The story behind Kito and AI Grimes’ new single Cold Touch

Have you heard? Grimes has a new song out. Cold Touch, a collaboration with the Los Angeles-based Australian producer Kito, is an icy, propulsive dance track, a two-minute blast of energy that pairs big-tent EDM and hazed out d’n’b with Grimes’ signature ethereal vocals. But Kito’s collaborator is credited as GrimesAI, which is a different entity to the human born Claire Boucher.

If you heard Cold Touch in a club or on the radio, you’d recognise the voice as Grimes’ instantly. But actually, the vocals were provided by Scottish vocalist Nina Nesbitt, then fed through Grimes’ new AI platform Elf.Tech, which allows anyone to input their own vocals and download a new version of the track, redone as Grimes.

Grimes unveiled Elf.Tech a few weeks ago and, since then, innumerable musicians have been posting their own songs featuring Grimes’ vocals on social media. The technology, which is similar to Holly Herndon’s Holly+ vocal deepfake, was trained using raw audio of Grimes’ voice. On the day Elf.Tech launched, Grimes proclaimed: ​“Grimes is now open source and self-replicating.” She’s said she will split royalties 50/​50 with any successful AI generated song that uses her voice.

Kito – who has previously worked with Skrillex, Empress Of and Jorja Smith – shared a snippet of Cold Touch across her socials last week, catching the attention of Grimes, who declared it a ​“masterpiece”. Now, the song had become the first official release with GrimesAI via Diplo’s Mad Decent label.

Recently, there’s been a lot of concern about AI’s potential for damaging the integrity of artists and songwriters (some of which is fair, given cock ups like FN Meka, the major-label-funded AI rapper that was accused of digital blackface). But Cold Touch is arguably a positive example of how AI tech can be used innovatively. ​“I actually think it’s going to be a tool for us,” says Kito. ​“I don’t think it’s going to be a hindrance for creativity at all.”

On the day of Cold Touch​’s release, THE FACE spoke with Kito to discuss the surreal process of collaborating with Grimes and her AI twin.

When you first heard about Elf.Tech, what was your reaction?

I think it felt very predictable that Grimes would embrace the experimentation of using AI for creative expression. I’m a fan of Grimes, so the idea of trying to make something that would sound like a Grimes/​Kito collaboration was really exciting to me.

What was it like working with the software?

I was in London a couple of weeks ago and I did a session with two people that are good friends of mine and amazing writers – Nina Nesbitt and Fred Ball. I pitched the idea to them to try and write a Kito/​Grimes song using Elf.Tech. So we wrote it together and it’s Nina’s voice that was then transformed into Grimes.

It’s kind of funny, because obviously AI is new for everyone and everyone’s a little confused about what involvement AI has in a song. There’s talk about lyrics being written, but obviously for something like this [Grimes has] trained her AI platform to know her voice. I think it did a pretty good job and it helped that Nina sang the way she thought Grimes would sing it. I think it’s probably harder if you have a really deep voice or something – it might not catch it.

What was the concept behind the song?

It’s from the perspective of an AI trying to understand love. There are so many films about AI having consciousness, or about whether AI has consciousness, so we kind of wrote it from that perspective.

When you posted the song on social media, what was the response like?

I think Grimes saw it pretty early on, because she’s obviously keeping an eye out for these songs, and she commented which was really cool. I started talking to her and that’s been amazing – in a weird way, doing this AI version has led to us talking about collaborating [in real life], which I think is pretty crazy.

What did she say about the track?

She loves the chorus and the mental energy of it. It’s kind of strange to be chatting – like, this song is a collaboration because it wouldn’t exist without being inspired by Grimes’ music, but it exists without her participation besides the platform and her voice.

Were there any false starts or learning curves working with the software?

I don’t know if it’s fixed now, but when I was uploading a vocal stem, it would only give me about 30 seconds, so it was a process of uploading and downloading each little vocal bit. And then, obviously, because it’s not giving you a processed vocal, you have to still process it if you’re trying to get a vocal to sound like Grimes. She EQs in a certain way, she uses a lot of reverb. I think the way that I’ve processed it is probably a little different to how she would process it. The software is definitely not this magic button that you click and all of a sudden it’s exactly like Grimes. I think a lot of people don’t understand the limitations of it.

Did Grimes have any tips on how to make it sound more like her?

She did give me some pointers [after she heard Cold Touch] on how she processes her voice. But we talked about lots of different things we could do – we were being very indecisive about, like, should we work on this further and then fully collaborate on it? Or is it a thing of the moment that we put out and this is Kito’s version, and then we work together on a Grimes version – which, I would hope that we do do that. It’s a funny conversation to have because it’s new for both of us.

A lot of people are a bit iffy about AI in music. What was your first impression of AI’s impact on the industry and how do you feel about it now?

I think it’s a bit scary because, for everyone witnessing it – how fast everything moves and the amount of versions of songs popping up without artists’ consent – it feels like it’s threatening our livelihood. I do think it’ll be a mess for a lot of artists! So much is attached to a voice – a story and an identity. Taking a very recognisable voice that we’ve grown to love and recognise, and having someone else take control of that is scary. Grimes was embracing [AI] and wanting to be part of this experiment. I would never put something up copying another artist’s voice unless they were embracing AI, because I do think there’s a bit of a danger to doing that. I would tread carefully in that respect, for sure.

[But] I actually think [AI] is going to be a tool for us. I don’t think it’s going to be a hindrance for creativity at all. I think it’s probably changed the way I’m gonna work on music this year, just because it was so creatively freeing for me to do something off the cuff and release it so quickly after making it. That in itself feels quite inspiring for me.