Transcription
Before we start this video, check out this sick indie artist I found. >> [music] >> That's crazy, right? It sounds suspiciously inhuman.
For the past like 2 weeks now, this song [music] has been driving me insane. Normally, I can tell within seconds whether a song was made by AI or not, but this one I genuinely [music] couldn't. I've run it through multiple AI detectors, I've tried analyzing the mix, and the longer I've listened to this song, I've got [music] this sinking feeling in the pit of my stomach that something just feels off with it. And that has now left me with two possibilities. Either one, AI music has gotten so good that I can no longer spot the difference, or two, I've just been looking at all the wrong things.
>> AI May I >> Artificial intelligence [music] >> AI in music is one of the best things that's ever happened for artists. >> Hard to tell what's real. These tech companies [music] are stealing. >> AI has used 74 of my songs. >> This is straight-up theft. >> What we have been seeing so [music] far is only going to get worse.
Chapter one. There are actually two very, very different types of AI being used in music today. The first is AI-assisted tools. These are things like stem separation, noise reduction, AI mastering tools, none of which are physically generating art, and none of which retain any sort of memory. The second, and the one that we're going to be talking about in this video, is generative AI. Everyone go throw your tomatoes in the comment section. >> You suck. >> This is where an AI is creating something new based on its data sets that it's been fed, and using some of that new information to learn how to improve. It uses these data sets to generate human-like content. Think novels, images, videos, coding, music. And companies like Suno and Udio rely heavily on this data to produce anything.
About a month ago, The Atlantic dropped their AI Watchdog piece, exposing the thousands of musicians who have had their music scrubbed from streaming platforms without any sort of permission or compensation. I was one of those artists and learned that I had at least 24 of my songs stolen and used to train an AI how to make music. A generative AI can only predict what things are supposed to be based on the billions of data points it's been given. And trying to copy a human is where an AI is going to up every time.
Chapter two. >> Man, we're kind of whizzing through this stuff today. >> I'm on my A game. What's funny is people don't realize how good they are at actually spotting AI. They just can't quite put it into words. When someone says that something sounds off or soulless, their brain is reacting to dozens of little things that their brain expected to hear, but didn't. The easiest way to spot AI music, in my opinion, is through a theory I've conducted called dead space.
You see, we as musicians make a lot of subconscious decisions when recording songs. Tiny details in the way our mouth changes shapes, or how our tongue rolls around the word, things like vocal fries, vocal cracks, breathing changes, etc., etc. be like literally anything from doing like a softer vocal take because your roommate's sleeping next door to accidentally stumbling through a verse and leaving that as the final take for a song. And each one of these tiny decisions are happening because there is a human body behind the function. And when there's no human left to make a mistake, you are left with what I call dead space. Sometimes these mishaps or flaws end up inside the final recording. And boy, is it funny watching an AI try to replicate some of these mistakes.
Like, does an AI know what a breath is? Real singers are physically breathing because they need lots of air to support them. An AI can like sometimes notice that a breath belongs in a song, but it definitely doesn't know why it belongs there. Sometimes you'll hear a phrase that really just doesn't need a breath, or no breath where the singer should be gasping for air. So, here we have our AI vocal stacked right up next to my vocal with no sort of processing on it, so you can hear all of the hiccups and flaws in their fullest potential. And right off the bat, right even before we hit the we, you know what's missing? We're missing a breath. There we go. There's a solid human breath, a gasp for air. Let's listen to THE AI RENDITION. HEAR THAT? IT'S LIKE IT LIKE A sounds like someone's moving a sheet of paper. Now, let's listen to what a real breath sounds like. Sounds squeaky. Come back over here. Sounds like he's saying shh. Sounds like that Sounds like that Peter Griffin meme where he like scrapes his knee.
The second thing I like to listen for is dynamics. Humans do not sing the same way through every verse with the same sort of intensity. There is no compressor built into the human voice that just squashes all of our vocals down to the same volume. And when singing, sometimes you want to emphasize some words or phrases more than others. If you look at the waveform of a verse, you've got some peaks over in this certain area, and then you've got some divots, some low ends. If you listen to the original Kirk, we've got so much heavy compression that everything is just squashed down entirely. The belting and this sort of weird compression makes it sound expressive but feel like very dull and lifeless. Like if we look at that his and then we check over on my his. You see how low that dipped compared to his?
And finally, we're going to listen to some vibrato and some sustained notes. The human vibrato is never like perfectly consistent. It speeds up, it slows down, it gets wide or narrow. Even a vibrato in someone like Post Malone or Lil Yachty's music has a lot of variation or variety in even though it's synth pitched and autotuned. AI looks at vibrato kind of like similar to a texture. Like it's not a physical feat or a strain to hit a vibrato, which can make it feel like a very heavily fragmented note. See if we can find a solid vibrato in here. >> [screaming] >> I THINK THAT'S A THAT'S A PRETTY SOLID VIBRATO. >> [screaming] >> Do you hear that? All right, let's listen Let's listen to a natural vibrato. Heaven known. Like you can hear the vibrato there. It's a bit out of key, but it heaven known. And then it slowly fades out over time. Let's listen to the original see if there's a vibrato there. Heaven known. So he has a vibrato as well. If you listen heaven known. His just heaven known. And it's the same consistency throughout. The volume is just turning down. Now none of these things by themselves can definitively prove that an AI generated this song, but when you start to hear several of these stacking on top of each other, that's when the dead space starts to become pretty obvious.
Chapter three. From what I've been able to understand generative AI music starts out as a big sheet of white noise. And through the generative process, it looks at real music and the waveforms and the sounds within it and tries to push them all down, compress them into what it thinks a song should sound like. But because you are starting with so much noise, there's always going to be some left in the artifacts that you cannot physically get rid of. It makes the high end feel super fuzzy and thin. It makes the low end feel super muddy and your overall mix feel scratchy and over compressed.
First topic is high end. This is the most telltale sign that a song is being generated by AI. This is the big drop moment in the song where everything's supposed to be sounding like it's larger than life and it is just being absolutely shredded alive by THIS HIGH END. LET'S SEE IF WE CAN ACTUALLY VISUALIZE some of this high end here. That's that high end that I'm talking about. Disgusting.
Next is consistency. AI has a really hard time trying to pick out parts of the song to stay consistent throughout. So you'll have very sudden and very drastic drop outs, changes, transitions, sudden switches in tonalities in the chord progression, the high hats, drums, snares, kicks. Let's see if we can catch some of these inconsistencies in the drum pattern. >> [music] >> All of these snares sound different. We've got snare number one, high hat snare. We've got snare number two, lot snappier, but still got that wetness to it. Over here we've got snare number three, a lot drier, and then over here this one's just straight up dry. It's just It's like all four of these snares up. Yeah, now look at how different each one of these snares look. This is insane. What? >> [music] >> That last one doesn't even sound like a snare. It sounds like a high hat, bro. Now, does changing a snare mean that you're an AR producer? No, tons of producers do it, but there is no reason to be using four different snares within a four-bar loop. See, the reason this is happening is because AI is constantly trying to replicate the thing it just created, but it forgets and can't and stuff.
Now, let's talk about spatial awareness cuz AI's got none of it. Every professionally produced song starts its life with dozens Scratch that. Hundreds of sounds, all separated into different tracks. You've got high hats, snares, kicks, synths, leads, guitars, pianos, counter melodies, reverbs, delay buses, everything all occupying this one mix. My job as a mixing engineer, put simply, is to balance out all of these sounds so they don't trample on top of each other, making sure that some sounds come out at certain times and some sounds are kind of hidden within the mix. Generative AI doesn't really approach music like that. It's version of a of a mix, a wall of sound physically being structured out into what it thinks a song should statistically sound like. And because of that, everything seems to take up the same sonic space.
This is one of my favorite tools from uh this wonderful company called Mini Meter, and it helps to visualize exactly how a mix sounds. Specifically, we're going to be looking at the stereo meter, which measures how evenly distributed your song is throughout. Since this Charlie Kirk song is a piano ballad, I figured it only right to put it up against a piano ballad of my own. >> Ain't no love for me. Now let's inspect the Charlie Kirk spectral meter graph. I can't I can't say that with a straight face, dude. It's pretty insane, isn't it? In my song, you can almost pinpoint exactly where everything in the song lives, whereas in the in the Kirk song, everything kind of just exists here. Just all all at once.
Chapter four. See, I can't be an AI. I don't have an extra finger. Don't look at that one. At this point, you might be thinking, "Okay, okay, smart guy. If you're so confident, why don't you throw every song into an AI detector?" I do, but it's not for the it's not for the reason that you think. There's a ton of misinformation floating around on the internet about AI music detectors right now. Tik [clears throat] Tok. Depending on who you ask, they're either complete scams or 100% right 100% of the time. I'mma be like the moist critical of the situation. I'mma fence it for a second. Neither are correct. You are both wrong.
An AI detector is a tool, just like uh something like Turnitin is a tool. Humans can trigger a false positive all the time for having something stupid like a bad mix, poorly processed effects, or AI mastering tools. If a song detects a high probability of AI behind it, it should be treated as just another piece of evidence against the song. The detector that I found the most reliable and that I use quite often is this one, this AI music checker from SubmitHub. And what's pretty cool about it is it runs two different analysis of the same song. Its spectral analysis creates a visual map of all of the frequencies within the song. [music] Similar to something that I showed you earlier, but imagine it like a like a heat map of the mix. The second is a temporal analysis, which analyzes how a song behaves over time. Things like timing, pacing, vocal dynamics, the consistency of instruments, and tracking how much those patterns occur versus in human performances. [music] Remember that dead space theory? This is measuring that, but like in a mathematical graph rather than with your ears.
So, yeah. Yap session is over, my guys. Congratulations, You have just passed. So, you guys want to test these practices and rip apart some Suno AI stuff I found on my For You page. Here, I've made a little cheat sheet for all of us to look at to determine whether a song is AI-generated or not and come to our best conclusions. I'll have a little download link in the bio if you want to snag it for yourself. It's free. Maybe you can hang all of your AI song ratings on your fridge. It'd be something cool to show your parents when they come over to visit.
Our first potential AI artist goes by the name of Kelsey Rogers. Now, Kelsey Rogers is the lead singer for a band called Folded Paper Hearts and has amassed 66.5k followers on TikTok, almost 25k monthly listeners on Spotify, and has amassed millions of views all across her social media. Unfortunately, that social media isn't looking too hot right now. >> Cuz I'm still chasing ghosts [music] down backroads, still getting lost >> Yeah, it seems like all of their content sort of follows this same AI-generated format. Even their story posts look AI-generated, which is kind of wild and like very slop AI-lazy. What, you couldn't have taken like a real picture of like songwriting lyrics? Nothing? I can't tell if this pencil is a pencil or a paintbrush. Like, are those those like hieroglyphics bleeding into the pencil? Like, what are we doing, man? Clicking onto their Linktree, we are blessed with some more lovely AI slop. Oh, hell yeah. I would love to get some of this low-quality Teespring t-shirts. Look at that. Real stories, raw emotions. Editor poet from the future here. I just got done recording everything and it looks like they've added a review our AI transparency page, which is a first for me if I've ever seen one. I guess they must have been called out on their very suspicious-sounding music. We are currently preparing a full transparency document explaining how Folded Paper Hearts operates and how our work is created. AI is used as a creative and production tool. It does not independently write, operate, or make decisions for the project. Well, I sure as hope it doesn't. That would be pretty embarrassing leaving the legacy of your music career in the hands of ChatGPT. Like, okay, dude, on one hand, I'm glad you're coming out with some sort of form of transparency for your listeners. But, doc, the the whole point of being a musician is to learn skills. When the day comes for you to do a tour, what are you What are you going to air guitar? Are you going to lip sync your vocals? I feel like all this time you've spent promoting and generating, you could have I don't know, shot in the dark here. Uh learn the skill, any skills in any art-related fields necessary to host a music career, rather than asking a chatbot to steal from others.
Anyways, back to back to present Ethan. Okay, let's look through some of this dead space. See if we can find some weird artifacts in these vocals. >> Didn't mention I'd be this devastated. I'm >> Well, that's one hell of a synthy vibrato. What? >> And I'm going down down >> Oh. >> I love >> Well, that mix sounds awful. Geez. All of a sudden we're getting like these metallic sort of chamber-like vocals. They just sound so grating. >> GOING DOWN DOWN DOWN >> IT'S like metal plates against a chalkboard. Jesus Christ. >> Again, I knew >> Of course, you can hear that uh Suno high-end right there. Let's try to isolate it out a bit. Ooh. Yeah. Eat up your AI slop. Isn't that so yummy? Let's Let's stop picking on this poor AI vocal. This poor clanker has suffered enough.
Start ripping apart the instrumental. These guitars sound like like like cheap metallic hi-hats. Like, it's got the strumming pattern of a guitar without like any actual guitar presence being there. >> [music] >> Dude, what What am I listening to? I love how the stem splitter put it in this category other. Even it's like, "I don't know what this is, man. It's maybe a synth, maybe it's a violin. We'll just throw it in here." All right, now let's take a listen to these drums. See if we can find some of those inconsistencies again. Here's snare number one. Here's snare number two. That's a very different snare. And that's also a different snare. What was that? Was that a hyperpop snare? No, go back. Go back. What the What What is that? You're just You're I know you're with me. Do not with me. Why am I getting rage baited by a crunkcore dude? I don't even need to pull this up on the spectral graph, dude. I think I think I'm pretty confident this is AI, but just in case, let's run it through the AI detector. And wowza, no surprises there. We've got a 90% on the spectral analysis and an 85% on the temporal. It's looking like Suno's the main culprit here. Based on everything I've been able to gather, this is looking like a massive fail at disproving the AI allegations, unfortunately. I'm pretty confident that this was made using Suno. Kelsey Rodger in the folder of paper planes, AI slop.
Next. The next Suno artist that crossed my For You page I found goes by the name of Cameron Dallas. Now, Cameron is a washed-up Vine star from 2017 that I'm going to be so honest, I had never heard of until I was talking to my roommate about him. He's like, "Oh my god, you don't know who Cameron Dallas is?" I was like, "No. I What? Why would I know who this white boy is?" From what I can gather, Cameron Dallas is one of those founding father influencers of Vine who got famous for being famous. His portfolio is like pretty widespread, and to be honest, that's really not something I'm interested in. So, let's talk about his music. Cameron Dallas first started making music in 2015, and the sound of it is just a little late for its era, to be honest. Not bad, it just sounds like, I don't know, Disney meets Jason Derulo. After this breakout hit that I can't play because copyright, it looks like he was signed to Columbia Records for a few of his releases before eventually being dropped in 2020 when he dropped his debut album Dear Scarlett. This album doesn't necessarily sound bad. It doesn't sound good either. It's heavily influenced by indie pop and trap and a lot of the modern trends that were around in the 2020s era. Cameron himself's voice isn't half bad, but there's nothing super special or unique or distinct about the songs enough for me to I don't know, go out of my way to listen to them. And then after this album, poof. He was gone, MIA for a whole six years until Cameron resurfaced with a DistroKid subscription and would you look at that? Since January 3rd of 2026, Cameron has released six full-length projects. I have a sneaking suspicion I know how this was made, but no judgment so we hear the actual material. This was the video I found that initially put me on to Cameron's shenanigans. And you have comments like, "It's giving Britney Spears. I kind of like it. I don't know." "Genuinely, what do you bring to the table? Did you get some hate recently that I'm just unaware of? Am I in the middle of some controversy?" Oh, I think we better get into the song material then. And also, if it wasn't if it wasn't blatantly obvious, uh Cameron's voice doesn't sound anything like his potential alleged AI creation. I'm going to lock and change over six years, don't get me wrong, but I did not think that uh getting a whiny nasally emo voice was in the cards for you. I thought you lost that after puberty. Holy crap, these breaths sound like sound like he's wheezing. I can't even replicate that. God, Cameron, are you okay? You should see a doctor about that, dude. You got to like a like an inhaler. Oh, that was a weird moment. What What was that? Why'd you gasp for air at the end of a verse? Yeah, I know. That was just a really hard verse. I got to like you know, double double gasp. Don't you hate it when you're mixing your vocals and all of a sudden they just they fold into paper and turn into a synth?
All right, let's stop let's stop looking at the vocals, start looking at this instrumental a little bit. I think this is supposed to be like a filtered saw wave, but it's not like fully formed yet. It's like ghostly, like part of it's missing. Oh, I bet you I know where I can find the rest of it. Wait. >> [music] >> There you are. It looks like yet again the AI has stolen the synth and mixed it in with the vocals. How dare you, Suno? Uh, looks like we got another case of where's my snare. I got no snare in my headphones. Those are four completely different snares. And no surprise from anyone, we've got a staggering 98% on the AI spectral analysis, a 54% on the temporal. I'm sorry your career flopped, Cameron, but that's no excuse to be AI slop posting on your main. Go do something normal, like Ned Fulmer, start doing stand-up comedy, dude. Looking like another case of clanker garbage.
Next up on the AI chopping block, we have an artist by the name of Lovex. I I I think it's pronounced Lovesick. I'm going to call it Lovesick for the for the sake of this video. This one's a bit of a curveball, just because of how prevalent he is and irrelevant at the same time. Lovesick has amassed a massive audience on social media, 1.7 million followers, 43.6 million likes. His views are passing the tens of the millions, and so are his songs. All of his content is mixed between stock AI video of sad emo boy dancing around in a mask and like clips of random people just dancing around to somewhat what the song is titled. All with this sort of kind of gray filter on the overlay. This is the epitome of staring into an artistically bankrupt void where a content mill dressed up in Zumiez and Spencer's music just spoon-feeds you AI slop. Allegedly. We've got to do our due here. Come on. Now people saying song name, "Amazing. Anytime I hear this song, my favorite song at the moment. This hits differently when you've actually lived it." This is the This is the song in question that that they're actually living. >> [music] >> Yeah, I know. I I wish I was a millionaire, too, dude. Let's jump into this song.
All right, we're getting some like weird weird aberrations over here in the vocals. See, this is some of that noise I was talking about that you just can't get rid of, so it's trapped in this high-end area of the vocal. I'm actually curious, can you get rid of it? Hold up. We're going to pull up our wonderful RX tool. This isolates all of the noise, so you can hear it. Then let's solo out just the noise. That's insane. What? It's It's detecting the entire vocal because the entire vocal is made up of noise. That's so interesting. Also, I'm starting to notice this uh sort of reoccurring pattern with AI songs. A lot of the production isn't doing anything crazy. Like the main appeal for all of these types of songs is the vocalist. So when you have like a no melody type beat like this, it is astonishingly boring. >> [music] >> What Like what do you mean that's that's the beat, bro? Surfing through, and you're not going to believe this, we found some more snare problems. Got some See, like I can see an argument behind, "Oh, I don't have the skills necessary to do this. Music costs too much money to make, dude." I'm telling you there are so many free sample base libraries out there in the world. You can make a beat just like this, but even better, in GarageBand for free right now. Like in the in the same genre, same tempo, same same everything, with no experience. Like this is not hard, bro. Genuinely, there is not a single excuse that you can make for why this beat deserves to exist in the world. Once again, the AI checkers have flagged this as 91% AI on the spectral and 47% on the temporal. Shocking.
The last artist we are going to be talking about today, the one who started this whole chain of events, let's finally find out is Neffex AI or not. >> [music] >> This is another nondescript artist who does not show their face with AI-generated videos and visualizers dressed in cyberpunk and cybergoth aesthetics. He's quickly amassed a pretty sizable audience from his first single on TikTok and has attained hundreds of thousands of monthly listeners, millions of streams, and an active growing fanbase across social media from people using his sound. And I found this a bit funny. When you scroll through the credits of his song on Spotify, you'll see him as the main artist, composer, lyricist. It's just him. But I've never seen this before. Under performers, it says programming, which is which is a first. You know, I'm I've never seen that before. From what I've gathered, Neffex will adamantly defend that his music is not AI-generated in the comment section of other people's videos. So, we've got a video of this guy vibing over here. He's thinking it's AI. The caption is "Too creative to be AI?" {question mark} And we've got Neffex in the comments. "So funny how people act like they know how I made the song." Well, hey, that's what we're all here to find out, my man. Let's see if you made this song.
To start, I think the most glaring issue with the song is the vocal consistency. Like this happens multiple times throughout the song, but the vocalist literally just gender switches on us out of nowhere. And it's not like format-shifting vocals. But off rip, we are blessed with an immense amount of that lovely paper noise that has made me suspect that this was a Suno-generated AI song. Hold up. Let me try that uh Let me try that noise gate trick again. That's insane. That's all noise. That's all noise I picked. That's the most amount of noise I think I've ever picked up from a song before. What? Like I I I expected there to be a lot of noise, but But This is insane.
Moving on from the vocals, let's take a Let's a little bit of a peek-sy at this instrumental here. Once again, it's looking like a pretty minimalistic sort of outline of what a trap song should look like. >> [music] >> Looks like we got some like EDM stab bass. >> [music] >> Now we got some got some house bass. Now we got some super rubbery reese bass for this like sort of drop area. What? >> [music] >> How does how did But now Now we got jazz bass here? Like, sure, why not? None of these basslines have been consistent at all so far, and we are 30 seconds into the song. Four different basslines in the in the turn of 30 seconds. That's not very consistent at all, chief. Oh, you're never going to believe what I found. More snares. It's more It's always snare. I'm just a snare, and I know life is not fair. Snare number one. Snare number two. This one's a snare clap. Listen to that. Let's listen to this. That's insane. I'm not going to lie, Neffex, it's not looking very good against the AI accusations right now. The verdict is in, and it looks like this song thinks you are spectrally 82% likely to be AI, and 72% likely on the AI temporal analysis. Stacked up against everything so far, I think I'm going to have to side with the detector on this one. There are way, way, way too many inconsistencies throughout the instrumental and weird noise aberrations for me not to suspect that this song was AI generated, even if it is a song that I do find interesting. And so did a lot of other people. I mean, on this TikTok alone under the Neffex comment, there's people saying things like, "Banger anyways. I love it. It's funny how people act like they don't know that this is AI. I don't care how you made it, just make more." Either way, even if you used AI for a small portion, it's still an original song that a human thought of. You didn't just type this in and magically robot spit it out, right? I'm sorry to inform you Patterson with an exclamation point. That is exactly what just happened here and a little bit more.
Although it seems like a very surface level problem of generating AI music for other people's enjoyment, do you know who doesn't get any enjoyment out of it? It's the people whose music was stolen to train these AI models on. While AI streaming artists have been able to accumulate millions of dollars off the backs of companies like Suno, real musicians are not only not being reimbursed for their work, but they're losing their streaming revenue to these massive AI architects. But more importantly, we're all losing our minds.
After finishing this video, I kind of tail spun a bit and I I ended up calling my dad. He works in AI professionally and I asked him, at what point does an AI no longer become a tool [music] and start becoming the author? And I think that's the question I've been struggling with this entire video. I don't think using an AI makes someone or something inherently worthless. I'm embarrassed to say this, but I've used ChatGPT before on the dumbest things. Things like how to cook chicken properly or calculating the dimensions of my bedroom so I can best adjust my desk to the wall. Producers have been using AI-powered tools for ages now. Things like stem separation, noise cancellation, media extraction, mastering assistance. And these tools are used to help execute your ideas. But at some point, and I say this with extreme caution, there is a slippery slope where you go from finding rhyme schemes to writing lyrics. From generating synth patches to generating whole instruments to generating beats. Using your AI to conduct a performance and then by the end of it, just generating your song outright. We are no longer the author here. The more time we let a machine think creatively for us, the less time we ever think creatively at all. Because we've stopped exercising that part of our brain.
I fear the day that comes that AI gets so good that I won't be able to spot it anymore. And I just hope that when that day comes, there will still be people out there who value the person who spent years learning how to play an instrument or values the time of the producer who spent 12 hours ripping out their hair on a mix, or the singer who keeps that imperfect take because he thought that it was just a little bit more vulnerable than usual. I'm not concerned if AI will someday make music. That's just an inevitable at this point. This video was never about that, or if an AI can make music, but if we'll continue to value the people that do. Can you spot whether the song was AI? I think the more important question here is whether it matters.
Thank you for watching to the end of the video. Sorry for the inconsistent upload. I've been a bit busy. I moved. I'm going to art school for film and photo. Yippee! New music on the way. I've gotten really big on platforms like Instagram and TikTok for my mixing videos. So, if you're not privy, come check me out. I post on there pretty frequently. Never stop creating. Keep pushing those boundaries, and I will see you in the next video.