As a songwriter and musician with tons of experience in the studio it is so refreshing to hear your perspective on using AI to generate your artistic vision. Making a decent AI track is difficult, it takes hour and hours of work and a significant knowledge of many aspects of music production, including the economics of music production.
For most creators, the sheer cost of producing a song like this in the studio would simply be overwhelming and unrealistically time consuming.
Don’t be fooled by those that poo-hoo AI generate music, while it is true that it can make crappy lyrics and a lack of musical vision sound better than it deserves, this has been true since the first reverb and multitrack recorder were invented.
I am equally as proud of my AI songs as I am my ground up studio productions
I'm looking forward to the golden age of AI when tools like Suno can be run on local hardware, thus bypassing whatever censorship TPTB may try to impose on AI.
The way you described what actually is making music with AI is probably the best I have ever seen, it's also how I think about it. I think the alternative rock aesthetic really fits the vibe!
There will eventually be a way to do it, and probably within the next 18 months. That's what the pros really want, because they don't want to leave it up to the machine to produce the final mix when they can reliably do better. There is no technological reason it can't be done, but it has to be properly designed and programmed.
The vocals and instrumentals can be stemmed, but the instrumentals cannot be usefully further divided. And yes, someone far more technical with regards to these matters tried it at my request. As he concluded, it's like baking a pie, then trying to separate all the ingredients from it and baking another pie with them.
"It’s more like working with hundreds of musicians and telling one guitarist that you like his riff, a bassist that you like his groove, and then selecting from between the various interpretations of a lyric from the fifty vocalists you’ve got at your disposal. In other words, it still requires making all the structural decisions as a producer, only with totally insane machines providing the input instead of moody musicians."
This sounds like the recording sessions with Steely Dan in the 70s. The SD guys were into sci-fi, they probably would have preferred insane machines to actual moody musicians.
Sounds pretty good, but how do you know that you're not accidentally infringing on another artist's music? The gamma lawyers who lawyer for these major labels seem to love trying to make something out of nothing.
Because the technology doesn't work that way. Lawyers are, by and large, stupid. Because music can be broken down into mathematics, and because it is broken down that way for the AI generation of music, it can be very easily shown how the concept is not even relevant.
The very concept of AI "imitating" human art is false. The training just involves showing it which patterns should be prioritized and which should be deprioritized. Think about the mathematical improbability of randomly getting even 10 notes in a row the same.
Throw in chord progressions and original lyrics, and it's not happening. And "look and feel" is not protectable anyhow. FFS, the guitar riff from "Smells Like Teen Spirit" is just a spin on the one from "More Than a Feeling".
I would love to see an example prompt. Not a serious musician, but i've played around with suno quite a bit for fun, and i cannot get it to cooperate if i have more than 3 suggestions.
Harder than it looks, isn't it! I don't use prompts, for the most part. I write the lyrics, then use one of the preset Personas I've saved to get a moderately consistent sound. The system uses the lyrics as the prompt.
It's like Twitter. Anyone can do it, but not everyone can do it effectively.
hah. truth. I've got a couple good ones but they literally took forever to get right. Your male vocals are so much better and different than anything I've heard so far. I've had difficulty with getting a deeper metal voice like James Hettfield or lemmy. Femal vocals seem a bit easier.
Primarily Suno. I've tried most of the others, and while they have their advantages, I find that even with its inferior tools, I can get better results out of it.
I don't think people will realize how cutting edge this is. Most of the time when you ask an AI to "bring back the last version" or something similar it will go, I don't know. SORRY. And sometimes minor edits or nuance will make it glitch out. Even AI advocates struggle to explain the black box. I finally tried video generation and it's clunky but I see the potential. Took me several prompts and testing different generators to get characters consistent. AI will race change quite liberally from scene to scene. Modifying vocal prompts must be quite the bear.
Yes, it's somewhat amusing how pretty much everyone dismisses it as "oh, it's just AI" except for all the serious musicians, who react with "Holy cow, how did you do THAT!"
The ability to create and control the vocals has been a dream for decades for a lot of composers because it's so hard to find anyone who can sing well, and who doesn't want to just imitate Whitney Houston or the singer du jour.
As a classical music buff, AI-generated classical music would be cool. Choose an approximate time period and genre, maybe enter your own theme or themes, and get a decent coherent piece complete with convincing sonata-allegro form or whatever you want. Haven't seen anything like that yet, but I expect it's just a matter of time.
Exactly this! I actually see independent artists having a lot of creative potential because you can now do all your own post production, editing out um and uh, bottom thirds, and you can still add in novel b roll or voice recordings. It's very fun as a hobbyist so I am excited to see more talent break into this area. Making something in a few days of focus when ive handed complete storyboards to a consultant and seen nothing for years. The Norm/Rothschild video the bears made is an example of relatively smooth video generation with an idea you could never pitch to a studio.
"Let’s just say this is not a hobby for the impatient or the indecisive."
Truth. getting a song to be 100% of what you want is a challenge I've not been able to overcome yet.
As a songwriter and musician with tons of experience in the studio it is so refreshing to hear your perspective on using AI to generate your artistic vision. Making a decent AI track is difficult, it takes hour and hours of work and a significant knowledge of many aspects of music production, including the economics of music production.
For most creators, the sheer cost of producing a song like this in the studio would simply be overwhelming and unrealistically time consuming.
Don’t be fooled by those that poo-hoo AI generate music, while it is true that it can make crappy lyrics and a lack of musical vision sound better than it deserves, this has been true since the first reverb and multitrack recorder were invented.
I am equally as proud of my AI songs as I am my ground up studio productions
Backed it!
I'm looking forward to the golden age of AI when tools like Suno can be run on local hardware, thus bypassing whatever censorship TPTB may try to impose on AI.
The way you described what actually is making music with AI is probably the best I have ever seen, it's also how I think about it. I think the alternative rock aesthetic really fits the vibe!
Pretty next level stuff. The guitars sound awesome 👌
Wow! Will the program spit out the tracks into Protools?
I think they are working on a way to make separate instrumental stems from the start. Maybe Udio has something more advanced regarding that.
There will eventually be a way to do it, and probably within the next 18 months. That's what the pros really want, because they don't want to leave it up to the machine to produce the final mix when they can reliably do better. There is no technological reason it can't be done, but it has to be properly designed and programmed.
Exactly!
The vocals and instrumentals can be stemmed, but the instrumentals cannot be usefully further divided. And yes, someone far more technical with regards to these matters tried it at my request. As he concluded, it's like baking a pie, then trying to separate all the ingredients from it and baking another pie with them.
Thanks
The songs sound good. Very well done.
"It’s more like working with hundreds of musicians and telling one guitarist that you like his riff, a bassist that you like his groove, and then selecting from between the various interpretations of a lyric from the fifty vocalists you’ve got at your disposal. In other words, it still requires making all the structural decisions as a producer, only with totally insane machines providing the input instead of moody musicians."
This sounds like the recording sessions with Steely Dan in the 70s. The SD guys were into sci-fi, they probably would have preferred insane machines to actual moody musicians.
Sounds pretty good, but how do you know that you're not accidentally infringing on another artist's music? The gamma lawyers who lawyer for these major labels seem to love trying to make something out of nothing.
Because the technology doesn't work that way. Lawyers are, by and large, stupid. Because music can be broken down into mathematics, and because it is broken down that way for the AI generation of music, it can be very easily shown how the concept is not even relevant.
The very concept of AI "imitating" human art is false. The training just involves showing it which patterns should be prioritized and which should be deprioritized. Think about the mathematical improbability of randomly getting even 10 notes in a row the same.
Throw in chord progressions and original lyrics, and it's not happening. And "look and feel" is not protectable anyhow. FFS, the guitar riff from "Smells Like Teen Spirit" is just a spin on the one from "More Than a Feeling".
I would love to see an example prompt. Not a serious musician, but i've played around with suno quite a bit for fun, and i cannot get it to cooperate if i have more than 3 suggestions.
Harder than it looks, isn't it! I don't use prompts, for the most part. I write the lyrics, then use one of the preset Personas I've saved to get a moderately consistent sound. The system uses the lyrics as the prompt.
It's like Twitter. Anyone can do it, but not everyone can do it effectively.
hah. truth. I've got a couple good ones but they literally took forever to get right. Your male vocals are so much better and different than anything I've heard so far. I've had difficulty with getting a deeper metal voice like James Hettfield or lemmy. Femal vocals seem a bit easier.
Off topic a bit but I would Love and audiobook version of Sigma Game book...can AI help streamline that as well?
Very interesting. What AI software do you use to produce this?
Primarily Suno. I've tried most of the others, and while they have their advantages, I find that even with its inferior tools, I can get better results out of it.
I don't think people will realize how cutting edge this is. Most of the time when you ask an AI to "bring back the last version" or something similar it will go, I don't know. SORRY. And sometimes minor edits or nuance will make it glitch out. Even AI advocates struggle to explain the black box. I finally tried video generation and it's clunky but I see the potential. Took me several prompts and testing different generators to get characters consistent. AI will race change quite liberally from scene to scene. Modifying vocal prompts must be quite the bear.
Yes, it's somewhat amusing how pretty much everyone dismisses it as "oh, it's just AI" except for all the serious musicians, who react with "Holy cow, how did you do THAT!"
The ability to create and control the vocals has been a dream for decades for a lot of composers because it's so hard to find anyone who can sing well, and who doesn't want to just imitate Whitney Houston or the singer du jour.
As a classical music buff, AI-generated classical music would be cool. Choose an approximate time period and genre, maybe enter your own theme or themes, and get a decent coherent piece complete with convincing sonata-allegro form or whatever you want. Haven't seen anything like that yet, but I expect it's just a matter of time.
Exactly this! I actually see independent artists having a lot of creative potential because you can now do all your own post production, editing out um and uh, bottom thirds, and you can still add in novel b roll or voice recordings. It's very fun as a hobbyist so I am excited to see more talent break into this area. Making something in a few days of focus when ive handed complete storyboards to a consultant and seen nothing for years. The Norm/Rothschild video the bears made is an example of relatively smooth video generation with an idea you could never pitch to a studio.