AI Covers breaking the internet

General Forums Off-topic Chat
1

Early this year I saw Mariah and Ariana fans were the first to do these AI covers, google collabs, created by vocaloid fans were shut down. Then new AI covers have been appearing made with other competing AIs. Kayne West covers made the trend gain more attention.

Here are some of my favorites I’ve heard.

This one was one of the firsts I heard with the first AI model that was later shut down.

This is very original, as it adds some adlibs using their real voices from different songs.

Some fans that have the skill have even made new songs, Blue Men is an unheard song and this is an imagination exercise of how this song could be

Another usage has been making live versions of songs

There’s also another trend of using the same singer older/younger self to sing their own songs, while we could have the Miley sings Hannah in real life, the most impressive to me are what I call “retrocovers” like the young MJ singing his own songs

There’s also singers able to sing in a language they’ve never have

And finally ficticious characters.

I’ve also seen some memes made with it, Spanish people is what they love

https://twitter.com/FemJenni/status/1671555975302504449

For now record labels have started to claim some of these, but the legislation is on a grey area as of now. Some artists like David Guetta have used it, Paul Mccartney has used AI, but for extracting an acapella, most news out there fail to mention it however.

Some singers have already reacted to it, some negatively, some neutrally, others like Grimes even said she would give 50% of the royalties and would give people audios if they needed more training.

https://twitter.com/KeshaLegion/status/1663735746048708611

https://twitter.com/euphoric_bangta/status/1652602629833326597

1 Like
2

Sonic singing Baby One More Time is hilarious!

3

Oh woaw!
Any tutorials on how to do so, also for instrumentals? Is there some sort of open stableDiffusion for music? :slight_smile:

4

The first one was diff-svc (created and shut down by vocaloid fans), then sovits came and right now people are using RVC which will soon release V3…

Now that I think about it this might be very useful for videogame voices, there’s the possibility to create hybrid voices…

5

A good subject to place into debate:

  • Is the actor and their role themselves? When for example you think of Arnold (free person) you immediately think of Terminator (copywrited role).

  • Is the face and voice of a singer their own? If you think of Michael Jackson, you think about the voice of the singer, or the record from the label company that is copywrited?

This is some how the problem where everything begins. If you think for example that an artist is a normal person with musical abilities. Then the other side of the coin is that the artist in terms of greatness and status is the construct of the record company.

Obviously you have both of the sides at the same time, but in the end since the legal system is very strong and has figured out all of the loopholes like these ones, you can’t do anything other than to follow the rules.

6

Just a quick warning. StableDiffusion is a trained neural network and you just run it to generate images. With those sorts of applications, you take a pretrained neural network, you have to collect data and train that neural network.
It takes a lot longer. Setting up the environment takes a lot longer. You have to collect data and get an understanding of what works and what doesn’t. Training it takes a lot longer and may require quite some experimenting.
But, if it works, it is awesome :slight_smile:

7

I couldn’t care less about the legal stuff for that kind of application. No copyright law was made with those sorts of technologies in mind. I’m confident it was heavily shaped by lobbyists in pretty much every country. I wouldn’t publish anything like that, because I don’t want to deal with unknown legal situations.

Personally, I am fine with it, as long as it is clearly shown to be AI generated. If people would start to commercialize it and start to make reasonable profits with it, that would certainly need to be taken into account.

1 Like
8

This needs to be criminalized. Its the death of culture and meaning.

For those Dune Fans…for lovers of Arrakis…please read up on the Butlerian Jihad.

1 Like
9

I am quite sure it will be criminalized. But not because it is the death of culture and meaning, but because the music industry has so much money and influence. I am more inclined to argue that the music industry itself already killed a lot of culture and meaning.

Edit: To be honest, I believe this sort of technology provides a ridiculous amount of artistic freedom and allows artists to experiment in never before seen ways. If an musician believes that in a certain song, if something would be replaced, the effect would have been better, they can actually try it out and experiment. It doesn’t have to be their own song, but can be someone else’s. They don’t have to publish it, but they can use it to learn and get a more in depth understanding. That’s amazing in my opinion.

1 Like
10

If you try to see what is the the problem with the datasets:

  • If the work is based on some IP (eg: Transformers, Batman)
  • If the work is based on some image produced by someone else (eg: a photographer or painter)

Pretty much you are safe, if you train your own AI from scratch, using only your own personal resources (your own renders, your own photos, and your own sketches).

11

Don’t want to be disrespectful, but I have no intention to discuss the legal situation about it. It is a piece of cake to construct situations that make no sense at all to me regarding the legalities of machine learning generated content.
I also don’t see how this could change in the foreseeable future, considering how those rulings and laws are being changed in general.

12

I think we should avoid silly phrases like “breaking the internet”, unless there is in fact clear evidence that the internet is somehow struggling to cope with some videos. :stuck_out_tongue: