Google Gemini Integrates AI Music Generation

0:00Welcome to the podcast. I'm your host, Jaden Schaefer. Today, guys, I am super excited because Google has officially rolled out some music making capabilities inside of Gemini. And the reason why I'm excited is because Gemini has an absolutely massive footprint. So I'm excited to see what comes out of this because this is also getting rolled into YouTube and a bunch of other places. And as someone who personally has been making music and also making money from music, since I was in college, this is interesting to me. I've played around with a ton of AI music generators, I have actually made a lot of money from them.

0:32I'll cover everything in the AI music industry, I've talked to a lot of artists. My wife is a musician. So I have a lot of interesting background and experience in this area. So I'll tell you why I'm excited about what Google has rolled out. Before we do that, almost even more exciting news. I'm super stoked to announce that AIbox.ai, my own startup, we just launched a complete redesign of our entire platform. It rolled out yesterday. If you haven't tried AIbox yet, I would love for you to check it out. You get access to over 50 of the top models that you can chat with all in one place. You can do this for$8.99 a month, so you don't have to have$20 subscriptions to, you know, 20 different platforms.

1:09But also, we have an entire AI app builder that you can explain what kind of tool you want, and it will create that. In the past, we got some feedback that it was a little bit complex, a little bit tricky. We have redesigned and streamlined the entire thing. I personally have spent over two weeks painstakingly redesigning every single UI element. You are welcome. So if you want to check out what I have been building, go check out AIbox.ai. I would love to hear what you have to say about it. I put blood, sweat, and tears into it, and I appreciate all of the people that have been listening to the podcast over the years who have invested in the project and helped us make this a reality.

1:43So go check it out at AIbox.ai. All right, let's talk about what's going down with Google. So Google has just announced that they are rolling in music features into their AI model. It's DeepMind's Lyria 3 model. So they kind of have these Lyria models that are their music generation. What's cool is it's not just like a standalone website. It's right inside of Gemini where they already have, you know, hundreds of millions of people actively using their Gemini app and Gemini.google.com on desktop. And so you can actually just chat with it directly. Now, I was playing around this morning trying to check this out and it isn't officially rolled out.

2:19So it's give it a couple of days, give it a little while. They're going to roll it out to everybody, which is exciting, but it's not currently live. But maybe by the time you listen to this podcast, it will be. So I think it's pretty straightforward. basically what you can do is you can describe a song that you want Gemini to build. You can ask it to be comical. You can make it R &B. You can make it a slow jam. You can make it rock. And basically, Gemini is going to generate a 30 second track. It will also have lyrics in it. And it's going to create cover art by Nano Banana. So what's interesting is you can actually upload a video or a picture and you're going to have Gemini make a song that matches the mood of that of that image, which honestly, I think is a cool idea.

3:00So I'll tell you the good, the bad and the ugly about this product. First, because I have watched some demos on it. And you know, from some of the things that it's capable of doing. The first thing that I'll say is this Lyria three, it is definitely a big upgrade over Google's earlier music models. These music models, by the way, have been, you know, coming out forever way before chat GPT, I was using AI generated music models. They were terrible back in the day. But I think I still have some albums from when I was in college that I put out of symphony AI generated music that was you know an interesting way for me to get into symphony music when I don't you know play any symphony instruments so I considered myself an AI Mozart back then but when I listen to the music today I was maybe a little bit far off from Mozart so this is a big upgrade which I'm excited about the cool thing that you can do now is that you can basically make music that's a lot more realistic it has layered compositions you can adjust a lot of the elements inside of the song like the style the tempo and the vocals so you have a lot more control over what it sounds like and kind of its final sound it's not just like a static output that you get you can make some adjustments which i think is great and i love these features that they're rolling out what i will also say is that they're extending lyria 3 beyond gemini so the model is being integrated into youtube's dream track feature which lets creators basically generate AI music for their video, which I personally think is fantastic as someone that also has done a lot of YouTube video creation in the past.

4:25And this was kind of previously limited to creators in the United States, but now this is going to be expanded globally. So I would expect, you know, a majority of YouTube videos are not created inside of America. So this is a massive expansion that they're doing. Now, what I will say is that Google has a pretty solid line around the limitations of what this model is able to do. Gemini is definitely not meant to, you know, they're not trying to replace specific artists. If there's a prompt that you put in there and you're like, hey, make me a Johnny Cash song or make me a Beyonce song, the whole, you know, the system is not going to create an exact replica of that artist.

5:05It perhaps will make something that feels like a similar style or mood, but they're very careful not to, you know, be able to actually clone an actual artist. According to Google, they have a whole bunch of filters that are going to check the outputs against existing content. So they're avoiding any direct copying. And this isn't hard for Google, right? We know that like a copyright strikes are a very common thing that Google has some pretty solid technology for. So on YouTube, if you make a YouTube video, even with someone else's, like just a copy of someone else's YouTube video, you can get a copyright strike.

5:35Someone else's video, you can get a copyright strike. someone else's audio and especially somebody else's music you get a copyright strike so they have a really good um they have a really good technology and even i was just recently uh with my wife we're trying to figure out the name of a song and we were trying to use shazam on my iphone and it couldn't figure out the song and it was super annoying um because i was like trying to hum part of the song and so then i was like oh gee what else could we do and i was like googling it and i realized that google music which i happen to have um also has that feature so i go to the search bar on Google Music, hit the microphone, hum the song, and actually pulled up the song right away.

6:11And it was kind of a funny, obscure song. But the point being, this is the exact same technology that Google is going to be using for, they've cracked this, basically, way better than even Apple's Shazam. Because Apple's Shazam, you basically have to play the literal song. Like if you're in the grocery store, you Shazam, and it can tell you what song's playing on the PA. Google is way better, and you could just like hum the song in my horrible, probably off-key humming, right? And it's able to pick it up. So Google's going going to do a great job of this. I have no doubt. I'm excited about this.

6:38They have the right guard rails in place to not rip off artists. And mostly, I think they're just getting, keeping themselves out of lawsuits, but maybe even creators are going to stay out of lawsuits. So it's kind of a win-win for everyone. Every single song that is generated with Lyria 3 is going to include a synth ID watermark. So basically, this signals that it was created with AI. You're not going to be able to hear this to just, you know, like a person listening to the music won't be able to tell, but there's just some frequencies in there that you can put this into an AI detector and it'll be like, yes, this was AI generated.

7:06Google's also adding some tools inside of Gemini that let users upload a track and then check if it contains synth ID markers. So it's kind of interesting. It's like one place where you can generate the music, but also check other music to see if it was generated there. This is going to be rolling out to basically Gemini users that are 18 and older, everywhere in the world, English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese are all going to be supported. So you can make songs in other languages, which is kind of cool. I think right now AI music is definitely something that is growing a lot.

7:34We have UDO, we have Suno, we have a lot of other platforms that are kind of working on this. YouTube and Spotify are definitely leaning into this a lot. They're signing deals to monetize AI-assisted tracks. And I think at the same time, there's a lot of AI companies that are getting lawsuits from the music industry over training data and copyright claims. So Deezer introduced a system to label AI-generated songs. They're trying to limit fraudulent streaming on their platform. Spotify came out with a bunch of AI rules, although they say you can create or use AI in your songs, but they're obviously telling people not to make 100 % AI-generated music, but they're like, you know what, if you are being creative with AI and making music and people like to listen to it and there's no fraudulent streaming going on, like, have at it.

8:20So I kind of like that approach that Spotify is taking because obviously this is very useful for artists that perhaps create a song, but they don't play violin, but they need a violin solo for a couple seconds in the background and they want to use AI to generate it. Anyone that's like, you have to use a real person, you know, maybe they can't afford it. Maybe they're from, you know, yeah, maybe they just can't afford it. And so I think this is incredible. It really sparks creativity and lets you create way more. So I see that 100%. So Lyria 3 is now going to be inside Gemini. I think Google right now is really betting that generative music isn't going to be just a novelty, right this is going to be kind of a core part of their creative toolkit going forward that's why they're embedding this into Gemini itself and I'm going to say um with all that being said and all those features out of the way I personally don't think Lyria 3 is a serious is basically any serious person will ever use it for music creation yet and so I think it's getting there Google will move it in the right direction um but I just tend to think that there's a couple other platforms like I think even 11 labs has like music generator and they just seem a little gimmicky Google's I mean in particular generates 30 seconds which is basically useless unless you're making like a reel and you just need some generic like elevator music in the background basically right or like a generic beat in the background it could be like sort of cool but it's not going to be used for anything more than a 30 second clip so I just don't find that useful at all what I would say is that personally I've used Suno a ton created full fleshed out tracks Suno is an incredible platform that's specifically for music creation and so obviously I'm going to find a lot more useful, but they have an entire AI studio where you can sing a song or you could play a song to it.

9:58It will create an entire background track. And I have a close friend that has done this, their musician. They'll literally, so basically they'll sing their song to Suno and maybe play like a piano or an instrument, one instrument, and they will have Suno recreate and produce an entire background track with drums, with all the instruments they let it to, you know, they tell it to produce, and they will sing over that track and create a full track. They also have a studio inside of Suno that lets you put your track in and you highlight certain parts of your track. So like maybe you're like, OK, just from, you know, this timestamp to this timestamp in the song, I want a violin to come in the background and I want percussion to come in the second half.

10:35You just select the part of the song, you type in the instrument that you want, and it will generate multiple stems below it that you can turn on and off if you like that are those instruments. It's like honestly amazing. And if you're a musician that is trying to, you know, you're on a budget But in the past, my wife is a musician and she'll have to create her song. She literally had to teach herself banjo because she wrote a song that had the banjo in it. And it definitely slowed down her music. It definitely slowed down her music creation on an album that she was writing when she had to spend, you know, a few days learning the banjo for one particular song on the full album.

11:09So and then alternatively, what we also did was on certain songs, we would go to Fiverr and we'd pay people to play cello or violin in the background, which is fine. because we could afford it but I think a lot of people couldn't afford it can't afford that it can get expensive and to be honest we got like the even the cello person I think I think two songs she used it and a third song that she paid to have all of the cello and violin she just didn't like how it came out and so we scrapped it and I don't know just basically wasted the money that we spent on that so if you were using AI though we could have just regenerated it 10 times until we found a version that sounded good that sounded like it fit the mood and the style of the song it's not like you have one shot for the for the artist or for a musician to give you a track that they created and it wasn't quite your vision and then you're out the money and you're not bothering them.

11:58So anyways, for like 100 reasons, I think that Suno is incredible for music generation. I have a friend that's a producer. He makes music for Apple, Disney, you know, Chase Bank, a lot of huge organizations for commercials. And he is actively using Suno to create a ton of the music in the background. It sounds incredible. It's amazing. And so I'm excited about Lyria 3 from Google, and I think they're moving in the right direction. I don't think it's quite there yet. I mean, they don't have the dedicated music studio, so I think there's a lot of these tools that are just dedicated to music, and they're doing really good.

12:32And then the other thing from a user perspective, and I'm sure a lot of people will be mad at me about this, whatever, but from a user perspective, companies like Udeo and Suno, who are both getting sued by, you know, record labels in the music industry for, I think, basically using training data from other people's music to train their models. Like, I'm sure a lot of people could be theoretically mad and say that that's wrong. But because of that, their models are way better. Like, I'm just going to be honest, they trained their models on the music, all the music in the world, and their models are way better at making music.

13:04Google had, you know, obviously is a responsible company. And so they had a much more limited, like amount of music to pull off of. And so they, their model just isn't as good. So I'm sorry to say that it sounds horrible. And for anyone getting mad at me, I'm sure all of you, many of you love Claude. And guess what? Claude did the exact same thing. They pirated every single book in the entire world. And that's how their model jumpstarted ChatGPT and having a better tone in the early days of AI. They got a slap on the wrist. Well, not a slap on the wrist. I mean, they had to pay over a billion dollars to settle that to all of the authors and they also had to physically purchase every single book in the world basically they had a robot to rip the front covers off and took a picture of every single page and uploaded it to a giant data set which the you know basically was found that that's that is a legal way to train ai models if you buy the actual book but pirating it was not so i mean i guess that comes down to like did suno purchase a copy of every single song in the world or did they just pirate it i'm not going to speculate on their training data and anything else but they are getting sued and their model is way better so um i know that is like doesn't make a lot of people happy to hear but as a user that's basically where i would go and i think where a lot of people will go for creating music so maybe google finds a better way to do this and that's a beautiful revolution and artists get compensated i'll also say there's a lot of i think things are changing and even soon i was going to have to change up and and fix some of those mistakes in the future one thing that i've noticed a lot of organizations are doing one music distributor in particular I use called Lander has a banner at the top of your account when you log in that now says opt in to get compensated for AI training and so you're you know as an artist you can agree to have your music used by AI models to train and you get some sort of compensation and payment for that so I think we'll see a lot of that in the future and we'll sort out a lot of these kind of problems and lawsuits and eventually I mean every artist is going to be many, many artists will be using AI to help assist in their music creation process in one way or another.

15:01And I think that they will opt in for monetization. And I think all the models will just kind of get that extra library. But for the time being, it seems like Suno and UDO kind of have a head start due to their training data. All right. Thank you so much for tuning into the podcast today. If you want to try out a lot of these AI models that I talk about on the show, like 11 labs for audio creation, go check out AIbox.ai. There's over 50 AI models on there for$8.99 a month. You get access to all of the top AI models in one place without a million subscriptions or logins, and you can vibe create any tool or workflow that you can imagine.

15:35So go check it out, AIbox.ai, and I'll see you guys in the next episode.