[AI art] I Made My First Song With GrimesAI-1 Voice-print Today
What it's going to take to slay.
I made my first Grimes-AI track today, and it was super fun. She's pushing some juicy edges by inviting the world in to "replicate" her. Notes from my first experiments:
What I love
• I don’t have to figure out how to publish to DSP’s like Spotify and Apple Music.
Perhaps it’s a quick and simple process, but no one I’ve ever talked to who managed to get their tracks up told me “oh yah, it’s no big deal you just upload your tracks and fill out a simple form, and poof that’s all.”
• Grimes splits the royalties 50/50.
This tells me she’s incentivized to notice which tracks she likes and promote them with her now-considerable social media machine. Presumably, she’s looking for breakout hits to emerge and actively wants to use her platform to make that more likely to happen. It benefits both of us. Good things naturally happen when two parties’ incentives are properly aligned.
This is HUGE for a creator like me who knows you can’t get to the top of the feed on social media platforms without a boost from big accounts.
• 16 GIGABYTES OF STEMS!
For those of you not up on the music lingo, a stem is the finished track of a single instrument or vocal line. Having 16 gigs of source material to remix and remake with…that’s an amount of happiness that’s so big I’ll be happy about this for days or weeks at least.
I plan to chop it all up into samples for my AI drumkit, Sensory Percussion. Stay tuned.
• She’s moving into curation.
AI is about curation. The core hard problem of AI is curation – both of inputs and outputs. I’ll substantiate this argument later, but just put this assertion into your stew pot on the back of the stove and let it cook for a minute.
Grimes has placed herself in one of the most boss-ass power positions an artist can be in right this moment and for the foreseeable next decade or three.
She did an NYTimes feature (link is paywalled, FYI) where she reviews her favorites of what songs have been uploaded so far. There will be more of this. She’s so very buzz-worthy and just gonna get better if her team plays their incredible hand of aces well.
Do you think Elon married a dummy trophy model wife? No, she’s brilliant as hell. Do not underestimate her just because she’s gorgeous and young and female.
What I don’t enjoy
I’m saying all of the following with love. I would tell her all these things to her face precisely because I trust she’s an adult who is capable of hearing constructive feedback.
I sincerely hope she or her team will read this. These things have to get fixed for us to have Grimes voice replicants who can do their job of reinventing the music industry. You can bet lots of other A-list artists are watching her to see how this turns out. I do not wish this to be a cautionary tale.
• Grimes isn't investing big on the tech, at least yet.
Speaking here as a 20+year veteran designer and product person who walked the vaunted halls of Sequoia Capital and Mozilla, the makers of Firefox… my standards here are world-class. She ain’t there yet, but she absolutely can get there with good guidance. I hope she will.
The website is pretty janky, I’m sad to say. For example, I had to refresh the page to get the embedded audio player to stop, and my 6:29 track was cut to a 1:30 track. *sad trombone* It’s a deal killer if that error happens more than every once in a while.
I hope she and her team get a big enough bump of whatever they hoped would happen from this experiment to invest cash to bring in a product visionary and UX/UI designer to do it right.
The bones of the vision of where she wants to take music are already EXCELLENT and the tech concept is solid as hell, so yes please let’s do eeeeeeeeeeeet fo’ real.
Brutal truth: to go big you require both great ideas and great execution. She only has the first at this moment. The buzz will curdle into a cautionary tale if the music made with Grimes-AI doesn’t slay.
• Does the AI voiceprint work? Well... so far not great.
The voice-print version of my track that I received back fell deep into the icky uncanny valley, at least for spoken word. The track that I received back was totally unusable, which didn't surprise me because I knew I was throwing them a curve ball right off the bat. I’ve attached it at the end so you can check it to out for yourself.
I’ll try again with an acapella instead of spoken word since that’s what it’s designed for. You get the best results by pushing what the tool does best instead of trying to get them to do what they aren’t good at.
• The Grimes-AI voiceprint uses a heavy auto-tune which I'm not wild about.
I imagine I'm not the only one who would wish for a dry vocal so that I have more stylistic possibilities in mixing. Feels to me like they're using auto-tune as a band-aid to mask the lack of emotive expression in the AI model. It’s gimmicky and thin.
ElevenLabs has a much more convincing result and has reached out to her to offer to make a spoken word and singing model for her. With a fan-girl’s heart, I hope she will take them up on it!
Without further ado: here’s the first voice-printed track. It’s terrible.
Before you go and blame me for inputting a crap recording as the source, here’s the raw audio of the original so you can compare for yourself. (FYI, the following includes backing instrumentals so you can hear the bigger intention of the piece I’m making, but that wasn’t present in what I submitted to Elf.tech.)
I would be so sad if she decides the experiment failed based on a technical implementation that didn’t make the mark.
Grimes, if you’re reading this – KEEP GOING!
We want/need your vision of remixability. We need the lie of the individual genius to finally be put to rest so we can make whole new genres and business models. Music can get a whole lot more interesting if we do this well.
I am gathering a 100-person AI art collective that works across media – generative image, video, sound, and text – to create art about intimacy and power.
If you’re not already a subscriber, pop your email in to hear more about that in the coming missives.
We’re announcing soon what we have up our sleeve for end of 2023 and into 2024.