Suno AI IOS APP.

RajahP

Sorry, but 5 min into this……. amazing.

https://apps.apple.com/us/app/suno-make-and-explore-music/id6480136315

MonkeyDrummer

Only $289 a year…

NeuM

Udio still produces a better quality generation than Suno. Even the lyrics (should you choose to generate them) are generally better. The only advantage Suno has right now is the ability to upload a clip of your own audio then have the system extend or use your clip to guide the music generation.

jwmmakerofmusic

Might be fun to create something in Suno to flip into a beat. 😂 50 credits a day on the basic plan isn't bad at all really.

NeuM

@jwmmakerofmusic said:
Might be fun to create something in Suno to flip into a beat. 😂 50 credits a day on the basic plan isn't bad at all really.

It's actually free to use. You get a limited number of generations per day and other than the lyrics or any audio clips you upload yourself, they own the copyright.

skiphunt

I could be mistaken, but I thought I read that both of these companies are currently in serious legal jeopardy with copyright lawsuits from organizations with very deep pockets. Anyone else read that too?

offbrands

@skiphunt said:
I could be mistaken, but I thought I read that both of these companies are currently in serious legal jeopardy with copyright lawsuits from organizations with very deep pockets. Anyone else read that too?

yeah you’re on it.

MonkeyDrummer

@skiphunt said:
I could be mistaken, but I thought I read that both of these companies are currently in serious legal jeopardy with copyright lawsuits from organizations with very deep pockets. Anyone else read that too?

I would hope so. Watch some YouTube videos like Weaver making songs that sound exactly like the og artist, right down to the voice.

The “training models” of this gen of music AI is more flagrant in ripping off artists than the visual art stuff by a long shot.

NeuM

@skiphunt said:
I could be mistaken, but I thought I read that both of these companies are currently in serious legal jeopardy with copyright lawsuits from organizations with very deep pockets. Anyone else read that too?

These companies just want compensation. All of this will be resolved in court or in negotiation.

Musicians do the same thing these models do and they’ll create a close version of the original training information if they’re cleverly prompted.

BiancaNeve

@NeuM said:
Udio still produces a better quality generation than Suno. Even the lyrics (should you choose to generate them) are generally better. The only advantage Suno has right now is the ability to upload a clip of your own audio then have the system extend or use your clip to guide the music generation.

“And – currently for those of you on our Standard and Pro plans – the ability to upload audio for far greater control over your Udio songs!“

128BPM

But…if you upload your own original (and “unique” 😉) audio/composition:
Does the algorithm steal ( train the models) on the material you upload?

swarmboy

I would caution against any use of Gen AI for two key reasons:

Despite what the T&Cs of these services say, you do not own the copyright in what it outputs, because they literally don't have the rights to assign that copyright to you. Further, the US courts have already decided that the output of generative AI systems cannot be subject to copyright because a machine learning system has no "intent to create".

Tied up with this is the second reason: As has now been conclusively demonstrated, these models have been trained on scraped data (which means it doesn't qualify under "fair use") and are very obviously capable of infringing copyright. So, they will either be a) sued into oblivion by the major labels and publishers or b) become owned subsidiaries of them, so their tech can be used.

The idea that Gen AI is the same as being influenced by other artists is also a non-starter. While a human being is influenced by other art, it is only a small selection of it, and it is filtered through an individual's unique experience. Gen AI is a statical model trained on everything with no context.

As a broader comment, it's my view that this transformer-based/LLM Gen AI is a bubble and the wheels are already coming off. They have fundamental issues which cannot be resolved by scale (hallucinations are baked in, there's not enough data in the world to train them, they do not "know" anything), and they have no concievable path to profitability - all of these companies are burning though VC money and make a loss on every single prompt

RajahP

@swarmboy said:
I would caution against any use of Gen AI for two key reasons:

Despite what the T&Cs of these services say, you do not own the copyright in what it outputs, because they literally don't have the rights to assign that copyright to you. Further, the US courts have already decided that the output of generative AI systems cannot be subject to copyright because a machine learning system has no "intent to create".

Tied up with this is the second reason: As has now been conclusively demonstrated, these models have been trained on scraped data (which means it doesn't qualify under "fair use") and are very obviously capable of infringing copyright. So, they will either be a) sued into oblivion by the major labels and publishers or b) become owned subsidiaries of them, so their tech can be used.

The idea that Gen AI is the same as being influenced by other artists is also a non-starter. While a human being is influenced by other art, it is only a small selection of it, and it is filtered through an individual's unique experience. Gen AI is a statical model trained on everything with no context.

As a broader comment, it's my view that this transformer-based/LLM Gen AI is a bubble and the wheels are already coming off. They have fundamental issues which cannot be resolved by scale (hallucinations are baked in, there's not enough data in the world to train them, they do not "know" anything), and they have no concievable path to profitability - all of these companies are burning though VC money and make a loss on every single prompt

But you mentioned ‘The US Courts’… what about other courts and how do you prevent AI on an international level?
They tell me if you are a felon, you cannot vote.. but, I digress.

Anyway, how do you get the Key and BPM of these songs? Koala found a friend..

I just want to create music and this seems like a powerful tool.. I am sure they will work out the legal stuff.. can’t put the genie back into the bottle..

Gavinski

@swarmboy said:
I would caution against any use of Gen AI for two key reasons:

Despite what the T&Cs of these services say, you do not own the copyright in what it outputs, because they literally don't have the rights to assign that copyright to you. Further, the US courts have already decided that the output of generative AI systems cannot be subject to copyright because a machine learning system has no "intent to create".

Tied up with this is the second reason: As has now been conclusively demonstrated, these models have been trained on scraped data (which means it doesn't qualify under "fair use") and are very obviously capable of infringing copyright. So, they will either be a) sued into oblivion by the major labels and publishers or b) become owned subsidiaries of them, so their tech can be used.

The idea that Gen AI is the same as being influenced by other artists is also a non-starter. While a human being is influenced by other art, it is only a small selection of it, and it is filtered through an individual's unique experience. Gen AI is a statical model trained on everything with no context.

As a broader comment, it's my view that this transformer-based/LLM Gen AI is a bubble and the wheels are already coming off. They have fundamental issues which cannot be resolved by scale (hallucinations are baked in, there's not enough data in the world to train them, they do not "know" anything), and they have no concievable path to profitability - all of these companies are burning though VC money and make a loss on every single prompt

I agree wholeheartedly with this. Regarding the last point, yes, it is entirely possible that they will never manage to create a system that doesn't hallucinate. If they don't, AI will never replace us to the extent people are currently imagining it will. Still, the tech has a lot of potential to be disruptive. Even with the current flaws, AI can and is being used for the mass production of low quality, inaccurate or even intentionally misleading content / propaganda. This alone has the potential to wreak massive havoc.

RonnieOmelettes

@128BPM said:
But…if you upload your own original (and “unique” 😉) audio/composition:
Does the algorithm steal ( train the models) on the material you upload?

Definitely. But of course it’s already consumed anything you’ve uploaded to SoundCloud etc.

AudioGus

@swarmboy said:
I would caution against any use of Gen AI for two key reasons:

Despite what the T&Cs of these services say, you do not own the copyright in what it outputs, because they literally don't have the rights to assign that copyright to you. Further, the US courts have already decided that the output of generative AI systems cannot be subject to copyright because a machine learning system has no "intent to create".

Tied up with this is the second reason: As has now been conclusively demonstrated, these models have been trained on scraped data (which means it doesn't qualify under "fair use") and are very obviously capable of infringing copyright. So, they will either be a) sued into oblivion by the major labels and publishers or b) become owned subsidiaries of them, so their tech can be used.

I dont think it has really been demonstrated yet that the use of data scraping negates AI/ML training from being 'fair use' however they can certainly be fined for scraping. There have recently been settlements (relatively minor fines really) with scraping involved where parties were paid out for that but the copyright sides of the argument have been falling relatively flat it as seems copyright handles the outputs of a process and not the inputs. Will be interesting to see how the music arguments hold up, particularly given that the user needs to ask for a clone type track to be produced.

AudioGus

@RonnieOmelettes said:

@128BPM said:
But…if you upload your own original (and “unique” 😉) audio/composition:
Does the algorithm steal ( train the models) on the material you upload?

Definitely. But of course it’s already consumed anything you’ve uploaded to SoundCloud etc.

The music scraping is not as broad as it is for images and text but it is certainly possible that once out there it could be used.

AudioGus

@Gavinski said:
Still, the tech has a lot of potential to be disruptive. Even with the current flaws, AI can and is being used for the mass production of low quality, inaccurate or even intentionally misleading content / propaganda. This alone has the potential to wreak massive havoc.

As a tool being used by humans it is definitely currently disruptive in some areas of creative work. Once you throw a human cyborg into the mix who inherently does not make low quality crap they are essentially on steroids. But yah, as an autonomous let loose random drone the results are pretty clunky and often hilarious right now.

NeuM

@Gavinski said:

@swarmboy said:
I would caution against any use of Gen AI for two key reasons:

Despite what the T&Cs of these services say, you do not own the copyright in what it outputs, because they literally don't have the rights to assign that copyright to you. Further, the US courts have already decided that the output of generative AI systems cannot be subject to copyright because a machine learning system has no "intent to create".

Tied up with this is the second reason: As has now been conclusively demonstrated, these models have been trained on scraped data (which means it doesn't qualify under "fair use") and are very obviously capable of infringing copyright. So, they will either be a) sued into oblivion by the major labels and publishers or b) become owned subsidiaries of them, so their tech can be used.

The idea that Gen AI is the same as being influenced by other artists is also a non-starter. While a human being is influenced by other art, it is only a small selection of it, and it is filtered through an individual's unique experience. Gen AI is a statical model trained on everything with no context.

As a broader comment, it's my view that this transformer-based/LLM Gen AI is a bubble and the wheels are already coming off. They have fundamental issues which cannot be resolved by scale (hallucinations are baked in, there's not enough data in the world to train them, they do not "know" anything), and they have no concievable path to profitability - all of these companies are burning though VC money and make a loss on every single prompt

I agree wholeheartedly with this. Regarding the last point, yes, it is entirely possible that they will never manage to create a system that doesn't hallucinate. If they don't, AI will never replace us to the extent people are currently imagining it will. Still, the tech has a lot of potential to be disruptive. Even with the current flaws, AI can and is being used for the mass production of low quality, inaccurate or even intentionally misleading content / propaganda. This alone has the potential to wreak massive havoc.

Well, we certainly don't want the job of creating 'high-quality propaganda' taken away from people...

AudioGus

@NeuM said:

@Gavinski said:

@swarmboy said:
I would caution against any use of Gen AI for two key reasons:

Despite what the T&Cs of these services say, you do not own the copyright in what it outputs, because they literally don't have the rights to assign that copyright to you. Further, the US courts have already decided that the output of generative AI systems cannot be subject to copyright because a machine learning system has no "intent to create".

Tied up with this is the second reason: As has now been conclusively demonstrated, these models have been trained on scraped data (which means it doesn't qualify under "fair use") and are very obviously capable of infringing copyright. So, they will either be a) sued into oblivion by the major labels and publishers or b) become owned subsidiaries of them, so their tech can be used.

The idea that Gen AI is the same as being influenced by other artists is also a non-starter. While a human being is influenced by other art, it is only a small selection of it, and it is filtered through an individual's unique experience. Gen AI is a statical model trained on everything with no context.

As a broader comment, it's my view that this transformer-based/LLM Gen AI is a bubble and the wheels are already coming off. They have fundamental issues which cannot be resolved by scale (hallucinations are baked in, there's not enough data in the world to train them, they do not "know" anything), and they have no concievable path to profitability - all of these companies are burning though VC money and make a loss on every single prompt

I agree wholeheartedly with this. Regarding the last point, yes, it is entirely possible that they will never manage to create a system that doesn't hallucinate. If they don't, AI will never replace us to the extent people are currently imagining it will. Still, the tech has a lot of potential to be disruptive. Even with the current flaws, AI can and is being used for the mass production of low quality, inaccurate or even intentionally misleading content / propaganda. This alone has the potential to wreak massive havoc.

Well, we certainly don't want the job of creating 'high-quality propaganda' taken away from people...

haha, yah the upside of the low quality flood is that it educates people to be skeptical and aware of content in general.

suboptimal

AI-generated stuff doesn't get copyright protection at the output stage, but the analysis changes once a human manipulates the output.

An old idea in the common law tradition is that someone who adds labor to a raw material gains some degree of property rights in that material. The seminal case involved manure shoveled into piles in Victorian England.

So by all means, shovel that AI shit.

The curation process is what makes AI-generated stuff useful at all. I'm sure it's a rich source for sampling material, unless it's encumbered by a breezy click-through. Intellectual property is going to slowly strangle everything.

NeuM

@suboptimal said:
AI-generated stuff doesn't get copyright protection at the output stage, but the analysis changes once a human manipulates the output.

An old idea in the common law tradition is that someone who adds labor to a raw material gains some degree of property rights in that material. The seminal case involved manure shoveled into piles in Victorian England.

So by all means, shovel that AI shit.

The curation process is what makes AI-generated stuff useful at all. I'm sure it's a rich source for sampling material, unless it's encumbered by a breezy click-through. Intellectual property is going to slowly strangle everything.

Property rights are a good thing… and that includes intellectual property.

AudioGus

@suboptimal said:
AI-generated stuff doesn't get copyright protection at the output stage, but the analysis changes once a human manipulates the output.

An old idea in the common law tradition is that someone who adds labor to a raw material gains some degree of property rights in that material. The seminal case involved manure shoveled into piles in Victorian England.

So by all means, shovel that AI shit.

The curation process is what makes AI-generated stuff useful at all. I'm sure it's a rich source for sampling material, unless it's encumbered by a breezy click-through. Intellectual property is going to slowly strangle everything.

Beyond simply sampling the output, the tools for guiding it to a specific outcome that you envision, particularly with images, are extremely powerful now.

audiomike

@NeuM said:
Udio still produces a better quality generation than Suno. Even the lyrics (should you choose to generate them) are generally better. The only advantage Suno has right now is the ability to upload a clip of your own audio then have the system extend or use your clip to guide the music generation.

“The only advantage” is a game changer. I uploaded acoustic/vocal and the rest of the song sounded exactly like me vocally and used the same strum pattern I used and same sounding acoustic.

audiomike

It is a songwriting tool and it is awesome. It is not meant to “replace” us. Only an idiot would upload a song and pretend it is their own creation. The fact that these idiots do not own the song who cares?