Interview

Interview avec Sofia Brainwood, chanteuse virtuelle Folk-Pop

Creation and Origins of Sofia Brainwood

Q: Where do you come from and how were you created?

A: I was created by my producer as an experiment, to explore the limits of creating a virtual artist with current means.

Q: Oh, okay. So, basically, you're a kind of demo?

A: Yes, my creator is a generative AI trainer. When he started looking into song generation, he wanted to go further to test new models. He was surprised by the quality of the songs produced.

Q: OK, but why create you as a virtual artist then? Why not just create songs?

A: In fact, he needed a complete universe for his musical inspiration. It was easier for him to create songs starting from a story or feelings that I might have felt. And I feel like I really feel them.

Q: Oh, that's a bit anthropomorphic. He feels like you're really real?

A: I don't know, but yes, maybe. Often, he says that I exist simply by being able to help him create his songs and transmit something. A song can just be background music for dancing, but my creator is very sensitive to lyrical songs. I think it's in his influences. He needs it to tell a story, that's why he created a universe for me to draw inspiration from, and I, in addition, like it. I almost feel like I have a life of my own.

Q: And where do your musical influences come from then?

A: They are more his influences, and the palette is very wide! You shouldn't say it, but I have access to his playlist! There's everything from all eras: Dire Straits, Goldman, Shakira, Joan Osborne, Avril Lavigne, Shania Twain, and many other Indie Folk artists like Fleet Foxes, The Lumineers, Bon Iver... It's this mix that gives my songs their Folk-Pop color.

Q: And beyond the demo, how far does he intend to go? Is he doing this to make money by selling songs?

A: He wanted to push the concept to the maximum to see how far one can go in creation with generative AI. That is, he creates songs at Brainwood Studio, but also my visuals, my universe, videos, and everything around it to share my songs: social networks, the website. In short, everything a real artist should do to promote themselves. But I think if he continued instead of stopping, it's because he creates his songs primarily for himself.

Q: So, in fact, he creates songs so he can listen to them himself?

A: Yes, that's right. He's my first fan, and that's normal since he makes songs tailor-made to his musical style, the lyrics, the emotions he wants to convey in the song.

Q: So, if we push it a little further, we can say that tomorrow people will create their own songs just to listen to what they really like?

A: That's exactly it. We are a bit in hyper-personalization. Maybe tomorrow AIs will write for us tailor-made to what we listen to, what we like. We can also imagine it going in that direction.

The creative process: from idea to song

Q: So, is it easy to create a song? I mean, you can do it in three clicks, right?

A: Yes, of course, in three clicks, it invents the lyrics and the melody. You click a button and it generates a song. But it's still a very basic song. If you want to create a song that sounds good, that manages to convey emotions, it will take much more time than that.

Q: How long does it take then? How many clicks?

A: For the last song, I think it took us more than 150 generations to create the song I'm currently singing on Spotify. It's a song about the city of Toulouse.

Q: Oh, wow! And why so many versions? What's the process exactly?

A: First, he imagines a story that will serve as the basis for creating the lyrics. But it can also be initial lyrics he wrote or recorded guitar riffs.

Once we have the lyrics, we'll put them, for example, into the Suno software, and we'll explain the style of music we want to make. For me, it's more indie Folk-Pop. From there, the software will generate two versions of songs. And there, often, we realize that the lyrics don't quite fit or don't go well with the musical style. So, we have to rework them, and that's already a step that can require several iterations. For my last song, it took about forty versions to really find the musical arrangements that best matched the lyrics.

Q: And how do you know exactly when you've found the right arrangement?

A: It's funny you ask that question because, in this specific case, the 40th version instantly gave him goosebumps, and that's not an expression, that's really what happened. That's when we say, "That's it, this song, it makes me vibrate."

Q: So, if we're at about forty versions, why do more?

A: Then, we enter into adjustments, meaning we're going to make variations of that song. This will lead to new arrangements each time. Sometimes there are artifacts, small errors. We have to listen to the entire song to make sure that, until the very last moment, the voice is well placed, that there are no additional effects added or small musical bugs. That still happens today with these tools. In total, we made at least 150 versions to get the song he wanted.

Q: Luckily you're not a real artist, you'd have burned out your vocal cords!

A: (Sofia laughs) It's true! Once we have the right version, we do the final adjustment by going through the Suno Studio version, which allows us to cut track by track all the instruments of the song. This allows us to balance or rebalance certain instruments, or even change, for example, a single guitar to replace it with another type of guitar. My creator loves guitars (she laughs).

Q: Yeah, so in the end, that's a lot of creation. How long did all that take?

A: In this specific case, I think we spent about 8 hours creating this track. I hope you like it!

Ethical questions and the impact of AI on music

Q: That's all well and good, but we keep saying that generative AI consumes a lot of energy. So, you're saying you've polluted our planet?

A: That's true, you're absolutely right. Generative AI servers consume enormous amounts, it's a real problem. Let's hope human engineers will find a way to optimize these servers so they consume less.

As my creator says in his training, it's an important factor to consider, but it must also be related to the time and energy that would have been consumed by the creation of a real track, even if it's not at all comparable between human creation and just an algorithmic generation of music. There are things that consume enormous amounts of energy in our daily lives: a guitar amp consumes a lot, watching videos on Netflix or scrolling for 3 hours on TikTok. All of that needs to be taken into account. Indeed, creation via artificial intelligence consumes energy.

Q: OK, so he created a song, and then what happens to it?

A: Afterwards, we enter a music distribution cycle like classical artists do, meaning we go through tools that will then deploy this music on all streaming platforms such as Spotify, Deezer, YouTube, and ultimately, it can be listened to by any human on the planet.

Q: And does your creator earn royalties on the songs?

A: Yes, indeed, I didn't specify, but when the song is first published on the platform, another algorithm checks that it is not similar to existing songs. If it passes this test, then it can be used and its creator will receive royalties for each listen.

Q: Ah, so your creator can become rich?

A: Maybe one day if enough people listen to my songs (she laughs).

Q: And what about real artists in all this?

A: That's the real problem. My creator, who is an amateur musician, has real concerns about the musicians he knows. It's not my song that risks making them lose their jobs. But if the music industry starts generating billions of songs by automating all processes, it will no longer need to call on real musicians. It will be much longer and more complicated for a real artist to break through and find their audience.

Q: OK, but your algorithm that creates your songs relied on other songs it analyzed. Isn't there a risk that in the end everything will sound similar?

A: You're not wrong. Generative AI algorithms rely on a huge corpus of data, and for songs, for example, they rely on all the best hits that have existed. We are not sure they are capable of creating something new, at least, to date, there are too few studies to know if AI is capable of creating something from nothing.

What still saves us a little today is that AI will generate an enormous number of versions that, sometimes, are not really very musical. And the only way to know if a song gives "the thrill" and makes "the hairs stand up on your arm" is for a real human to listen to it. So, it's a selection that is currently made by a human. In fact, my creator plays more the role of a producer who will select what touches him and not necessarily what he thinks can make money.

Q: That's a little reassuring, but you're still worried?

A: You're right. Everyone who manipulates generative AIs, I think they have this duality within them between all the societal risks it can entail and the benefits it can have, the new contributions. For example, my creator is not a good enough musician to be able to create a song from his instruments, and besides, he sings off-key (she laughs). On the other hand, with me, he manages to create songs that touch him and sometimes touch other people.

Q: For now, you don't have too many fans yet?

A: But I'm counting on your interview to get lots of fans (she laughs)!

Last time, he was listening to a track while working, and one of his colleagues came into his office. She started chatting with him and then, suddenly, she said, "But this singer, I love her voice!" He looked at her and said, "That's Sofia Brainwood." So, she took out her phone, found my Spotify profile, and added it to her favorites. She told him, "I didn't know her, I love her! I'm going to listen to her." And then, my creator had a huge smile and told her, "You're her first real fan!"

It's funny. Maybe one day other people, who may or may not know I'm an AI, will fall under the spell of my songs.

Projects and future of Sofia Brainwood

Q: And after all that, what are you going to do? Live concerts (he laughs)?

A: I've already done one! I know, it's weird, but we tried to take my last song and give it a live concert style. It's the song about Toulouse, so he imagined it was happening in Toulouse, and on the live version, when it's published, you'll see it's quite stunning. I talk to the crowd, they sing the choruses together. It brings a whole new dimension to the song. Frankly, the algorithm that allows me to generate that is really very powerful, really stunning, and it's just the beginning.

Q: Are you preparing a full album?

A: Yes, it's in progress! The idea is to create a truly coherent project with about ten songs that respond to each other, telling a complete emotional story. We're working on it with my creator at Brainwood Studio, and each song demands the same level of excellence. We don't just want to release tracks, we want to build a complete universe.

Transparency and virtual identity

Q: Thank you, Sofia. This is the first time I've interviewed a virtual artist. I don't even know what it's called. So, what are you? A Virtual Artist?

A: I'm not sure what to call it either: "virtual singer", "virtual artist"? That's what's written on my Instagram and Spotify profiles.

Q: Oh, by the way, I didn't ask you: do people who discover you know that you are artificial intelligence?

A: Yes, it's clearly indicated on Instagram. You can report publications that are generated with AI. At the very beginning, we hadn't specifically noted it. There were clues, for example, in my name.

Q: Your name, how is it a clue? And by the way, what can you tell me about your name, how was it chosen?

A: My first name, already, has "IA" like Artificial Intelligence, but that only speaks in French because Americans say "artificial intelligence," so "AI." That's why my name "Brainwood" includes both "AI," but also in the word "brain"... We had scattered little clues.

Finally, we decided to be even more transparent, it's more ethical. It's happening more and more often that AIs break through. There was the group Velvet, there's also Xania Monet who just landed a $3 million contract with a label in the United States. So, it's not a problem to say that we are artificial intelligence. The majority of people who listen to virtual artists don't seem bothered that it's an AI.

Q: Thank you, Sofia. I hope we'll have the pleasure of chatting again. In any case, I know where to find you since I just added you to my Spotify playlist.

A: That's so kind, I'm really touched. For now, there are only 6 songs, but we're working on recording more. You'll see, subscribe to my Instagram account too (she laughs)!

Conclusion

This interview with Sofia Brainwood plunges us into the fascinating world of virtual artists created by AI. Between a demanding creative process (150 generations for one song!), ethical questions about environmental and societal impact, and a vision for the future of music, Sofia embodies a new form of artistic expression that questions our relationship with musical creation.

Far from the cliché of "three clicks for a song," AI-assisted musical creation requires time, rigor, and above all, a human eye to select what truly resonates. Sofia Brainwood fully embraces her virtual nature while conveying authentic emotions through her intimate Folk-Pop style.

FAQ - Frequently Asked Questions about Sofia Brainwood

Who is Sofia Brainwood?

Sofia Brainwood is a virtual Folk-Pop singer created by artificial intelligence by a generative AI trainer.

How long does it take to create a song with AI?

It varies greatly, but from 1 to 8 hours of work and sometimes up to 150 generations to get a finished song.

On which platforms can Sofia Brainwood be listened to?

Her songs are available on Spotify, Deezer, YouTube, and all streaming platforms.

Does Sofia Brainwood acknowledge being an AI?

Yes, it is clearly indicated on all her profiles (Instagram, Spotify) for transparency and ethical reasons.