech is the industry of mining and scraping data, especially personal data. When you create a free social media, you are allowing them to exploit pretty much everything post, say, upload, your browsing history etc...
It seems there are many Paul's out there, that have no issues with this.
This is why opt in is important.
Paul can opt in to have his 1,500 cues trained on by AI companies who won’t compensate him
And we can all opt out. Win win!
If AI music gets popular or at least ubiquitous in advertising and movies, it will move that music in the direction of what it was trained on - specifically, not your stuff.
@@Desirsar good. So I’ll stand out then
@@jmi_music Noisecore stands out, it doesn't get sync licenses.
For software engineers we were suddenly and swiftly thrown into "retirement" even for those in their mid-20s, I don't think anyone CAN plan on retiring at 27.
I've finished Bansuri Dreams and it really did sound like after 3 hours of manual crafting graft that it's 100% a track that is all Lakha Music Sonoton/Weststar/Royalty Free and Custom Unreleased all in the same track. I thought it was Lakha himself playing Indian not AI. I think the production music artists might have something to put forward a lawful case for a PMA, PRS and Cocatalog licence where the owners of the site don't pay it would be law breaking because not paying is opting out like the TV Licence in the UK where you can't watch any TV unless you pay the BBC to watch all channels.
Paul is right that regulating this will be next to impossible as generative AI will calibrate to be just one notch below violating copyright.
I think you are probably right. Seems unstoppable to me.
This was a great interview!
Great talk guys! Tons of good points to ponder...
:-)
Big thank you! Great discussion.
Love it. This was a wonderful piece of dialogue about a hot topic. I really like Pauls attitude towards creating music and learning new tools to create better. I am definitely in camp Paul as a creator. Ai is a superb tool for me. I can throw ideas at it and get a result back in just seconds. It instantly gives me a way to judge if my idea is good or not and if it gives me material that I can bring into my DAW to work with. It allows me to try things that I would otherwise waste days on before finding out it was a crappy idea. Real timesaver.
But. Looking a bit into how the legal contracts of the Ai services it seems to put all responsibility about if the content created is copyrighted material on the end users. I am making music for fun and I don't care if I don't make any money off it, but I sure as hell don't want to get sued for something that I have never heard and had no idea it was not legal to use. The logical solution would be to impose on Ai services a part to the algorithm that makes it impossible for it to generate pieces that is too much alike the original piece to be a legal problem. But listening to Paul it seems like it is not even clear in the law how to judge what is legal and not. Am I right here? Is it impossible to make sure the outputs of Ai stays clear of copyright issues?
great discussion! appreciated the details of how the Google LM trained on music - random 10sec samples, easy to defend legally i'd imagine. Training model protocol (essentially explaining HOW input is selected) / and subject matter should definitely be public information. that would be very educational and interesting.
Really interesting, great conversation... two good knowledgeable experienced talented and wise people banging out their thoughts and advice. so much to sift thru, but bottom line we've no idea what the future holds, but embracing/learning the new way of doing things is at hand, and the sooner the better, because it's the future, and its legalities will work themselves out because they have to.
#OdedFriedGaon #OdedMusic #OdedInformation
This generation's heroes are the last generation's hacks. Paul made a great point about the musicians we think are innovative today, who are really taking influence and borrowing from previous musicians. Every sync opportunity starts with, "We need something that sounds like X, Y, or Z." They never ask for something unique. Entire genres are based on the same guitar riff or bass lines.
As soon as musicians started making complete songs with loops and library content, this was inevitable. How many times have we heard a song with the AMON break?
On the Red Lobster example, I think you are missing the point that it is a parody, and it would be protected the same as any other parody would be today... not a violation. Also, it doesn't take away from the original; it doesn't compete with the sales of the original or turn people off from listening to the original song. People have been making tracks like this for years before AI; now, it's just better at it. Also, AI is used in every aspect of the product process... where would you draw the line. Companies have been using covers of popular songs for decades because it is cheaper.
Great chat!
Excellent discussion!
I'm with Paul on this, and I know it's not a popular viewpoint. Imagine, though, that copyright worked the way most musicians seem to want it to: where training on copyrighted works first requires permission from and payment to the copyright holder. Like Paul pointed out, we all study copyrighted works. We don't get permission and we don't necessarily pay the copyright holder. And most of us take what we've learned from this studying and write new music inspired by what we've heard and if we're professional musicians, we're using that learning for commercial gain (for-profit). If the law works the way most people seem to want it to, no one would be able to listen to any copyrighted music and then write music unless we first get permission from everyone whose music we've ever listened to. It's untenable.
AI is doing the same thing. It's studying music (copyrighted or otherwise) and then making new music based on what it's learned. In my view, that doesn't require permission or payment.
Jesse, with respect, I don't think you are viewing the 4 elements of fair use properly. Those are for copying. But the AI systems aren't copying. They are making new music that did not exist before. It's not like sampling, for example, where exact copies are being used. In most cases, these systems are outputting stylistically-generic music that's bound to sound a bit like existing music because that's what "country" or "rock" music sounds like. I challenge anyone to make Suno or Udio produce an actual copy of, say, "Bridge Over Troubled Water." That's not what these systems are designed to do, and for the most part, they don't.
The "copying" is at the input side--into the training set--not at the output side: the music the AI system generates. Copyright has no concept of "unauthorized studying"; it regulates the output, not the input.
And that's what we want because otherwise we are all guilty of "stealing" other people's music. Because all of our music careers are based on hundreds of hours of listening to and studying other people's copyrighted music. But those are inputs to our brains. What copyright cares about is if our OUTPUTs are copies of other people's copyrighted works, not if our inputs are.
That said, wouldn't it be nice if we made special rules for use of copyrighted music in AI training sets? I'm for that, if we can figure out a way. Maybe something similar to the compulsory licenses for recording cover songs?
Just wait till you see my video next week. You might have a new understanding of what these models are doing...
@@SyncMyMusic I look forward to the video! But even if one is able to force the system to generate a copy of a song, that is an outlier case--it's not what the systems are designed to do, and tarring them with the broad brush that they are copyright infringing overall is still wrong in my view. That said, my experience with these tools is that the more specific you need the output to be, the harder it is to get the system to generate it. I would love to see a video of how someone coaxes such specific output from these systems because those techniques could also potentially help me get specifics I need in songs I create with these systems. My use case is a bit different in that I am a training professional and my interest is in creating instructional novelty songs--sort of on the "Schoolhouse Rock" model--to help convey some of the key messages of the training sessions in a fun way.
It’s like if you want to iron out the conversation of charging ai to train you then who and how many gets to charge you from training them? For example we know hip hop pioneers. We know when you aren’t one but you’ve made a living studying training on it. So I believe when he says it’ll be a mathematical mess it’s giving those who fall under if ai is to be charged then you should too a way out if I’m understanding him correctly and it’s definitely a conversation as a whole if anyone wants to continue because what’s the difference between their training for use of for profit company and yours? Speed and connections. The same structure and process is still there and no one expects you to license everything you trained on even if it was a lifetime full of work and a computer chip has multiple lifetimes full of work. This was like a oversight at the beginning of internet when you choose to opt in. Maybe that’s why so many fought on this topic then about what’s still yours on the web. Very interesting topic that shows great courage to at least try to find fairness at least going forward for yourself. Nothing is to hard to talk or figure it out especially when on some level it’s already being done. If we agree that the computer is so powerful then why not rewrite the code so that all training can be licensed even on the micro. That would be great and traceable to pay homage to those being trained on even longer. Let’s do it. That means schools too would owe for teaching those same tracks. My school constantly encourages me in active assignments to download Spotify to learn from something. I’m just not into it because of other work that clouds my priority then listening to music but that’s for a grade at a for profit school so if ever it’s a time to consider changes it’s now. Let these people and their legacies get paid now even for training by all. It could be a major flex to be on file as trained using one of the greats.
I think the part 4 of the “fair use” agreement still stands.
I could be very wrong but in my own opinion typing prompts into a computer to generate music especially if the prompts are similar will result in similar music being generated and ownership battles along with copywriting problems will begin.
Udio's terms don't even deny this. They concede their outputs may infringe on copyright but it's the users who are liable. They also state multiple users can generate the same outputs from similar inputs. They seem to be creating a total copyright minefield. It's why comparisons to the visual art industry is misguided. Music IP has pretty cut and dry ownership and people will be wayyy more eagle eyed (eared?) when it comes to AI music. Musicians sue each other all the time for copyright infringement which is almost always accidental. Imagine hundreds of thousands of people generating near infinite tracks with the same generic prompts. It'll be such a mess.
@@SC-ew2fc Thanks for confirming my suspicions I'm definitely not opposed to AI in certain roles in fact I love Izotope and everything they offer. I fully agree this is gonna get messy before it gets sorted out.
It would be incredibly unlikely that this happens. There are exponentially more varaibles involved in creating the outputs... just look at the odds of winning the Lotto. They are infintisemal. Well, the statistical likelihood of AI generating the same track are a fraction of a fraction of a fraction of the chance of winning the Lotto. :)
@@YoPaulieMusic I saw on a forum that someone had generated a track on Udio, but not published it. Then they heard portions of this track in another users generation on the homepage under “staff picks.” They were trying to figure out how that could happen, but it has. It’s probably why Udios terms explicitly state it.
It’s probably why you’re also seeing composer contracts with studios forbidding them to use any generative AI in their scores. If they can’t own the music, what’s the point.
@@SC-ew2fc
It's not a copyright minefield because Udio effectively claims they have ultimate ownership of all the creations generated in Udio, as per their "Attribution" subsection in the Terms of Service. Attribution is required whenever each piece is used for any reason other than mere personal, private use.
Also, claiming that Udio generations can potentially infringe on Copyright works as being a tacit declaration of guilt is not entirely correct. One major reason is because the reality in which we reside has a principle whereby things which exist or were created in one place can also show up or be created in another place, their creation being almost entirely independent of the other. This phenomenon is called "Parallel Development" or "Independent Invention."
Parallel development is the process by which similar innovations, technologies, or cultural trends emerge independently in different regions or societies, without direct contact or influence between them. This phenomenon highlights how similar needs, challenges, or conditions can lead to comparable solutions or advancements across diverse contexts.
Udio being a massive neural network juggling equations to create music in seconds surely makes this much more likely to trigger that effect, especially when you include the limiting factors or humanity and musical styles as parameters.
Although it suggests potential copyright infringement, it alone is not proof of copyright infringement.
Just starting the video but wanted to say, if it wasn't for the AI thing I wouldn't have found either of your channels. Been watching Paul's content for about a week as well. And that's awesome, I love both of your channels!
Paul is conflating human and machine learning. They’re two completely different systems with different processes and outcomes. And because of that, in my opinion, they should be regulated differently.
And yet the output is the exact style of music that you're so afraid of being generated by AI. Clearly, whether a human or machine do the output is hardly as relevant as you are claiming. The real problem is resource flow. Because resource flow is so very unfair in so many cases in general in society, we see AI as the problem. The problem is not the technology--it has never been the technology. Technology is a neutral force. It probably will always be this way, too. The real problem is how we handle money and access to resources, and their distribution in society. Effectively, the problem is Capitalism. In our modern era, although Capitalism does some good things, in other ways it severely handicaps many of us in our ability to not be crippled by advancement in science, technology and general progress. Fairer distribution of wealth and access to resources would significantly ease this problem; and, as a bonus, would allow social power distribution in society to be more equalized, thus, effectively, create a fairer, more peaceful planet for more people. AI is the PERFECT solution for this kind of world, and yet we fear it because we know the system we live in would only use AI to relegate us to lives of destitution or worse.
That is the real problem, and not much are actually talking about it. The time to start is now. Enough of the games, they're over.
@@MagnesiumEnterprise Actually I have no idea what you're trying to say in this wall of text
Live music is "my answer" for artistic expression, not so much for $, though :)
There is maybe literally a million times more recorded music being created than 30 years ago, but there's probably no more than twice the live venues.
There are some that see the writing on the wall and some who don't. Midjourney 2021 is a perfect example of where Udio is now. People are laughing, but they won't be laughing for long, because within 3 years Udio will have millions of "subscribers" creating billions of songs that you will not be able to tell if it's AI or not. The only hope there is for human creators is the 4th pillar of Fair Use. They need to trash the databases and start from scratch, which is what the NYTimes is attempting to do. These companies will not be able to tease out specific artists, they'll have to start from scratch with licensed works, in the very same way that Sample Libraries do now.
I know I just said it really can't make production music like wrestling very well. But then today I did another trail and I reckon I made professional grade Indian Production Music in it as an English and with Lakha Music's on Sonoton, Weststar and Royalty Free as well as unreleased custom tracks distinctive older mans voice. It trains off of Lakha Rathnakumer for sure this thing does.
This was in custom mode Hindi, 2024 Prompt and in custom starting off with [bansuri solo] and chaining other indian instrument solos into it.
But this guy though he also puts select stuff on Soundcloud and Spotify. But then so might other production artists have to to make ends meet getting normal people to spend £1 to keep on their hard drive or stream a track of it to make almost the same as a side hustle as well as live at the proms for instrumentals and live at a festival for vocal ones in the future to make a living on recorded music.
But if I being British and just liking to stream production music for free on the websites sometimes for private listening can make my own production music in the Udio then the licence does have to come out.
Both PRS For AI and PMA For Ai must both become taxes so we ca have fun in Udio 100% legal I'm liking the sound of this Bansuri Vision track I prompted this is full grade production music and the instruments I'm convinced are played with a lot of soul by a real person as well that track is scary.
What would Star Plus/Utsav Plus say about 1-1 soundalikes of their custom orders from Lakha Music being able to be made on Udio including his older mans voice doing the ah ah ah generating in them as well. I don't think Disney would be happy as they have been using the real man for years and probably have a high percentage of his custom tracks publishing rights.
Hey, one thing, remember all the top sports brands, addidas, nike, etc went crazy when chinese companies copied their stuff identically and sold it for little money. Now these companies changes their view. They profit by these cheap copies because they keep rhe brand in the publics focus. Just a thought
Great discussion. My 50 cents, the idea that an AI music generator that’s been trained on copyrighted music is an innocent tool and all copyright infringement falls onto the user for using certain keywords is a little bit strange. The tool spits out ready made product, instead say a string library where the user decides what to play? Like you guys say an AI music generator is pretty useless without data. And these companies intend to have for profit models while claiming fair use for the training data they use. Interesting times. Keep up the good content!!
Even if a song is a "soundalike" to another popular artist, unless the song is similar, do they really deserve compensation? Is it not just inspiration? At least these models are trained on hundreds of thousands of artists, thereby making the output more varied and drawing from countless sources of inspiration. What about bands like Greta Van Fleet, which essentially sounds like AI modelled on one band only, Led Zeppelin? Do they owe compensation to Robert Plant?
Paul must be a mind reader because everything he said is exactly my opinion on every point. Obviously, I think he’s right about all of it. I also share the experience of going from being totally against AI music to diving in head first after actually using it.
I mainly agree with Paul.
Of course I’ll continue to play and make music. The question is: Is it still à sustainable source of income. That’s the question… Respectfully, maybe Paul doesn’t really need it since it’s just a second source of income for him?
what’s worse is all the unethical people that are going to generate a song with these tools and then go “remake it in their daw" and copyright it. The service should have to publish a publicly available API that can both report any generations they have created as well as compare to anything submitted.
Would you prompt generative music from 9 to 5 at minimum wage? Cause we won’t get paid much more than that to do it…
Computers don’t have rights. They shouldn’t be allowed to “train” on copyrighted works without permission under the doctrine of fair use.
Ai not being copyrightable is not correct, why would Adobe and all these others be putting it in their professional products? It won't standup because the premise is that humans weren't involved, yet there can be a lot of work involved in creating an Ai image or video or music. etc. You can certainly argue that a lot of human input was put into it even if it's purely generated by AI. And this is going to get even more true as more microscopic detailed interactions we're able to perform with the AI. If i hum every part into an Ai that renders all those lines to instruments, then what is that? Is it substantively different to spending a lot time generating a ton of variations prompting lyrics and directions for it's arrangement and so on, to construct the perfect track? The answer is no. The only way the "it cant be copyrighted" works is a situation where it's just generating tracks or images all by itself and no human has done anything other than maybe provide one prompt (that may not even be their own) that you can get infinite variations out of. But don't rely on the idea that udio tracks can't be copyrighted, and if that turns out to be true, it wont last and before that you'll just need to make enough alterations and write over it and so on to make it. I see it similar to construction kits and loop libraries saying you can't use the sounds as-is. Like using the old Percussive Adventures had a full performance percussion bed ,so you cant copyright that and isnt licensed for that. Suno has made some great videos for filmmakers chosen for their Beta, but those are obviously going to be copyrightable because they've done a lot of work on their short little films not just pressed go and that's what they got from it
00:40:25 Is using the AI to create more music is a real strategy though? If TV networks, movie studios etc can use AI to generate whatever they need faster and cheaper, why would they even look for our music?
44:08 Maybe I could be the person Red Lobster reach out to??? All my dreams come true! Sign me up to the AI. Let’s go!
I have a different view. Imagine when Udio have their AI models to a point that they can serve CBS, Neftlix,etc, these companies will say you can only use Suno because I have an agreement with them and I save tons of money bypassing royalties. So all tv shows will have to use this models because its what the company requires. Even worst, CBS has invested in buying or developing their own AI composer for this manner. The danger I see here is that companies will find a way to save money and if AI is one way then it will grow the interest over time for this to be real.
Honestly, I think Udio will be there by the end of this year and what you're thinking is exactly what will happen. So, it's to a musician's benefit to learn everything you can about these generators because the person who does the prompting for these companies is probably going to be a musician who knows how to manipulate them quickly. Udio actually gives you a lot of control within a prompt by using brackets. I've got it to do a lot of things I didn't think were possibly until I tried. Also, stems are coming and you'll be able to work with clean files in your DAW. What I think people who haven't used the generators don't see is that you actually have to do some work to get a good result. It's faster to search a library at the moment. But that's going to change quickly. Best to learn now.
@@alexadigitalradio I also think video software will have AI music integrated so now you can have music generated for the scen you are working and will fit perfectly on time and emotion. So I do think it will get hard for composers to make money
If art artist couldnt stop Midjourney, Dalle 3, and Stable Difussion. What makes you guys think Musicians can do anything about Udio and Suno? A bunch of the stuff you're saying is what Art Artist said last year. Thet spent all last year trying to fight Ai, don't make the same mistake they did. If you can't beat em, join em.
Now if you're main argument is how much one should get paid and focused mainly on money then you guys wont last 1 year with this argument.
Yoo bro I loved your content I loved that you shared about sync music but i am not really able to understand will this work for me i am a hip hop music producer. Please help me 🙏🏽 i will be waiting for your reply bro can i actually generate monthly income??
very interesting conversation, I completely agree with you, thank you guys! and don't be afraid that AI will replace you, the only ones who will use this kind of tools for professional purposes will be people who are already artists and or people who have been thinking about it for a long time but who never had the motivation or the time before that, and people who can't afford to buy "real music" for their works like indie video games developed by a small team for exemple.
maybe even that some people will become real artists after generating music and this will awaken the flame within them., even if it is easy to generate music you must have the passion, the equipment, the knowledge, the mastery of professional language, taste, the time etc etc... and the world continues to turn people are not going to change jobs to sell music generated. have a good evening/day
oh nice strawman arguments. I was one of the few arguing against people making similar arguments 20 years ago, that copying mp3s was harmless. I don’t pirate music. I don’t pirate software. It’s precisely because I don’t that I hold we are being robbed.
Training should cost a larger multiplier, e.g. every track you train on you need to pay as if you had bought it N number of times.
, which should make it cost prohibitive to do this on the cheap. Also, any AI generated song should have to demonstrate lineage in showing which weights contributed the most, and those should be demonstrably mapped to the training data, and all of this is not currently possible which would gate the immediate ability to monetize this tech. Once It is technically possible to credit back the community of contributors in the way that BMI, etc., track and pay, and once these companies have paid upfront for training on copyrighted music, then go crazy.
Paid mode came in today meaning less people are going to make less tracks on Udio now. And one thing I tried it can't even do anyway is the Wrestling Genre with the Production Music sound. Udio can't do production music. In Instrumental I even tried Wrestling as well and below Production grade instrumentals came out of that one.
Production Music itself has reused Production Music for years because DAW Stems are production music material and I think Udio does train off of those though with the [Instrument solo] prompt used to generate a stem to use from it.
But a Sony Stems or a Sound Ideas Stems and whatever else pack goes under Production Eliments rather than production music and even humans train on them to an extent so Udio can too.
As it can't do wrestling I don't think it trains on the real pre built production songs but everything production and mainstream human and now AI trains on the eliment packs.
The vocalists are fishy though like they do train the mainstream tracks but in the past packs had generic vocal stems in them as well from non namers. And then you have the people on real deals for vocals like Will I Am confirmed to be a vocalist on a real deal for Udio.
They definitely train on melodys with their stem packs, vocalist deals thats for sure but then so do humans as the other guy pointed out and your custom music orders come from prompts as does AI producing.
Where does Scurry from Non-Stop turning into What Does Everybody Want Head for Al Snow in WWE and many of the Disney Star Plus tracks being made to sound like an artist from Sonoton go into it those TH-cam dub themes on Imlie although to be fair the real artist of the original Sonoton tracks works on the show himself maybe he replaced his own production track with his own custom seeing as he makes custom music for that show Lakha Music does but then maybe Tapas Relia ripped him off again like he made the midi version of the theme with the other guys vocal stem still in it.
Many humans copy production music and with it off with custom versions without paying the custom track single payment lisence off which you are supposted to when they don't want to renew the annuals and want to make their own custom version.
What has been happening for years is now happening in AI but then the way I think is waht happening in AI is whats happening in Production Music itself with the custom orders on proper lisence anyway as well as the making their own custom version without paying the lisence at all it's all made using the same stems packs that is in both a DAW and an AI and both the human and AI uses prompts.
The problem is the lack of a yearly tax that AI Companies shoud pay to train off of the old melodies using these stem packs.
Ok, many humans don't pay it and do their own with the orignal production tracks still in the background or even going as far as reproducing it through a human guuitar playing but there is a human version called the custom music orders lisence these AI companies need their own tax and they must pay it even though the humans often don't get caught not paying their custom orders lisence the humans in charge of AI though must pay to get their melody training rights or in other words opt in with paying or opt out and not being able to train on that library at all.
But also as part of opt in opt out on the lisence AI can get a genre on the libraries and only the most skilled prompters get their tracks released by the companies AI range.
I do think AI has a future in even ordanary tracks because [instrument solo] is a stem so I can see them being good DAW stem generators in fact there is already audio sound generators to do the stem creating but Udio might be higher quality stems on prompts in the custom mode.
Yoo reply to my comment bro don't just leave it liked
AI is doing the same thing as musicians, except it is faster and more efficient. Any musician can listen to other people's music without having consent to train themselves off of copyrighted music. Humans don't need consent. I know musicians who can hear a piece once and then be able to reproduce it forever. There are protections in place that make it so they can't profit off playing that exact piece, but they can train on it and profit off of their own version. Not much different than AI.
The reality is most music, especially popular music is iterations of older music we have heard before. Labels have been following a model for decades to keep reproducing the same music. Labels almost always pass on something that is different, and there are a hundred stories where that "different" band went an independent route to release something that became a major use.
Labels are influencing TH-cam to demonetize everything and it really has distorted what is allowed and what isn't.
There should be a way to license another artist's voice or style.
But how did those musicians access that copyrighted music in the first place? They bought CD’s/Cassettes, Vinyls, heard them on the radio, streamed them on Spotify, bought concert tickets, purchased sheet music, etc etc. Human beings don’t scrape the entirety of recorded music for free, Eric…
Also, objectively, AI doesn’t do what musicians do at all. People can keep saying it does but it doesn’t make it true.
@@SC-ew2fc While it's true that humans don't scrape the entirety of the internet for free, are you really saying that this is the big problem with AI? That Suno and Udio would be perfectly acceptable if they had only first purchased a $20/month TH-cam premium account? I jest, because obviously, that isn't what most people are finding objectionable about AI music systems.