Last month, when “Heart on My Sleeve,” a track credited to A.I. versions of Drake and the Weeknd, became an unauthorized hit online, many in the music industry loudly fretted about the legal and creative risks to come. But Grimes, the producer and pop singer who has long been enthralled with visions of the future, saw opportunity.
For years, she had been dabbling with fledgling technology in the realm of generative A.I., using the imperfect tools available to create a lullaby; a set of meditations; a Grimes chatbot à la ChatGPT; and plenty of sci-fi and anime-inspired visual art with services like Midjourney and Stable Diffusion.
But the rapid mainstreaming of passable voice-emulating filters — tools that allow users to tweak existing vocals to sound like someone else, notably famous artists like Drake, Michael Jackson or Taylor Swift — struck Grimes as more than just a novelty. They could be a teachable moment, a source of inspiration and even a side business.
“I’ll split 50% royalties on any successful AI generated song that uses my voice,” Grimes tweeted to her more than one million followers, referring to the royalties for the recording itself; she clarified in an interview that the songwriter would be entitled to all profits from the composition, or publishing. “Feel free to use my voice without penalty. I have no label and no legal bindings.”
Then, she and her team rolled out Elf.tech, easy-to-use software that aids producers and songwriters — amateur and professional alike — in making it sound like Grimes is singing their song. So far, there have been more than 15,000 vocal transformations using the tool, called GrimesAI-1, and more than 300 complete songs submitted for distribution to official streaming services with the help of Grimes’s behind-the-scenes apparatus.
Daouda Leonard, her manager and one of the developers of Elf.tech, called it a moment “when preparation meets opportunity.” He added, “People are struggling with this and it’s obviously controversial. How do we showcase what’s possible?”
Over a recent Zoom call, Grimes — who has two children with the entrepreneur Elon Musk — discussed the project so far, musing that her out-of-body celebrity status and longtime obsession with A.I. have combined to make her the perfect vessel for experimentation. She also provided her thoughts on five tracks created with the GrimesAI software. These are edited excerpts from the conversation.
“Heart on My Sleeve” seemed like a tipping point. What was that moment like for you, as someone who has been playing around in this space for years?
I was excited pretty much in every way, even with people talking about the risks. I love A.I., but I am kind of worried that this isn’t more of a discussion, so I think it was really useful. And I was excited that we could definitely get access to this technology now, because we tried to make the Grimes voice five years ago and it kept just being not quite there.
Where does your journey with A.I. start?
Honestly it began when I was a kid, which is weird maybe. We were going through my old college sketch pads last year and we found a bunch of A.I. theory. I’ve always been talking about this; it just wasn’t possible before. But I started getting into the possibilities of art a bit before the crypto times. That’s when we were trying to open-source Grimes for the first time — 2018 or 2019.
What does it mean to “open-source” Grimes?
I’m really interested in the art of identity. We tried to sell my soul — 10 percent of it — in a legally binding agreement. But no one cared, and also it’s at a ridiculously high price that no one will ever buy — like $10 billion. But if they do buy it, then I accept my fate and it’d be worth it.
Only one person you know can afford that.
Yeah, I don’t think he’s going to pay for that. But my soul is already gone. I’ve already completely lost control of the Grimes narrative. Like, I’m accused of war crimes all the time.
So how do you go from that to open-sourcing Grimes musically?
I feel probably less pain than the average person would about such things, because the amount of ego death that I’ve had to go through in order to even just continue being functional is pretty high. The sort of weird, icky feeling a lot of people get when they hear their voice being used in a way that they did not intend — I’m just subject to more crazy press than the average person. I’m so used to it.
Grimes started because I was in a very punk scene and it seemed edgy to put on a pink dress and dance around and make pop music. Part of what I was interested in doing at the time was upsetting people. Even now, what are the boundaries? What is the Overton window of art? What is allowed?
How would you explain to, say, your grandmother, what you’re doing with A.I. now?
People keep getting really upset, being like, “I want to hear something that a human made!” And I’m like, humans made all of this. You still have to write the song, produce the song and sing the vocal. The part that is A.I. is taking the harmonics and the timbre of the vocal and moving them to be consistent with my voice, as opposed to the person’s original voice. It’s like a new microphone.
How was the tool trained?
It was trained on stems of my voice. Some dry vocals [without effects], but it would definitely be better if I would send it some vocals with less reverb. I wish we had more. We don’t have anything old, which is tragic. But I didn’t understand that was useful to keep back then.
How are you confronting the idea that somebody could make a hateful or obscene song in Grimes’s voice?
The good thing about the music industry being so against this is that it seems pretty easy to strike things down. But I also think it’s good for there to be one edgelord moment. With regards to making A.I. more safe and more culturally productive and helpful, it’s good to get things out of the system when they’re least damaging and least popular. I sort of don’t mind if Grimes is the mechanism.
Where would you personally draw the line?
I think slurs, hate speech, advocating violence that’s clearly not in jest. Conveniently, no one’s really done anything bad, and I sort of feel like it’s not even that exciting to.
Do you think that Drake A.I. or Grimes A.I. negates the need for real Drake or real Grimes?
No, I don’t think so. Maybe for me, but I kind of want that. Feeling really amazing from making beautiful art is something that has typically been behind a gate for a lot of people — extreme amounts of time and energy, years of technical training. I think it’s valuable that there’s a tool with which, if you have a beautiful idea, you can make a beautiful thing and access that.
Kito featuring GrimesAI, ‘Cold Touch’
The chorus hook is really good. I could probably be convinced that I worked on it — that would not shock me at all. Especially when the techno comes in. It was immediately very euphoric and very Grimes-y in a really pop way. I would change a lot about the verses, and I’m probably going to do my own version. But I like that she feels really strongly about her artistic vision and wants to stick to it. That’s why we’re doing Kito’s Version and Grimes’s Version, using the Taylor Swiftian nomenclature.
Ravi Parikh featuring GrimesAI, ‘Friend V. Enemy’
The drop is so sick. It’s just fun someone put my name on it, because it’s a really good song. I think the vocal doesn’t sound like me at all, but I can tell why: The singer is exceptionally different from how I sing. She’s enunciating really well, she’s got a lot of vibrato and she might even have an accent or something. You can tell Kito tried to sing like me in the hook, in a really breathy way. This person did not, and it gives credence to the human behind the thing.
OtterlyMusic & Säfira featuring GrimesAI, ‘Concept of Creation’
I love this one, it’s probably my favorite. I think it’s the best depiction of something I would very much make, even down to the production and this image. It really feels like Grimes is self-replicating. It’s so organic and Celtic around the hook, but so A.I. in the verse. She’s even doing my lisp.
Nick Webb featuring GrimesAI, ‘Ether’
I love how weird this song is — it sounds really inhuman. You can hear the A.I. My favorite music is Vangelis because it sounds so early synth. You can hear the technology very profoundly. What I like about the early A.I. stuff is that you can hear the technology very profoundly. I think people will appreciate that more in five years when they realize people only made stuff like this for a couple months.
Kotomi featuring GrimesAI, ‘In Another Life’
These lyrics drive me crazy. It’s really good besides that. “Scream out your name” — anything that feels bordering on sexual makes me really uncomfortable. That’s where I can sort of relate to other artists’ itchiness — lyrics causing me mild knives in the back of my brain. But I also appreciate the feeling of being uncomfortable.