Ep. 75: How to Make AI Images That Don’t Scream “I Used AI”

0:00:00Who is Finn McKenty?

Play from 0:00:01 8:88:88 0:00:01 What's up, everybody? Fast Hours Podcast, episode 75. Drew Brucker, Rory Flynn, and we got a special guest today, Finn McKenty. Welcome this man to the show today. We're gonna have an awesome conversation, kind of digging into his specific experience, talking about what's happened in AI. Probably give you some really hot takes along the way. So I think this is gonna be a great episode, but Finn. a mediocre conversation, that's the best I can do.

Play from 0:00:27 8:88:88 0:00:27 Okay, fair enough. we got a very large ceiling here. The sky is the ceiling. Space is the ceiling. So why don't you get us started? Tell us a little bit about yourself. Describe yourself and what you do to uh the listeners today and kind of get us rolling. Uh well, I I always uh I I always get annoyed when guests on podcasts go on too long about their life story. I'm like, uh dude, I don't care.

Play from 0:00:53 8:88:88 0:00:53 Just tell me something tell me something interesting. So I'll keep it brief. Long story short, grew up making like punk zines um, you know, in my parents' basement and stuff and that was sort of my introduction to, you know, being a creator, which wasn't really a word back then, or being an entrepreneur, which also wasn't really a word back then. And then I realized like, okay, I can do this for a living. I can be like a designer and a marketer. So uh I did that sorted out in the agency world.

Play from 0:01:20 8:88:88 0:01:20 worked on uh some stuff like Nike Nintendo Red Bull in the beginning and then worked at an industrial design and engineering agency for about five years mostly Procter & Gamble stuff like Tide, Febreze, Bounce, Swiffer. Uh, then went to in-house role at Abercrombie & Fitch. I there for about four years. First did design for Hollister girls t-shirts when it was kind of, you know, the number one mall brand. Yeah, no, it was yes, it it was it was peak. That that brand was No, this was like late two thousands.

Play from 0:01:48 8:88:88 0:01:48 It was still like 'cause Hollister didn't come out until 2000. Yeah. So so like what are talking? Like two thousand eight, two thousand nine ish? was there from two thousand nine to twenty thirteen, yeah. so I did design for Hollister Girls T shirts then. Um and then I moved to a marketing role there where I did a lot of stuff. We were opening a lot of flagships and stuff at the time in like Europe and Asia. So I spent a lot of time Over there also like factories and stuff like that. So that was cool. Um and then went to a startup called CreativeLive, moved back to Seattle, uh startup called CreativeLive that did like online education for creative people.

Play from 0:02:20 8:88:88 0:02:20 Um we raised about sixty million bucks from like Chamath and Reed Hoffman and you know Richard Branson and people like that. and by we I mean the CEO Chase. Um so you know, I was there for the ride, um, credit to Chase. Uh and then uh after that was doing YouTube and some other stuff. A lot of people know me from YouTube. ended up getting about 750,000 subscribers there between my two channels. Um and then I quit doing that.

Play from 0:02:46 8:88:88 0:02:46 And since then I've been doing, I guess what we'd call like a fractional CMO kind of thing. Um for you know last time companies most recently started working with a company called Brotege, which is a skincare company started by Gabe from Cobra Starship. If you guys remember the he did like the snakes on a plane theme song back in the day. super, super super smart guy, very good entrepreneur though. So I've been working with them for a couple months now and that's been a lot of fun.

Play from 0:03:12 8:88:88 0:03:12 And now I'm talking with you guys. Dude, that YouTube subscriber count is nuts. I'm curious, like, wh why quit that? Or have you found a channel to be better for you and your business and and maybe revenue or just exposure than YouTube? no, um, but that's okay. I mean, you know, that that's that's not, you know, I don't wanna be a YouTuber, you know what I mean? Like that that's cool.

0:03:13Why did Finn quit YouTube?

Play from 0:03:38 8:88:88 0:03:38 I'm uh grateful that that I had that opportunity, but like I like making shit, you know? And uh, you know, what I've found is like Twitter to me is where I meet it's not the biggest, but it's where I meet that's like where your one fifty IQ autists are and that's me. So you know. We're also gonna, you know, don't want to downplay Finn's just like, I don't want to say, I don't want like, I don't want to blow smoke, say brilliance.

Play from 0:04:09 8:88:88 0:04:09 But I, the reason I met you was probably like, I don't know, two years ago, I saw a post and it was the most self-aware, like straightforward, brutally honest post. was like, I think you put a picture up of yourself on stage and your hook was like, I'm on stage, so you should listen to me. And I was like, Here's a picture of me on the stage so you think I'm important. And I was like, this is so self-aware. And like, this is like the perfect type of content for me. And I was, I reached out to you right away.

Play from 0:04:35 8:88:88 0:04:35 I was like, that's the best like sales pitch, non-sales pitch I've ever seen. So like, cause that's exactly how I would do it. I'd be like, this is an ad, like for my business, know, like, dude, it's, you know, if anyone doesn't follow Fin on X or LinkedIn, you know, very unique, I would say not, not a unique, but like a very tailored content strategy. There's, there's thought. behind everything. To me, it's very well put together, whether that's the visual content or the written content.

Play from 0:05:01 8:88:88 0:05:01 Some of the memes are incredible also. So throwing that out there for anyone who hasn't seen it yet. yeah, so. oh he's you're kind of spread out across channels here. You got me you got a really good follower base in all places. So I was you know, I pulled up your ex here. Uh I know you post there from time to time. 20, 28, 29,000 followers there.

Play from 0:05:26 8:88:88 0:05:26 You've got an Instagram with ADK, you've got the YouTube thing, which lot of these people are people that follow me from YouTube that don't give a shit about anything I do now. Which is fine. Like that's okay. you know, that's sort of one of the things you have to decide as a creator is like who who do you want to talk to and like right now I'm optimizing for I want to connect with people like you guys that are doing cool shit.

0:05:30Do creators need a large audience?

Play from 0:05:52 8:88:88 0:05:52 They're like on the bleeding edge of sort of, you know, making interesting things and having interesting ideas and you know, I mean I lost like twenty-five thousand followers on Instagram over the past year or whatever. I I don't care. Like getting one follow from like a creator I respect is worth more to me than ten thousand people who don't like me anymore, you know? Like literally one person. I I don't I don't I don't care. Yeah. I mean the bots.

Play from 0:06:19 8:88:88 0:06:19 They're there for you. whatever. that though. I I think we're all kind of on that same page here. Like we, Rory, we've talked about this a lot, but it's never been about follower counts for us. Like, and and what's driving me nuts right now, like God, Instagram is probably the worst with this right now because of the ManyChat thing. I mean, every single content I see that's in my feed this day is asking for a comment of this to receive this.

Play from 0:06:46 8:88:88 0:06:46 And it is. yeah. disgusting, you know, what's happening there. I mean, like it is just, I'm so fatigued of seeing that. And then, you know, just from like the marketing days, you know, like the first couple of times I saw that, I was always curious, like, what what are they? What do they have here? I was kind of analyzing it from two lenses. One, I might have some interest, but two, like from the creator side, what is the value attached here? And nine times out of ten, it lets you down.

Play from 0:07:14 8:88:88 0:07:14 And to me, that's that's a huge just yeah, yeah. Yeah. meet, you know, that standard of, you know, like the way that I explained this in the video gives you this. And then it's just like a, it's a, it's a doc file. It's a notion or, you know, something, you know, it's just like, what are we doing here? Yeah. Yeah, exactly. Like e exactly. I just I don't wanna play that game. I don't want to play that game. I get that some people, if that's what you want to do, so be it.

Play from 0:07:41 8:88:88 0:07:41 But it's not what I want to do. I only want to connect with people who like I respect who are doing like cool bleeding edge shit. That's it. And by the way, for anybody listening, you do not need to have a big audience to have a great career doing shit that you like and making a lot of money from it. Cause, you know, so w whatever. Uh I got 130 million people to watch my videos, which is amazing and great, and I love it.

Play from 0:08:07 8:88:88 0:08:07 I'm grateful for that. But you know, the the people who hire me now, it's probably f it's probably like twenty people that really like matter, you know what I mean? And I'm not like devaluing the people that watch, but my point is like they're not they're not going to hire you. So if you want to sell a low end product, yes, you need like a huge number of people, but if you want to do, you know, client work or sell a high-end product, you don't need a big audience.

Play from 0:08:33 8:88:88 0:08:33 You need a couple dozen people that know you and respect you, and you're good. Well said. And I think that's also sort of how, when we talked on the phone the first time and you sort of laid it out for me, like, oh, Finn understands this game. He understands the content really well. at that point, was really just like, I'd say it wasn't like this wasn't, AI wasn't like the thing. It was like this sort of like niche, it felt like more of like an underground sort of movement, right?

Play from 0:09:03 8:88:88 0:09:03 Like it was. credit. You're the person who got me to see it in a different way. 'Cause at the time, like it was kind of shitty for the most part. Like GPT was bad. I don't remember if GPT four was out then or not, but like GPT was just not very good then and you know, Stable Diffusion, like it it it's cool and I respect it, but it wasn't like that good at the time, you know? And you were the first person to like show me the potential of what AI could be in the hands of someone who was an actual professional.

Play from 0:09:30 8:88:88 0:09:30 So I gotta thank you and give you Well, thank you, sir. It was mutual. I felt like when we got on the phone the first time, was just like, we didn't even talk about what we were supposed to talk about. We just went down the rabbit hole. And then I was like, once I found out you had a design background, which I think is a good segue into this, because it's changed so much. Me coming from a graphic design background, not at any level of Abercrombie & Fitch or Hollister, more so making t-shirts for Discovery Network.

Play from 0:09:58 8:88:88 0:09:58 But like, difference. I mean, you know, not the same level of pop culture-like influence as, you know, if anyone on here do working for like a big company of any kind is hard. And like if you do literally anything for like a Fortune five hundred company is hard. And you know what you're doing, you know. It is and I think like, you know with the one thing that that definitely you know Wanted intrigue me more was that it wasn't just that it was like you had like a bunch of other skill sets that you were tying into this and I think the design thing is you know for me from a question standpoint like Back then tools were you like, you know, was it primarily Photoshop?

Play from 0:10:35 8:88:88 0:10:35 Like what were you using like in that giant corporate space? Like, you know, it's all different obviously you stuff yourself, but what was it? Okay Illustrator, Premiere, nothing fancy. You know. the skills started, or you started something earlier than that. well I started using Photoshop before it had layers because unfortunately I'm really old. Um Photoshop th Photoshop three I think is when they got layers and the first one I used was two point five.

Play from 0:11:02 8:88:88 0:11:02 so yeah. Yeah. disks back then? I'm kidding, but seriously. of course I was. Yes, of course. Um, I was super into like and you know, so I I've been my like my stepdad was he fixed airplanes for Boeing, so he had he had like an electrical engineering degree. So I remember when I was like eleven or twelve, he sat me down at the kitchen table and explained like a lot of computer science and electrical engineering basics like bus width and eighty eighty six architecture and how to do hex math and like all this kind of stuff that, you know, I didn't understand at the time how important that was, but because of that like I was always also into technology.

Play from 0:11:39 8:88:88 0:11:39 So I made my first website in like nineteen ninety five or ninety six. and I got real into like flash and shockwave and stuff at the at the time, you know, in like late nineties, early two thousands. I was making games and stuff like that all the time. And so, you know, to me it's always been about like and and why and segueing into sort of the current moment with AI is like that now anybody can like basically do anything.

Play from 0:12:04 8:88:88 0:12:04 And so to me I I I I'm tool agnostic. It's always like, what do I want to do? And then what do I need to learn in order to like make that thing real? You know? Um, and that's the great thing about AI now is that's so much more attainable. Like, I was like, I want to make a game. Okay, how do I do that? Okay, well, Flash is one way of doing that. All right, I'll learn Flash. Okay, well, that took me years. Now it doesn't take you years, you know. But if you do, the great thing is also that if you do have a background in anything, like that is your unique.

0:12:05Which existing skills create an AI edge?

Play from 0:12:36 8:88:88 0:12:36 advantage with AI because you know anybody can prompt anything, but that knowledge you have of whatever that specific subject matter is, whether that's design or art or construction equipment, it doesn't matter, like that is your edge. Because you're gonna be a hundred like, I don't know anything about bulldozers. So yeah, you know, I can prompt about bulldozers to Claude just like anybody else, but I guarantee it's gonna be bad. If you know about bulldozers, you're gonna be a hundred times better at it than I am.

Play from 0:13:03 8:88:88 0:13:03 Subject matter experts are just as valuable now, if not more valuable now than ever too, right? Because you you know, like a background in you know, I'm like I'm not an engineer, but I'd say I understand computer science pretty well. So like the fact that I have that background plus like a design background was kinda like it it felt like um it it felt like a perfect storm of like, okay, I can use like my powers of autism, you know, on the technology world to also like do creative shit.

Play from 0:13:38 8:88:88 0:13:38 This is amazing. Yeah. nice blend that happened and it happened like, I don't want to say it happened like organically, but it kind of did. It felt like there were kind of two separate paths that someone was just like, we need to start like mixing this together. Cause it wasn't, I don't feel like in the beginning of AI, like the LLMs were like, we're going to do design, you know, or the, or the creative tools are like, we're going to end up integrating with the LLMs to do design and code. I know there's like a big debate all the time about, you know, design versus code, but it feels like it's sort of having its marrying moment now.

Play from 0:14:09 8:88:88 0:14:09 which is nice. ah to me, but and to I think to a lot of p there's a certain type of person that is sort of wired to to to think that way and those people are thriving right now, which I I think you guys are are like that too. D. I have go do you wanna go, Rory? I was just going to ask on that one. you think like now because because you have the experience, right? Coming into it with a lot of these, these different sort of let's just say areas of expertise, little micro skills or things you've just learned over the accumulation of a career.

0:14:26Does expertise accelerate AI mastery?

Play from 0:14:41 8:88:88 0:14:41 Do you think it was easier to get like into AI and, and is that a, is that a actual benefit having that sometimes or is it like a hamper? Cause uh I like I got I would say pretty good at it really fast because of that. Because because I know what I want and I know how to get there and you know, I think AI is basically a function of your Subject matter knowledge on one hand and systems thinking ability on the other hand and uh I'm naturally, you know, a strong systems thinker and I have a lot of random subject matter knowledge.

Play from 0:15:20 8:88:88 0:15:20 So like i it was like this is amazing. I I just wish I wish this was around twenty years ago. The only, the only reason I ask is cause like I've done some stuff recently in like the, in like the 3d pipelines. have no business doing stuff in the 3d pipelines. I have no experience. don't know anything, but like, I don't think in the way that the 3d designers think and the 3d artists think. And because they're, they're thinking about it from like how their process would typically work, right?

Play from 0:15:49 8:88:88 0:15:49 Like, how are we going to take this CAD model and turn it into a production asset? They have like a typical like process steps that they go through. Right. just how they've always done it. That's how they were taught. I come in, I'm like, how about we just have Claude do this part? They're like, what? I'm like, it can write code. can, those tools can receive code. Like how do we marry this? they're, they're just, because I have no experience in the process, I found it like easier to integrate and like find little ways instead of thinking it has to be done like the meticulous every step of the way.

Play from 0:16:18 8:88:88 0:16:18 Um, piece, which is there's nothing wrong with that, right? It's just like, I don't have the experience to know that. And they're like, well, why do you think about it? I don't know, because I don't know what I'm doing. Yeah. it this I I think I think there's like a meta skill here which I would call systems engineering that I would say that's what you do. You know, sort of system systems engineering, you know, in the domain of creative work, um, which is like a meta skill.

0:16:27Are creatives too attached to process?

Play from 0:16:43 8:88:88 0:16:43 And I think there's probably you know, there's lot of people in creative roles that I would say are um maybe a little too tied to a specific way of doing things. You know, like they think their job is to operate a particular tool in a particular way. You know, whether that's in whereas I would say, like if you're a photographer, your job is not to operate a camera. Your job is to produce images that fulfill, you know, whatever need.

Play from 0:17:11 8:88:88 0:17:11 Sometimes you're gonna do that by pressing a button on a camera, other times you're gonna do it by prompting. Probably a lot of times you're gonna do a little bit of both. And maybe there'll be a different way to do it next year. I don't care. You know? And so I I think that's kinda what I've observed is like people that are tied to a process versus people that are interested in a result. And I think that might be what you're talking about. Like I d I'm completely indifferent to the process.

Play from 0:17:37 8:88:88 0:17:37 I don't give a shit. Like, as long as we get to a good place, I I'll I'll change the way I do everything tomorrow. I don't care. That's the weird thing I found about this, right? Is that like, because some of the old, not the old, I shouldn't say it that way, but some of the grandfathered in systems that have always been in place, right? And everyone uses them and that's just how the whole organization runs. Like I find myself when building systems, have to take that into account so that there can be either checkpoints or things that still line up with that so that it's like comprehensible, right?

Play from 0:18:10 8:88:88 0:18:10 a big company you can't just like throw out everything overnight. You know. The integration is hard. That's why I think You know, I I think that, you know, they're calling it forward deployed, which is kind of a corny name. Um, you know, given that like you know, my whatever, like my brother in law's an actual like forward spotter, you know, forward observer in the military who, you know, so like his job is like literally to be forward deployed and you know, get get shot at versus engineers, meaning you go to the client's office.

0:18:17Why do companies fail to adopt AI?

Play from 0:18:40 8:88:88 0:18:40 Um, but I I think this uh I think this idea of forward deployed is gonna be really critical because There's just no chance. Like the idea that every company is gonna bottoms up adopt AI, I think is just fucking complete fiction. It is not going to happen. They don't have the culture, they don't have the expertise, they're too busy like doing their jobs. Like there's no fucking world where the vast majority of companies are going to ad like if they're going to adopt it, they already would have by now.

Play from 0:19:10 8:88:88 0:19:10 Like it's just not gonna happen. So I I think the fact that there's people like you or, you know, that I know OpenAI and Anthropic have and FLORA also has done this, like they're like spinning up, you know, whole teams of like four deployed people. I I think that's gonna be the way. Truly the only way, I think. I think you have to have someone like boots on the ground and like hand holding and some of that's also just visible into the whole thing because you don't have, know, that was one thing I realized early on when building systems for some of these companies is that they're, they're, they're looking in, they're, not looking in the right places because they're so used to the environment, right?

Play from 0:19:47 8:88:88 0:19:47 So it's like the environment is so comfortable. And so like, they don't look in the areas where I'm like, why is, why there? Why are we doing that? Yeah. this is over here. Why don't you just have the API do this? Oh, yeah, good idea. And then they don't want you know, they're also so like, you know granted Everyone is so goddamn busy with what they have to do to begin with and their their current role and their current deliverables that it's like there's not someone that's thinking about this 24-7 of how to like Continuously figure out how to integrate this stuff.

0:20:17Why does advanced AI work cause burnout?

Play from 0:20:18 8:88:88 0:20:18 So it's already tough to be someone's basically full time job and there's just it's just no way you c it's it's crazy to think someone's gonna do their job and then also like do this extremely cognitively demanding work. It's just not easy. Like I I think, you know, again, I've been like doing this kind of stuff for a long time. I think be interesting what you guys think. I think AI it like I feel cognitively tired for like the first time in my career.

Play from 0:20:45 8:88:88 0:20:45 Like I can't do this shit like eight or ten hours a day. Like I I can't. Which is the first time I've ever felt that. know what else to it's just the the context switching was always a problem. But now it's, you know, we we thought that was going to be solved. But the truth is now it's just like you're running separate threads and you're just go you're waiting for certain things to to finish with the output. Meanwhile, you're reading the one that is finished with the output and then thinking what to do next.

Play from 0:21:13 8:88:88 0:21:13 And then you've given it that next prompt or iteration or direction, and then the next one's ready. You know, so like I'm bouncing between maybe three or four like high leverage things at a time. And so it's like, how deep am I really thinking about that? And what kind of strain is that putting on me mentally as well? I think that's a really overlooked point. I I mean, I've definitely felt more pockets of mental exhaustion and burnout in the last two, three years than I had previously.

Play from 0:21:45 8:88:88 0:21:45 Maybe that's some some of its old age and getting hitting forty now, but I'm just I everybody that's like using AI at a high level. Everybody says this and and it it hasn't been talked about a lot yet, but I think it will be very soon. Yes. And it going back to your okay, so I have two things for you. Going back to your point about some companies just not being able to deploy AI efficiently. Do you think though, and you and you said that it's it's not possible and maybe some won't.

Play from 0:22:12 8:88:88 0:22:12 I think like where there could be a trap there is if you want to, but you just never get caught up, versus making a declarative statement or decision that you're just not going to do it at all. Right. Like I mean, I think there are decided they don't want to, you mean or? mean, I think there are probably like limitations or thresholds that are appetizing for how they would do it. And maybe they're just realistic in terms of like, hey, we're just gonna do these low table stakes things and move on.

Play from 0:22:41 8:88:88 0:22:41 And you know, maybe eventually it catches up to them. But I feel like maybe the ones that are in the most danger are the ones that actually wanna do it, but they don't do anything proactively or even reactively to solve for it because it's too late. Right. And to your point, having been in-house and knowing how these things work, I mean, they There are rarely meetings set aside to talk about integration adoption and these big changes. And even if they are, they take too long for so many reasons, mostly because make a decision there's been like three model updates and whatever you decided on is irrelevant.

Play from 0:23:13 8:88:88 0:23:13 Yeah. mean, I experienced that with clients, you know, where it's just like, hey, like we need to get approval on these models to use it. And by the time they get approved, there's already a new model. So, you know, like there's that sort of delay that happens. And I think that happens across the board because all these people that are in-house, they also have five or six other projects they're working on at any given time. So it it they're stupid or lazy or something like that. I'm just saying like It's just not no, it's not, it's not realistic without really going in and to your point, going into like this enterprise ready integration adoption or hiring somebody, you know, like one of us to go in and do it, right?

Play from 0:23:52 8:88:88 0:23:52 Like it's just not gonna happen unless you were to commit at that level. Um I I do want to go back to one particular question I was thinking about for a little while, which was we were talking about your skills, right? And you're Career arc, you've, you know, you've got many things that you're interested in, many things that you've done, the punk zine stuff, the design, the education, the YouTube, the AI, art, you know, all this, all this stuff. Um, is there a skill that you think of that has compounded over the course of your career and has helped you in that sense?

Play from 0:24:24 8:88:88 0:24:24 Right? Like make those You know what? I knew you were gonna say that. I knew you're gonna say that, but joking. I I mean I would not recommend that people do what I yeah, right. Well I uh no I wouldn't, but I wouldn't recommend that people do what what I've done. Um I like I you know, without getting into too many too much whatever stuff that doesn't matter, like You know, like I grew up around like basically a bunch of like drug addicts and criminals and shit like that.

Play from 0:24:57 8:88:88 0:24:57 I did not know anything about the way business or money or anything like that works. Like anything that I have done is because I like brute forced it and figured it out on my own or maybe met somebody that told me how to do it and I listened to them or whatever, but like nobody should take what I've done as an example of like the smart way to do things because it's definitely not. It is the best I could do with, you know, my extremely limited understanding, you know, or whatever.

Play from 0:25:26 8:88:88 0:25:26 I was born fortunately, you know, with this brain, so I was able to like figure it out, but like I would suggest anybody else be more focused than than I was. I'm just like taking a wandering path trying to find my way out of this fucking scary forest. know it's not a maybe an optimized path, right? Like I mean it rarely is, but like what what do you like is there something that you try I mean, is it grit? Is it curiosity? Like thinking, you know, that's just sort of, you know, like in in like uh as I've learned, like in in cognitive science, like one model is like there's there's there's a model of cognition with E-type and S-type cognition.

Play from 0:26:03 8:88:88 0:26:03 E-type is empathizing, S-type is systemizing. And you can be good at both. Or you can at you can sc you can rank highly in both, but usually people are sort of one or the other. but like the the ability to like uh understand a system, deconstruct it into pieces and see how they interoperate, and then, you know, pull the levers, I I would say is like the thing I I'm good at and and I don't as just I was born that way.

Play from 0:26:32 8:88:88 0:26:32 So I don't think I get any kind of amazing credit for that. Does that make it does that make you feel like is that like a comfort thing like when you can do that like from a you know if you're looking at something it's like oh like that's like Not the not like you're your zone of genius But like that's where like you hit flow state kind of stuff is when you're in that systems mindset, okay it ju it comes to me like I just it just it comes naturally to me. I didn't like it's just like being able to like jump high.

Play from 0:27:00 8:88:88 0:27:00 You know, my cousin's six ten, he played pro basketball for a while. He's just born that way, you know? So how I can see this sort of like how what you were saying, where it goes from that to like the I've been on a 10 hour bender with Claude and now my brain is going to pour out of my ear. Right. Like once you get going with something, you keep getting results and you keep like finding the little nuggets in it. You're like, that's another little clue.

Play from 0:27:26 8:88:88 0:27:26 that's another little clue. The other thing is it's and I don't know if this is two questions here. One, because I do this now. I read everything that whatever these LOMs put out so meticulously because I know that one word screws up the whole thing, right? And if I didn't catch something that they say or I don't catch something that I say or if I wrote something in the wrong context, it's not going to understand it and it's going to be the wrong thing. Like you have a software background.

0:27:33Where should creators invest brainpower?

Play from 0:27:51 8:88:88 0:27:51 Do you still read the code that it outputs or is that like a, is that something that you just like trust? Well, um, it's not possible to like read all the code and it's not necessarily desirable, but I think the the skill is like I mean, it it's just humanly impossible to like look at every piece of these systems in detail. You know what I mean?

Play from 0:28:17 8:88:88 0:28:17 So I think the skill is like knowing where to invest. Like think of your like you're you know, like we've talked a lot about like getting mentally exhausted. Like your mental stamina is a finite resource. And I think it's about like understanding where to choose to invest that. You know, like which part of this am I just gonna kinda cross my like, I don't really know exactly what's going on here, but it seems to be okay, so I'm gonna let it go. Versus this one I'm gonna go way down the rabbit hole and make sure I understand every single piece of this.

Play from 0:28:45 8:88:88 0:28:45 I think I think that is like a a meta skill of like knowing where to invest your brain power. Um one thing I do is like I read the manual and I talk to like I use like I use LLMs to like you know, I have like a thread of like going super super like I want to know like so with like images, for example, like I have a thread where I've been going on back and forth for like months, like understanding every single part of like image architecture from like you know the the like embeddings like clip versus sigglip and like how those work and blah blah blah blah blah.

Play from 0:29:20 8:88:88 0:29:20 And I want to understand like you know what I was just asking this morning, what are the different dimensions of a tensor and like how does that work for like diffusion versus autoregressive models and blah blah blah. And I don't understand all of it, but You don't need to understand all of it necessarily, but I do think there's like a lot of value in going down the rabbit hole on these things because Um where you get into you know the the the the problem with AI is that it is probabilistic.

0:29:31How do creators steer probabilistic AI?

Play from 0:29:47 8:88:88 0:29:47 And so like The more you understand how to steer it, you know, think of it as like there's a literally a trillion steering wheels in this thing, you know, and the more you understand what each of those steering wheels do, the better you can get where you want to go. And so like the the the slop is just when people are sort of pressing on the gas with no hands on the wheel, you know. And so if you're listening to this, you want to move fast, but you also want control.

Play from 0:30:16 8:88:88 0:30:16 And I think like reading the manual, like literally any any image model will put out like documentation. Read it. And I don't mean ask, don't like, don't ask it to fucking summarize it for you. Read it with your own eyes and go slow and think about it. And if you don't understand something, you know, research it. And then you do that for a couple hours and you will understand the model better than everybody else, which means you'll get better results out of it.

Play from 0:30:45 8:88:88 0:30:45 So is that sort of how, because I definitely want to segue into some of the Midjourney and the Weavy stuff that you've been doing. Is that sort of how you attacked it right from the beginning when you first jumped in? Yeah, yeah, like yeah, exactly. I'm like, well, okay, how does it know that there's a blue shirt in this image? Okay, that happens in a thing like okay, embeddings. Okay, well how does that work? Blah blah blah blah blah. You know And when you understand all of that, again, you just you understand how to steer the car.

Play from 0:31:14 8:88:88 0:31:14 And I feel now very confident that I can basically make anything I want. You know, you give me enough time and you know, enough credits, I I can make anything because I understand all the all the levers that I can pull. Give us the Midjourney. You started on Midjourney, right? That was sort of like the intro to it. Give us the moment. What was the like, I'm gonna be on this for a little bit.

0:31:27What made Midjourney click for Finn?

Play from 0:31:41 8:88:88 0:31:41 I uh I used your You you put out like a PDF or something like that, I don't remember what it was, um, like a free thing of like how to like describe an image and like recreate it and stuff. And I did using this it was like some sepia image of like witches in the woods or something like that. And I I recreated it and then I was like, what if I added one more witch? And I did and it worked in the same style.

Play from 0:32:06 8:88:88 0:32:06 I was like, fuck. Whoa. Like I just added a witch to this in like two seconds. And then I was like, What if there's a horse in the background too? And it worked, and I was like Like, my God. And I think that was Midjourney V6. ti timestamp that? Yeah. So okay. So maybe Midjourney V6 then? V6. V6 was like a December twenty twenty three or something? Yeah. Yeah.

Play from 0:32:32 8:88:88 0:32:32 It was right before. Yeah. Yeah, well I just researched that and I I was wondering that too. And was like, that was the moment where I like, oh shit. Wow. Um I could do a lot with this. Cause you know, you think about with design Creating high quality, you know, this is sort of where you came, I think, Rory, is like the binding constraint for design, like creating high quality images has always been really, really, really, really hard and expensive, you know?

Play from 0:33:03 8:88:88 0:33:03 Whether you're doing like meta ads or editorial or whatever, it's like you're lucky as a designer if you get three decent photos from a client, you know? Like literally, you know? And and one of them might be a fucking seventy-two DPI JPEG, and you're like, All right, well we'll do the best we can. You know? Yeah. You know, and then you're Okay.

Play from 0:33:30 8:88:88 0:33:30 yeah, in your Google images and stock photos and like it all sucks or it's not right or whatever and you know, so like having that experience in Midjourney I was like, Oh, I will I never have to worry again about having high quality images. Like that is it's not perfect, but like I I I I feel like that is more or less a solved problem now. This changes literally everything about the entire creative process.

Play from 0:33:55 8:88:88 0:33:55 When was the first time you deployed some of that, not customer facing, but just public facing? Because I know you did some cool stuff with it. yeah, so for I'm a partner in a company called the URM Academy that does like on online education for music producers. A lot of people know our product called Nail the Mix, which um basically we teach like metal producers how to how to make music. And so uh I I had wanted to do some SEO stuff for us for a long time, but uh paying humans to write you know, it's a it's a pretty niche.

0:34:00How did AI produce 10x SEO traffic?

Play from 0:34:28 8:88:88 0:34:28 It's pretty niche topic. So paying humans, you know, hundreds of dollars to write each one of these articles was not feasible. And creating like thumbnails and stuff for it, you know, by hand, also not really feasible. And this was like right a w this is when Gemini 2.5 came out, which I think was the first model to be like good at writing. and so I set up a basic to make a long story short, set up a an automation that took all all of the s the scripts from our YouTube videos, which we had hundreds of, and turned those into blog posts and I used Midjourney to make images to make header images that looked cool because like in music, the you know, the standard for like visuals and music is super, super, super high.

Play from 0:35:14 8:88:88 0:35:14 So we couldn't put out stock bullshit, you know, it would just like be a credibility killer, you know. So I used Midger to make stuff, make images that were really cool. And I used Gemini with at the time I was using Make, uh, because you know, this is before Clock Code and stuff existed, to automate a few hundred articles that were like very, like very good. I showed them to all my partners, like, dude, like hey I wrote this.

Play from 0:35:41 8:88:88 0:35:41 This is good, right? And they're like, yeah, this is like really good. And the images looked cool. I shipped all that and 10 extra traffic in like three months, and I was like, Whoa, that's wild. Probably wasn't what anyone was expecting, really. mean, like, you know, if you don't, if you try to do that without any sort of skill in SEO, by the way, or have any sort of experience of that, never were worked.

Play from 0:36:08 8:88:88 0:36:08 But, you know, if you have any sort of marketing brain or any sort of understanding of how this stuff works, it can work because you've given it rules. It's like. And by the way, we have not lost that traffic. A lot of SEO, a lot of the like programmatic SEO, whatever, like you see that shit fall off a cliff, which I I did some other programmatic SEO stuff on another one of our sites. That did fall off a cliff because I basically was like, what if I just do this like the cheap, dirty way? Got nuked.

Play from 0:36:33 8:88:88 0:36:33 But this other site where I did it sort of, well, I'll call it the hard way, still with AI, but like a lot more effort. I only did like 300 articles versus the other one. I did like 6,000. The site with three articles where I made, you know, those images and r uh you know, sort of took my time with the I content, that traffic is still there. It's interesting now because it's like you have to build uh the websites are almost like a a two part thing now.

Play from 0:37:00 8:88:88 0:37:00 You have to build them for humans, which we've always done, but you also need to be building them for machines now. I mean, more than ever. I mean, I think the early SEO days, you were kind of building it in that way, right? Schema and you know, you had sort of these strategies, but now it's it it seems imperative, you know, like you you almost have to have things working on both ends because a lot of times like You're not gonna have the humans going to the website anymore.

Play from 0:37:25 8:88:88 0:37:25 You know, you're gonna have L L Ms and agents going to the website, you know, I mean I had that exact realization around this time I was like, this is like websites now are basically just a data source for, you know, agents. Or that's what they're going to be. So let's build it with that in mind. Yeah, it's a different intent. And the cool imagery thing, by the way, I'm doing a master of segues over here because, know, Drew and I, Drew and I talked about this last week, unprompted.

0:37:44Why reject polished AI perfection?

Play from 0:37:58 8:88:88 0:37:58 were just talking about how like there's something there's something that's going on. I feel it right now a lot is like I'm being pulled towards like a nostalgic style because I'm so tired of the AI perfection and seeing it everywhere. So I'm like going back. I want. I want pixelation. want distortion. want glow. want shifting. I want spectral. I want all this stuff again because I'm just so tired of yes. Yes. And I think you you've gone particularly deep on this stuff and especially some of the other thing.

Play from 0:38:28 8:88:88 0:38:28 If anyone here is a fan of like doing their own sketches to to image and like playing with that, this is a really fun stuff. Some of the stuff you've done with fonts, RPG style things like you bring that nostalgia back. I feel it immediately when I see your stuff. people use that word nostalgia uh on a lot of the stuff and it's it's not it's what it it I don't agree that it's nostalgia. It's what you said is like I just want it to not look like AI.

0:38:41Why does handmade design feel better?

Play from 0:38:55 8:88:88 0:38:55 I want it to look hand done because that is interesting. I could make I mean I've done lots of super clean stuff. That's not a problem. But I think it's more interesting now for things to look hand done and organic. So that's sort of what I'm focusing on is like, how do I make content that that looks and feels organic but is still with AI? So it it's not like it's not nostalgia per se.

Play from 0:39:21 8:88:88 0:39:21 It's more just like wanting the kind of or subtle like the the the reason Why hand-done well let me back up? So the hu this is going back to like cognitive psychology, is like The human brain is trained to recognize patterns. And we are we are highly like we recognize disruptions to patterns. There's a thing called pre-attentive processing, for example. Like if you look at a bunch of bushes, you tune it all out, but there's one part where the the branches are disturbed, your eyes gonna instantly go to that because it disturbs the pattern, because that might be where the tiger lives, right?

Play from 0:39:55 8:88:88 0:39:55 So the reason why and and how that applies to design is that the reason why uh organic anything is interesting is because it has that subtle variation that disrupts patterns. You know, instead of a perfect grid of images, a perfect grid of something, it's a little bit wobbly. You know, and one of them is a little squished or whatever. And yes.

Play from 0:40:20 8:88:88 0:40:20 It's got more personality. And and you know, the other we've we've talked about this is like one of those common threads that always shows up. But when you step away from the realism aspect too, you have more liberties, creativity, or liber uh creativity. Create creatively? Jesus. This wouldn't be a guest podcast without me stumbling over something. But right. I think the value of those is basically zero. You know? We've talked about that was like a a key topic we were talking about just last week, right?

Play from 0:40:48 8:88:88 0:40:48 Like it's just the step away from realism because there are just so many there's you know, almost infinite styles that could exist out there. And then when you put your own unique spin onto those, right, they become standalones. And it's just it it feels like you're playing more with raw materials, right? And there's look let's look at some of these. I'll give you an example. Sorry, I'm gonna cut you off, but this is a great example of like these are I mean, you could say these are AI, but I don't really consider them to be AI because there's probably four or five different assets in each of these, maybe more.

0:41:06How should AI and manual craft combine?

Play from 0:41:27 8:88:88 0:41:27 So there's like the frame um in all of these. I made those with Nano Banana in the style of like stone and then used uh a plug in called um it's the Doran Studios DitherTone Pro, I think, and plug it in Photoshop to like dither them and recolor them. And then the borders with like the barbed wire and the skull and stuff, that's another thing. The monster in the middle is another thing, and a lot of those I recolored.

Play from 0:41:53 8:88:88 0:41:53 Like the one where the it's the guy pointing with the wings, like I probably spent two hours with my fucking Wacom tablet like recoloring that whole thing because it came out of Midjourney as like you know just line art and I was like well I would like more color in this so I it's like it's like it doesn't occur to people like you can just color it like you don't have to prompt everything like hey remember people you can like just draw shit That's what we were I mean, we were talking about that.

Play from 0:42:23 8:88:88 0:42:23 That was like a key point where it was just like we know now that there are so many things that have been unlocked with with AI and have hijacked the creative process. You know, and and we've all kind of gone down this pathway of seeing how far we can push it, right? What can AI do? Rory Rory, I mean, this is your thing, right? I mean, Rory's always edging on those things. But it's just like then you get to this point too where it's like, we're slowing down now.

Play from 0:42:51 8:88:88 0:42:51 Yeah. Because we want to provide our own personal touches in that process, the parts that we like the most, right? We I mean, yeah, we know AI could do it, but if we like that, why not us do it? I mean, yes, it costs more time. Yes, maybe it allows us to do like fewer things, but I feel like we need that internal satisfaction now. Yeah, it does it's not fun to do everything. You know, there's a difference, like obviously when you're doing something and creating a system for a client, you know, fun is not the not the objective.

Play from 0:43:22 8:88:88 0:43:22 But um, you know, when when you're doing your own shit, like I don't think it's fun to do everything with AI. Like it doesn't like it can't like it's it's interesting in its own way, but like it's more fun when I feel like, you know, when I had my hands on it, I enjoy it more. And you started a lot of stuff, you know, outside of this, we're looking at one specific sort of thing that you've been doing recently, but as you know, there's been some some fonts and some things that you've sketched and then transformed.

Play from 0:43:51 8:88:88 0:43:51 Yeah. up the letters there. Like there's zero AI in that. I drew all of this, you know, because I did graffiti for years before. So like that's all I I have I probably drew that twenty times on paper, you know, before I even opened up Illustrator. So then I opened it up and you know, traced it in Illustrator. Drew some textures in in Photoshop and then I used either Dither Tone or Dither Boy, I don't remember which, to to make all these and turn it into a GIF.

Play from 0:44:17 8:88:88 0:44:17 So there's like zero AI in that. And that's specifically because I was like, I just wanna like draw some letters again. It's like I haven't done that in a long time. It's like I just wanted to do it. And yeah. using AI like it untaps that again. Like I've noticed that too. Yeah. I wanna draw letters again and I haven't felt that urge in years and now I wanna do it all the time.

Play from 0:44:45 8:88:88 0:44:45 It's am gonna be fascinated to see when the time comes that people posting content that are using AI as a part of the creative process stop explicitly calling out that they're they used an AI tool for it. And it's just we're gonna have to figure those things out. I think there's that's gonna be good in some ways and that's gonna be bad in some ways because I mean, how skilled are you? You know, like how much of that can I envision.

Play from 0:45:11 8:88:88 0:45:11 being attributed to that model or the technology versus your creative thinking or expertise, right? Like that also could be a very scary period of time when that starts to just go without saying. You know, people don't even say that anymore, right? It's just like here, here's this. Was AI used or We don't know. We got to figure it yeah, I I don't usually say it unless I wanna like teach some unless I wanna explain it to people. Like there's probably people that think those letters are AI.

Play from 0:45:40 8:88:88 0:45:40 I saw a couple comments about that. I'm like literally not a single pixel of AI in this. And it's the other thing, right? It's like with this stuff, I've found myself, but because I can now, and it's like all these ideas I had in my head, like I want to add like, you know, this level of like pixelation distortion and scan lines and all this other stuff that like, this glow, but like, what would take me hours of playing with the blending and the layers and like the double stacking and like, how do I feather this?

Play from 0:46:07 8:88:88 0:46:07 You know, just like would make me not do it because I, as you know, yeah. adult, you end up not having time to just sit there for seven hours on a design, right? But I go to Figma now and create a shader for exactly what I want to do and then have it at my fingertips. And now I want to design more. Right. Like I think it's I think the net positive in that sense. Right. Like. And and maybe you might explore it a little bit in Figma and and then maybe you do decide, you know what, I I do want to take seven hours to do this once.

0:46:24Why does making more have zero value?

Play from 0:46:36 8:88:88 0:46:36 You know? 'Cause but but it it's it it's not either or like you know, I I think what we'll realize well there I'll tell you one thing I've definitely realized is that doing more is basically has zero value. Like Okay, I all of us on this call right now could probably make a thousand images today that were pretty good. They're that are like very good.

Play from 0:47:02 8:88:88 0:47:02 We could i if if you gave us the rest of the day, we could each make a thousand images that are really damn good. So what? Yeah, exactly. That's it. That's it. And that's what everyone fails to imagine. Like, this is what's getting me upset with a lot of the content, too, that I'm seeing. It's just like, like, no one is evolving past the point of uh fast and cheap, right?

Play from 0:47:27 8:88:88 0:47:27 Like, it just seems like everyone's, I did this in 15 minutes. I did this. This would have cost a billion dollars. um know, and everything I see is like, can you believe this was made with AI? And it's like, it's. I don't know if it's that or if it's laziness, you know, at this point, like, because are people really still? Yeah. Yeah, yeah. they wanna brag about how long it took to do something and how they, you know, were able to do this with little to no barrier to entry and then kind of like boast about it and at the same time they're too lazy to go about it the longer way.

Play from 0:47:58 8:88:88 0:47:58 Yeah, it's both things. I think both things are true. But it do and I've seen a lot of people that you know, so like I do a lot, you know, for our our company, I you know, our product team is like me and an engineer, so I'm I'm like pretty deep in code stuff too. I've seen a lot of people who have legitimately shipped ten times more features than they did before. Like it really if you're good with AI, you can definitely be ten times more productive in the sense of shipping more.

Play from 0:48:24 8:88:88 0:48:24 They're not growing 10x. You know, they're not. Like they're not. a lot of stuff too. like y you shipped ten times more things, so what? You didn't ship you didn't ship the right thing. That's all that matters. You know, same and the same with like images. You know, you ship it the idea, like having a clever idea Yeah, right.

Play from 0:48:50 8:88:88 0:48:50 Exactly. And half the number and more attention into the depth of it, right? And asking the right questions. yes, one of the right images is worth more than literally a thousand that are even we'll say very good, not even bad, like very good. The difference between like very good and like great is everything. You know, that like one percent, five percent difference is everything.

Play from 0:49:17 8:88:88 0:49:17 I want to take a step back on this one all the way back to the Abercrombie and Hollister days when you're when you're designing t-shirts that are going to possibly be purchased by millions of people, right? Like sure my designs have sold at least five million units. I I don't know how to track it exactly, but at least that. here has probably owned something or purchased something from those stores during that time. Whoever's watching this. like, you know, how, how much scrutiny does one of these designs go through before it gets like, some of these.

Play from 0:49:46 8:88:88 0:49:46 I'd have to look it defin yeah, the the yeah, I recognize some of these. The pink one that says Hollister SoCal with like the green eagle there, I did that one. This is so insane though. long sleeve, it's hard to say, 'cause we did a lot like that. But how much scrutiny do they get? Um, infinity. Uh you know, in both good ways and bad ways. Yeah, I I did a good amount of these.

Play from 0:50:15 8:88:88 0:50:15 in both good and bad ways, you know, like it's like anything else, like people will like fixate on the dumbest shit, you know, sometimes. But then in other ways, you know, they'll hold your feet to the fire in a good way. of like this embroidery isn't it could be softer. You know, and you're like fuck I just wanna like I just wanna be fucking done with this, but you're right. It should be softer. you just reminded me of one of the things that you posted.

Play from 0:50:41 8:88:88 0:50:41 I think is a great also a great thing for anyone that's creating images as well. You did a great post on text sort of like I think it was sort of like texture relationships from like a screen printing or like embroidery and things like that. learning about like uh about materials, yeah. Yes, yes, yes. the actual mechanical process or the physical process of that. example, so I did printing when I was a kid.

0:50:57How does material knowledge improve prompts?

Play from 0:51:08 8:88:88 0:51:08 Like, well here, go back to the like that pink one there. Like, yeah, click on that. Um, I drew all those cracks in. There's not that's not a prompt. That's not a fucking texture I downloaded. I just did the pencil tool in Photoshop and I drew them. You know, like I learned that at Abercrombie. That's one of the to your point about the scrutiny. I was like, like how how do you do those and make them like all line up in exactly the right way? And he's like, I just draw them.

Play from 0:51:33 8:88:88 0:51:33 And I was like, oh, yeah. uh There's no special button for that. Yeah. you just draw them where you want them to go. Okay, yeah. Um, but that's an example. So like what what I was talking about with like materials is um so I did printing when I was a kid from when I was like fifteen to twenty two or whatever, before I could get a design job. So I learned a lot about that.

Play from 0:51:59 8:88:88 0:51:59 And you know, if you were to prompt like vintage poster, who knows what you get? You could rerun that ten times, you get ten different results. But if you were to say like vintage s you know, a screen printed poster um printed with, you know, water based ink on um uncoated paper, sun bleached, you know, uh letterpress, whatever, you will get a different result.

Play from 0:52:26 8:88:88 0:52:26 Like th those things all have semantic meaning, you know, in the models, and you will get a specific result from that. So the more you understand, you know, what the specific words are, the more you'll be able to control the model. So I think, you know, if you have that knowledge already, use it. If you don't have it, it I would suggest getting it. You know, like go take a screen printing class.

Play from 0:52:53 8:88:88 0:52:53 you know, or read a book about it or whatever, like learn all that shit. Same with photography. Like Rory, you know, this is one of the things that you did early on is like you had like the Midjourney prompting guide with like that massive block of like um like uh film terms, you know, like angles and camera moves. Yeah, like that stuff. Yeah. Yeah, but they still they do to some extent, you know. Like the the point is that those words have actual meaning.

Play from 0:53:22 8:88:88 0:53:22 And they do change the output. And the more you understand the detail like that. Here go back up to the orange. The orange one there. Yeah, like that. Yeah. Like this would be a good example of it. Like I knew exactly what to ask for. Um uh because, you know, I I spent thousands of hours doing those graphics in Abercrombie, like distressed twill applique. Like twill is a specific type of fabric. I know exactly what that texture is.

Play from 0:53:48 8:88:88 0:53:48 The other one, those like the orange one, that's called chenille. Like that's chenille with a felt layer underneath. Like I knew what to ask for there and it I I was kind of surprised how well it came out. But I was like only getting fifty to seventy five percent of that if you don't have this the SME knowledge. You know what mean? Like that's that that's that extra layer. Yeah. layer of felt and and chenille, that looks perfect.

Play from 0:54:13 8:88:88 0:54:13 But see, that's oftentimes like the differentiator between something that looks like slop and something that looks professional. It's like that level of like chenille and felt, right? Like instead of just, you know, embossed, not embossed, whatever, embroidered, you know, jersey font on a, Like I said. But it's cool, man, because I'm definitely interested also, because we want to check out some of this, this Weavy stuff.

Play from 0:54:39 8:88:88 0:54:39 that you've been doing because there's been, you like said, a lot of sketch to image, a lot of, you know, systems thinking here that you've then taken and then we'll push it past towards the end of this episode. I to definitely check some more of the deterministic tools that you're creating also. like, so, you know, now knowing a little bit more on the system side, it feels like when something like Weavy or, you know, any of other tools that are node-based comes along, that's right up your, you know, right in your wheelhouse.

0:55:03Why choose Weavy over ComfyUI?

Play from 0:55:05 8:88:88 0:55:05 I and I had tried using ComfyUI before and I love ComfyUI. I think it's awesome. Like I love what they're doing. But it's just it's it's overkill for the majority of things. I don't want it like I'm pretty technical, so can I figure out how to get the right VAE node for this CLIP encoder blah blah like can I figure that out? Yes, but I don't want to spend an hour and a half on that. I want to spend that time, you know, on the creative part. So like I kind of bounced off of ComfyUI.

Play from 0:55:32 8:88:88 0:55:32 Again, I love it and respect, but it's just it it's there's too much friction for me in most cases. Unless you use the partner nodes, in which case that's basically the same thing as using Weavy or FLORA or whatever. Um So yeah, I mean that that to your point, this is exactly I just felt like totally in my zone of like you know creating these systems with a bunch of levers you can pull to adjust it and and get the output you want.

0:55:44How do sketches improve AI images?

Play from 0:55:58 8:88:88 0:55:58 And uh like with the sketch thing, I've I've been very interested in using that using like sketches that I make as an input in the image models because There's um well this is one of the things I learned about from like basically reading some of the research papers and stuff. Like there's an idea of like out of d well here's a good example. Yeah, there's a good one right there. Like I made this on my phone, like I was just drawing this while my kid was you know, eating a snack.

Play from 0:56:29 8:88:88 0:56:29 It was like, I'm gonna make like a cool what if Paris Hilton? I just thought had this idea in my head like what if Paris Hilton is like a fantasy knight? Like what if she had this cool like I had like Y2K shit. Yeah, what would her sword look like? drawing by the way. I mean, Jesus, not this ain't no, you know, fifth grade drawing here. I mean, Jesus Christ. Well, I mean I almost symmetrical too. I'll give you credit on that one. That's a tough one. mean, some of that stuff is pretty much, if you fold that over, that's pretty close.

Play from 0:56:56 8:88:88 0:56:56 It looks so mean, whatever. I do to me that's like a doodle, but like it doesn't have to be It doesn't have to be amazing, you know? And and I j and I did this on my phone, which I usually hate doing anything on my phone, but my kid was like jump literally jumping on me. So you you know how that goes, Rory. So uh I was like I just had this idea of like, what if Parasilt and had this cool sword? I'm gonna just do instead of trying again, like you don't have to prompt everything.

Play from 0:57:22 8:88:88 0:57:22 That's like the big thing I want everyone to take away is like a prompting is probably like five percent of my attention. I d prompt like I I don't I don't care about prompting for the most part. like if you have an idea, just draw it. It doesn't have to be a great drawing. And it came out awesome. Like that's the first one. Yeah. references are great. Doesn't they could come in many forms. They could come in a drawing.

Play from 0:57:47 8:88:88 0:57:47 Maybe you've got like a raw material somewhere that you find on a third party site and you just want to use the shape of an object, right? Like these are the that's the way you got to be thinking versus, oh, how do I prompt this from scratch? How and one, um, the one we're looking at here. Uh I like I I I I started out like I again so the idea here is I I went from like a sketch to like you know these cool like chrome whatever kind of effects.

Play from 0:58:15 8:88:88 0:58:15 I could do all that stuff in Photoshop, like you're saying, Rory, but like it would take me a long time and it's just realistically just not gonna happen. So but I I like the part I care about so and getting back to what I was saying before, it's like where do you choose to invest your time and your skill? It's like okay, well I'm good at drawing letters because I've done it a lot. I can do that better than any model can. So I want I'm gonna do the I'm gonna put my time into making the letters exactly what I want, and then I'm gonna let it put the Chrome on top, like literally.

0:58:33How should creators divide work with AI?

Play from 0:58:45 8:88:88 0:58:45 So in this case, I drew the letters, I traced them in FigMun or Illustrator, whatever I used, um, and then used that as an input to you know do style transfer. You know, just took a bunch of different things from Pinterest and mix and match. So it takes my letters but puts these other effects on. top of them, see what I like, I might end up redoing it, you know, like but at least it it enables me to like try out twenty different styles in half an hour instead of spending seven hours on each one of them only to find out I don't like it.

Play from 0:59:20 8:88:88 0:59:20 That is see like that's a that's a way of thinking I think that should resonate with probably just about anyone here right as if you have that initial idea and want to get to 20 because you because you want to do the 20 and I want to just keep going from there it's like you don't have to sacrifice the first stage to use AI for the middle stages right like that drawing is still part of it yeah too, because you it was it was like you wanted to focus on the part that you wanted to build the muscle or or continue to work out, right?

Play from 0:59:48 8:88:88 0:59:48 You didn't want it to atrophy, you wanted you got some excitement or creative spark from that. You didn't waste the time on the chrome, which maybe you're thinking about the return on that. How long is it going to take me to learn an a a certain threshold of a level of this? And is that going to be something I ever use again. Right? Like so it's like I already know a lot of that. It's just it takes time to do it, you know?

Play from 1:00:14 8:88:88 1:00:14 But that you know, you made the judgment of what part you were gonna use, what part you were gonna just discern from and and move away from. measuring gradients and reflections and stuff like that on the chrome stuff. I remember trying to do chrome on the gradients and just being like, my God, is like, it's an unreasonably long amount of time to get it to math correctly, if that makes sense. uh is a good example. The first three are all me.

Play from 1:00:41 8:88:88 1:00:41 Like that sketch. This is just the straight outline. And then this is the outline plus a shadow. And then the next one is, you know, a chrome effect from GPT Image, I think. Could I do that myself? Sure, but it would take me a long time. Go ahead. so like that's that's like that's the one thing that I also keep coming back to also with these digital the digital sort of like 2d vector font that I'm getting so What's the maybe boards not the right word?

Play from 1:01:11 8:88:88 1:01:11 Maybe it is. I'm just I don't know if you look at like typography I mean anyone who's in design knows typography just like it's your whole life It's like I'm just so tired of this 2d realm. It feels weird to me. I don't know how else to explain it But it feels like everything's so flat and digital and like that's it. Like there's someone with that hand drawn sketch. You just, it looks better. It feels better. Yeah. to the um to the little bit here.

Play from 1:01:38 8:88:88 1:01:38 Sc scroll up to there's like a black and white letters. Um Scroll up a little bit more. Yeah, there it is. That that. So this is this is a file that we had at Abercrombie called vectorstuff.ai that got passed around. I don't know who made it, but some but probably a bunch of different people. These are like a bunch of scan. You can scroll through here and see if there's some of the other ones. This is like scans of old like vintage samples that we would buy at, you know, flea markets or whatever, and vectorize them and make fonts out of them.

Play from 1:02:08 8:88:88 1:02:08 And you know, like you can see there's like, you know, Rory, I think you you replied. To the something about like how much time I've spent trying to find the perfect jersey font. Well, so did we. That's why there's like 30 of them in this file because we found all of them. And this idea, what I'm very interested in with AI, is like there's a there's a an idea called out-of-distribution, meaning that any given model, there's a distribution of potential outcomes that it that it can do, and it cannot do anything that is out of that distribution.

1:02:12What does out-of-distribution mean?

Play from 1:02:40 8:88:88 1:02:40 So, like if the model has never seen a semicolon, it cannot. Create a semicolon, right? It is impossible. Um, and like with images and video, you see sometimes it's really good at doing certain things, and it it's there's also a gradient here. Something can be technically in distribution, but it's at the edge of the distribution, so you can get there, but it's hard. You know, like we all you've all felt that when you're fighting the model, you're like trying to get it to do something like fuck, I have to do this like 12 fucking times before it would just make this part green, you know.

Play from 1:03:09 8:88:88 1:03:09 That's when you're feeling like you're at the edge of the distribution. And so what I think where the value is is in doing things that are either out of distribution completely or at the edge of the distribution. All that stuff in the middle of the distribution is worthless. That's all like the you know, people are making fun of like the small business, you know, GPT flyers, you know. That's the middle of this and and I don't fault those people for doing it. They're small business, like I'm not I'm not yeah, I'm not dunking on them, but as a designer, like that's not acceptable, you know.

Play from 1:03:39 8:88:88 1:03:39 So th what I'm interested in with with stuff like you this vector file is like, how do you do stuff? Like these fonts are like scanned from literally like a high school football jersey from 1968. It does not exist in any image model because there's one of them, you know, and we bought it at a flea market, you know? And so that kind of thing that's like out-of-distribution is is what's valuable now, you know, because like different when it comes to creative stuff is yes, yes.

Play from 1:04:11 8:88:88 1:04:11 articulated that way. That's that's nice. in machine learning called out-of-distribution that you can look up. And so this is like these are the light bulb moments that I have. This is the reason why like I read research papers and like read about the computer science stuff, is because I had that light bulb moment of like, like out-of-distribution is valuable because AI literally cannot do it. It's literally impossible. You know, or extremely, extremely hard. And and also like when you use and another another thing with like image references or like sketch references is it pushes it to the edge of the distribution.

Play from 1:04:44 8:88:88 1:04:44 If everyone else is prompting or using whatever Google Images for as like they're gonna be um the their outputs are more likely to be towards the center of that distribution. But if you draw something by hand that is a unique set of inputs that does not exist anywhere else in the world because you just made it, it's gonna steer like all of this is like s the the way to think about like AI in this regard is attention steering is like the term for.

Play from 1:05:12 8:88:88 1:05:12 You are steering, imagine you're steering a ship. Like imagine the potential outcomes are this big 3D cube. It's not exactly that, but imagine as that. And you are steering through it. That's what you're doing. And so um prompting with text is sort of the most basic way of steering it and the least interesting. But when you were doing things like giving it a hand-drawn sketch, that's like putting a crazy person at the wheel and like swerving in a totally different direction than everyone else, where you're gonna end up going down some road.

Play from 1:05:42 8:88:88 1:05:42 That no one else has been down in in that latent space of possibilities. We're like, there's this down at the end of this little road that nobody ever looked at before. That's cool. You know? Yeah, its predictability starts to to shake and tremble. Exactly, exactly. That's what you're looking for is like how do you give it different inputs so that it steers in an unpredictable different direction? And that's where things like go take like with Midjourney, like Go take a picture of something on your street and use that as an image reference or a style reference or whatever.

1:05:56How do original references beat AI slop?

Play from 1:06:18 8:88:88 1:06:18 Like cause that is a one of one reference that does not exist in any training set because you just made it. So like anytime you can do like a one of one and and like steer, you know, that attention vector in an unexpected direction that's different from what other people are doing, that's more interesting and valuable. That's a good point. I mean, there was one thing I remember, we did it on Midjourney a while back.

Play from 1:06:43 8:88:88 1:06:43 I think we probably did it on one of the episodes. was like, people couldn't get like that raw iPhone realism, so we just made a mood board of our iPhone photos from our phone. And it was like, well, now it looks like raw iPhone realism. Same thing with everyone that's just like, you're trying to make, know, everybody's trying to make UGC with AI, which I think is like the dumbest thing of all time. It's like... The reason people like UGC is because it's real. I know. gonna be so it's a race to the Also like s b lots of potential FTC violations there that people should be careful of, but that's another conversation.

1:07:03Why does fake AI UGC destroy trust?

Play from 1:07:17 8:88:88 1:07:17 Exactly. I'm just like, why? I don't understand that the whole race to get to the to the UGC. Like if I found out that a brand was doing fake UGC, like I just immediately ignore them now because I'm just like, it's not the point of UGC. Yeah. in unless you're some dropshipper fly by night brand that's not gonna exist in three months, um, in which case like I don't want to work with anybody like that, but okay, use the growth hacky crap for that, whatever.

Play from 1:07:42 8:88:88 1:07:42 But if you're any real brand, like that should be just absolute cancer. Hell no. Yeah. That it's just like I mean, Jesus. Right. Like shots. You want to vary some e-comm assets so you have a front view and a side view? Like, I don't care. But if you have a good... Yeah. Yeah.

Play from 1:08:07 8:88:88 1:08:07 Yes. When you see a face and hear a voice, that shit still needs to be real, you know? Don't no like my my rule of thumb for that stuff is like I never want people to feel like they were deceived. I never want the viewer to feel like they're I'm trying to trick them. Which is another reason why I don't like realism all that much. Is like, you know, like the the sort of pixel art kind of stuff I've been doing.

Play from 1:08:33 8:88:88 1:08:33 Like Nobody like there nobody's gonna feel fooled by that because like it's obviously you know, it's obviously what it is. Um and I think like there's people like the the thing for everyone to remember is that consumers like audiences have very long memories for this stuff. We can all think about stuff that a brand has done or said once years ago that people bring up for like decades sometimes.

Play from 1:09:04 8:88:88 1:09:04 Yeah. Yeah, yeah. i i like they have long memories. If they feel like they got burned once, you're never getting them back. Maybe maybe think about this too. I'm also like this is an interesting avenue of this. It's almost like how long the exposure is to the photorealism, also. What's the placement, right? Because we were talking about the AI UGC. Well, those are usually 15 to 30 second little things. That's a long time.

Play from 1:09:29 8:88:88 1:09:29 Versus if you use that, that is maybe in a b-roll of a video where you've where the images or the stills were AI. It's not. you know, where someone's just scrolling past it. Yeah. it fill it's a filler, right? You need maybe that content for there. Um, you know, so it's like almost the placement, the exposure in terms of the surface, right, that it's on. they feel deceived if someone's foot in the background was AI?

Play from 1:09:58 8:88:88 1:09:58 they don't probably don't care. Yeah. But if it's the person speaking to you in the ad where implicitly it's you know, you're asking you're you're trying to persuade yeah, right, for tr exactly. Now they're gonna feel burned. I just it's it's just i it's such a foolish like you know, like you were saying earlier, this race to the bottom of how can I do it faster and cheaper? Okay, but at what larger cost?

Play from 1:10:23 8:88:88 1:10:23 Like yeah, I can make a thousand shitty fake UGC ads for you tomorrow, but And and and maybe you'll get some you know, that maybe they'll drive some s conversions in the short term, but it might also burn down your whole brand and destroy it for for decades. You know, like think about like Nike, people still bring up the sweatshop stuff, which that was in like nineteen ninety seven. You know, and I'm not saying that was like okay or it it's not, but that was twenty five years ago or whatever, thirty years ago almost, and people still bring it up all the time.

Play from 1:10:54 8:88:88 1:10:54 Like the people we're invested, we spend our money, right? So you feel like a little bit of attachment or you feel like I'm putting some level of trust by exchanging a dollar, right? And then it's like, then you feel like you've bought into something and if that something is rotten, you're like, speaking, you know, I'm sure that's how everyone feels with Higgs field right now. seeing you right there. uh like I I I I when I saw that they had so many things that appeared to be, you know, uh FTC like violations of like uh undisclosed influencer marketing, uh definitely soured my opinion of the company.

1:11:31What went wrong with Higgsfield?

Play from 1:11:31 8:88:88 1:11:31 Yeah. they've hit a couple they've had their milestone moments for sure. Like a couple of those bumps along the route. And then our our buddy, our buddy uh theoretical media just posted last night and he just drops the terms and conditions privacy policy for them, which I think to be fair, after like I did some more digging in here because I was like, dude, this is hell. But I you know, I think a lot of these things actually are more common, not just Higgs field, but still I think the when you couple that with Higgsfield's already damaged reputation here and lack of trust.

Play from 1:12:04 8:88:88 1:12:04 This just adds gasoline here. I mean, so this was posted last night. If you haven't seen this, it just it it dives into, I think, s five or six areas specifically that are very concerning if you're a, you know, a Higgsfield user. Um, and so if you've spent time generating assets there, there's a lot that I would highly recommend reading through this. But, you know, they're talking about sort of I I think let's just let's just maybe talk talk talk about number one here, which is just like you grant Higgsfield a perpetual, irrevocable, worldwide, royalty-free, transferable, sub-licensable through multiple tiers license to everything you upload and everything you generate.

Play from 1:12:47 8:88:88 1:12:47 I mean, wow. is what it's like with everything but still seriously like if you're a business right and you have if you're trying to like if you're an enterprise using Hakesfield god bless like I don't know what you're thinking but that's what I'm saying you can't can't. Like no attorney yeah, you can't any any corporate attorney that looks at that is gonna be like d uh cancel this account immediately.

Play from 1:13:13 8:88:88 1:13:13 and this is where we're this is a big problem too, because like when you get into commercially accepted things, what do you have? You got Firefly approved models most of the time. I mean, the bigger companies I've worked with, that's the go-to. And we we already know, sort of in terms of like, you know, quality tier where that falls. Um, and I think that's That's a problem that exists right now.

Play from 1:13:39 8:88:88 1:13:39 Also, you know, it's funny with these terms and conditions. I saw like a documentary a long time ago, which basically was this idea that they create these terms and conditions and privacy policies to be as boring as possible on purpose so no one read them. This was like something that, you know, was manufactured decades ago. But they specifically talked about the thought that went into that. That's why they always put them in the typography of this boring typography.

Play from 1:14:05 8:88:88 1:14:05 There was never any. sort of markdown formatting, you know, is these long files that right before AI. No one is spending the time to ever go through those. Ever. You know, like especially at the uh friction point of signing up for that tool or when you're trying to use it. Because when you think about going into any tool, the last thing you want to do, because you're activated to use it, is go and read through 30 pages. Right. And it's just uh like it's just so conniving.

Play from 1:14:31 8:88:88 1:14:31 And then You know, you got this breakdown, which again, maybe we maybe we link this or maybe you guys go find this, but this this post is taking off as we speak. And um Higgsfield's got something something to deal with here. Credit to Tim. This is the this is the most subtle Beautiful blast post without like being you know rage baity. I have seen a long time. It's just like Yeah Before you bro me And then he said something else, which is like a perfect sign up.

Play from 1:15:00 8:88:88 1:15:00 He's like, I am not he's like, I'm not your dad. I'm not a lawyer. You know, make up your own mind. Actually, did you read the whole thing? You know? Yeah, that's it was like so well, like he just so predictively analyzed what everyone was going to react to and how they're going to react to it. Just great. Tim's a great guy. more, it it kind of does like like you said, it it seems like these T&Cs probably aren't necessarily super different than what you'd see in a lot of other places.

Play from 1:15:27 8:88:88 1:15:27 But like you said, Drew, because they've burned people's trust so many times before, you know, you come into it um sort of looking for the worst possible interpretation of it. And then they're and they're buying it with money. They're still doing the influencer stuff, right? They're going hard. I mean, I I see several people on my feed that are still promoting them. And I I gotta be honest, right? Like you at some point you have to question who you're taking the money from.

Play from 1:15:54 8:88:88 1:15:54 I know some people's circumstances are different, but it's just I think you gotta ask yourself: when do I use the tool? What are people gonna think if I promote this product? Do I want to attach my equity to this equity? Right. And then I just think that. Ladder half is has shipped. You I I don't think you can go back on that. I mean, let's not remember what the other you know, hiccups along the way were, Rory. Remember the like dude, they had that whole campaign of like we're sacrificing our salaries this month based on, you know, we're sorry and blah, blah, blah.

Play from 1:16:25 8:88:88 1:16:25 You know, dude, it's just so one that they did that really was the first that really kind of made me wonder what's going on is there was something along the lines of like, you know, we're destroying every creative job or whatever. Do you remember that one? And I'm just like, what are you guys doing? Like what? naming their Chrome plugin, Steel. then just like their whole marketing, we've said this, I've probably said this in Dossier, their whole marketing strategy is RageBait.

Play from 1:16:50 8:88:88 1:16:50 And it worked for a long time to get users. And then they switched to just give them free stuff. And then there's been this whole thing that, know, one of the things that I think it was maybe Blacklight uncovered here and posted on X a while back was that they made, and again, I don't know how, what the validity of this is. I'm just telling you what I read. was that when they offered the full, the free month of Seedance that it was a distilled model from Seedance and it was, you know, coming out at like 480p and a lower like Seedance light model.

Play from 1:17:19 8:88:88 1:17:19 So yeah, you were getting it for free, but you weren't getting what they promised in the ad, which was the full Seedance experience. So like they've done that. offered Nano Banana free for a month. Yeah. Unlimited. And then you had to pay extra because they make, they'd make the generations like, you know, it'd take like two hours to generate an image. It was like, but if you want to, if you want to pay for the. the speed then you can obviously put your credit card in. It's like that kind of stuff. ah Yes.

Play from 1:17:45 8:88:88 1:17:45 Their cancel flow is uh if I was their attorney, I would be very nervous about it. Let's put it that way. it's not easy to cancel. Um I I also like I had an annual subscription. I had a Black Friday like annual one, and I upgraded for like one month because I needed more use, and then I canceled it and just canceled the whole thing. And even though I'd paid for a year and there was no support link that I could find, I'm like I I I would not be surprised if there's a class action lawsuit in their future.

Play from 1:18:22 8:88:88 1:18:22 Ooh. Here we go. We got a post from them. okay. I can't wait to hear this corporate speak. We are updating our T terms of use to more clearly reflect our commitments to creators and ensure they feel confident, protected, in control of their work. We'd like to be clear on what is true and will stay true. Blah, blah, blah. They are free to use it commercially and does not claim ownership of our use of user outputs, but it's in the deal.

Play from 1:18:48 8:88:88 1:18:48 So are you changing that? Our latest update added terms for new products, including MCP and Super, we will revise legacy language introduced last year to make clear that your private content stays private. Licensed user climate will be limited to what's necessary, but they're not that that's not taking it away. You're just saying we promise we're not gonna use it against you. Okay, then put it in the put it in the contract.

Play from 1:19:13 8:88:88 1:19:13 There's there's some I bet I bet if we went and looked through all of them and did like compare and contrast we could find like some really outrageous stuff, but like when you're like in a you know, uh sometimes this come up with a an agreement with a company or something like, Well, it's just like a boilerplate thing. We're not gonna you know, we'll we'll do that. I'm like, Okay, well why don't you put why why don't you say that then? Yeah. that's yeah, it's gross. It's gross.

Play from 1:19:39 8:88:88 1:19:39 oh okay. So so you do mean it. Tim's on this though. mean, like he's just, he did this in a very, very like, this is going to go, I'm sure he'll do it on his channel and he'll get, you know, lot of good feedback from it. you know, they're just like, I get, they're like a consumer tool. Like, I mean, they try to build themselves as business. I mean, I feel like it's for more small business and people that can take risks like that.

Play from 1:20:06 8:88:88 1:20:06 Like you can't have, you know, you can't be Coca-Cola on that. You can't be, you know, there's no way. Rory's bagel shop, who cares? But yeah yeah. It's unfortunate because they're... I like to praise them every once in a while for the way they think about things. They think about things the right way. They're like, oh, everyone has all these models. They don't know what to do with them. We'll basically tell them what to do and we'll build these little easy systems for you to tap into.

Play from 1:20:35 8:88:88 1:20:35 So it's well packaged. They obviously know how to draw attention better than anyone. personally I I'm maybe I'm a little more harsh, but I think like the UX is not good and like can yeah, conceptually, yes, I I agree that like it's like telling people how to use it is good, but like uh it took me a long time, like before I even understood what the product I was like, oh, this is an API wrapper with a bunch of these like use cases, like and it was like I like literally didn't even understand what Higgsfield was for like months.

Play from 1:21:08 8:88:88 1:21:08 It was a video model. That's where it started and they couldn't compete. Yeah, they couldn't compete so they had to start bringing the aggregators. Yeah. of money right now is if you know how to provide this condensed user experience that is optimal for people using your platform and just breaks it down and simplifies it. Google the naming problems, right?

Play from 1:21:34 8:88:88 1:21:34 Like you just you get so many capabilities, so many products. You just don't like it, it becomes too much to bring somebody in new. It's it's they don't want. to be walking into somewhere and be like, where the fuck am I? What am I gonna supposed to do first? How do I do this? And then like to build the repeatable muscle to get people to show up and figure that out. I mean, like, yeah, I mean that that's that's a problem.

Play from 1:21:59 8:88:88 1:21:59 and I've been using Photoshop for 30 some years and I get a prompt. It's like, here's how to use the select tool. I'm like, okay, thanks. Yeah. Like you know that I've been using I've been a subscriber for like over a decade. You know this about me. Yeah. in a personalization ecosystem. How about that? Welcome to layers.

Play from 1:22:27 8:88:88 1:22:27 This is. What's what is this now from? What is this now for Midjourney? Okay. Yeah. So I was gonna bring this up too. Midjourney uh just posted this a few hours ago. They're acquiring uh Co-Star. Uh it says, uh Banu Güler, uh sorry, I can't pronounce, is now Chief Design Officer at Midjourney. She's building cross-functional design product front end team. So somebody with them was asking, Okay, well, what is Co-Star exactly?

1:22:33Why did Midjourney acquire Co-Star?

Play from 1:22:56 8:88:88 1:22:56 'Cause it's like, what is that? It's a popular astrology app. builds hyper personalized birth charts, daily horoscopes using your exact birth date, time and place. NASA planetary data blended with AI and human astrologers. Social features let you compare charts, compatibility with friends, sleek minimal design, sharp, blah, blah, blah. They acquired it. Integration is mostly talent and design synergy to start. But you know, like I I think of a few of the connective dots here.

Play from 1:23:21 8:88:88 1:23:21 We we know that Midjourney is redoing their entire UI UX, but it this feels very late to be a part of that. Because I think that's shipping soon. But does this roll into the the hardware aspect at all? Like these secret projects? Um mobile app. Mob yeah, mobile. it has anything to do with astrology. I I don't th if people are like going in that way, I I I don't Wait.

Play from 1:23:49 8:88:88 1:23:49 Time out. I've seen some interesting questions asked, so I I mean David clearly marches to the beat of his own drum, so who knows? We weren't expecting a spa, but we got a spa. know, that was, that was, it wasn't expecting the full body imaging scan. Maybe we're getting, you know, something with, I'm not going to go that far, but. Finn brought up a really interesting point that I want him to to sort of reiterate here.

1:24:09Has Midjourney had the right vision?

Play from 1:24:15 8:88:88 1:24:15 It was it was pre-show, but it was basically your take on Midjourney's roadmap, their vision, people maybe doubting where that goes or the validity or or sort of even the relevance to them um based on you know the publicly available information. But tell us kind of what your thoughts are there. Yeah. Well I think the narrative around Midjourney i is oftentimes that they missed the boat. You know, that because Nana Banana and you know, GPT Image or whatever are are so much better at prompt adherence and realism and blah blah blah and like they don't have an API.

Play from 1:24:47 8:88:88 1:24:47 Like what happened to Midjourney? I I I actually feel the exact opposite, which is like I think Midjourney had I think we will look back and see that David and Midjourney had the right vision all along and everyone else is gonna end up in the spot where they are now, which is that in order to like have anything well first of I Nano Banana and GPT Image are awesome products and like Those should those should be what they are.

Play from 1:25:14 8:88:88 1:25:14 But for everybody other than that, you need to have your own model and your own harness for it, meaning the front end, in order to like actually have any sort of competitive advantage and do anything interesting. Because like, what would a Midjourney API even be? Like that doesn't even make sense. Like you have the the the the product and the model are like intertwined in a way that can't separate.

Play from 1:25:41 8:88:88 1:25:41 Like, you know, using the different references and edit and remix and blah blah blah like all that stuff. Like that's what makes Midjourney what it is. And nothing else can do that. Um, so I think that's the part everyone is gonna realize sort of is is is like everyone's gonna come around to realizing that David had the right vision all along. And and also that he is deliberately, you know, people always criticize Midjourney for you know not being the best at realism and prompt adherence and all that kind of stuff.

Play from 1:26:11 8:88:88 1:26:11 And I And I think they'll also realize that that that was always also the right call to make is like well but there's already at least three, you know, nanobanana GPT and Flux, I would say, are all very good at, you know, sort of those things. Why in the world would they want to be the fourth? Right? And like to this day, like Midjourney is the only model that it that can do what it does. It just completely fucking smokes everything else at aesthetics.

Play from 1:26:36 8:88:88 1:26:36 Like, yes, prompt adherence is not, you know, its strength. Okay. Like that that's fine. Um so I think seeing and and I I I've done a lot of research of like trying to figure out the technical decisions that they made to get there and I I don't think it's publicly documented anywhere, so I don't know how they did it. But I think we're gonna see that the smart move is actually, you know, to do something radically different than everyone else in terms of sort of have image models with a vision rather than we'll say Nano Banana, GPT and flux, I'd say are very good like generalist models.

Play from 1:27:10 8:88:88 1:27:10 I think what's gonna KRIA's kind of of starting to do this is like have a specialized model with a particular focus and again Kriya I think had the same realization because they're sort of like cloning Midjourney of like you want to have the harness and the model. So that's my rant, but I think I think Midjourney will be vindicated I tend to agree with all of that only because it just it feels like if I'm feeling myself as like a ebb and flow of like the market which is probably pretty egotistical where I feel like I feel like most of the time when I'm like Shifting one way or the other maybe I'm like with popular consensus at the moment But I'm back on Midjourney like ripping a million images a day, which I hadn't done It's because I'm just again.

1:27:30Why are creators returning to Midjourney?

Play from 1:27:57 8:88:88 1:27:57 I'm so tired of the other stuff And I don't want to control everything all the time. Like I don't want the control. Like I don't want to have to prompt every little detail of a person's face in Nano Banana so they don't look like the standard Nano Banana face. Like I just want to get my idea across, maybe bash a couple things together, not have to like... Yeah, correct. yeah. Yep.

Play from 1:28:23 8:88:88 1:28:23 until the edit model. And if that's a totally separate model, is the creative little magic that you get from you know, trading off a slight bit of the coherence. I'll take that any day, all day, based on the quality that that provides. And to your point, the creative direction that maybe you spin off into. And I think when the edit model comes. That to me is is really interesting because then you need that coherence for that model to be damn good.

Play from 1:28:52 8:88:88 1:28:52 The text has to be good, right? Because if you're gonna take people back into that ecosystem in a way from Nano Banana and ChatGPT, it needs to match the coherence uh and the capabilities there at a at a bare minimum. You don't necessarily have even have to blow it past that, but you need to do at least that. I think that in another very interesting thing about Midjourney is that they're bootstrapped and profitable. So he can do whatever the fuck he wants.

1:29:10Why does bootstrapping protect creativity?

Play from 1:29:19 8:88:88 1:29:19 And I think they only have like a couple dozen employees or something. It's a really small company. So he's like, I want to make a scanner. Okay. Then he just do it. And it's so he doesn't have to like he doesn't have to do the the the things that would that that a a venture backed company would demand because that it boxes you in. Like when you take venture money, they expect venture scale returns, which forces you to go down a certain lane.

Play from 1:29:50 8:88:88 1:29:50 And, you know, he had already sold the company before he started Midjourney. So I think he deliberately was like, I don't wanna be I I don't want to do that again. I wanna do he had a bad experience too, which also played into it. I don't think very many people have a great experience with venture. Like it's not a thing where like I love VCs, that was great. don't think anyone's ever said that. VCs suck. I think VCs in creativity don't get along. You know what I mean?

Play from 1:30:15 8:88:88 1:30:15 because it's the opposite. It you VCs are looking for something that's codified, that speaks to something that's ultimately gonna give them, you know, reasons to invest further into things, right? That are probably the more similar path, right? It's not this path of its own. It's something that people have probably already done and it parallels to other things that are happening. So like I I love that David is just like, I'm gonna do whatever the fuck I want, because we're bootstrapped and profitable and some of that stuff is gonna work out, some of it's not, but you know, we need people like that who are forging a different path, otherwise we're gonna have seventeen versions of the same shit and you know, like two of them are good and then the other fifteen are why do they even exist, you know?

Play from 1:31:02 8:88:88 1:31:02 they're all competing and doing the same thing, which is it's starting to get like there's no, there isn't a lot of differentiation I don't find between GPT and Nano Banana 2. Like I can do a lot of the same things with both, right? Like it's, pretty much like one looks better visually with color. One looks better with text. One looks better with this one. It's better with that, but they all kind of do the same thing, which is why I keep going back. Cause I'm again, and I'm also like, now I'm just, I want more interesting aesthetic.

Play from 1:31:29 8:88:88 1:31:29 Like I want something that's a little bit out. side of the distribution, like you're saying, right? And that's what I'm looking for. Yeah. models. Like 8.2 Preview, like that was the first time in a while where I did that and and I was like, holy shit. Like, whoa, they cooked with this. I think if I did my usage over the last like a month, it would probably be more than the last seven months combined like Yeah It's one of those times where I've hit the button.

Play from 1:32:02 8:88:88 1:32:02 I just keep hitting it. I'll hit our 10 You know, let's go repeat 10 on this one go which I it's only what I'm having a lot of fun when I'm really in the really in the zone, but it's a It's good. glad that they're like, I don't know why they haven't just released it because it feels like everyone's only just using preview now at this point. Um, regardless, it's, it's great. Yeah, I was. just, you know, again, I don't know the guy. I've never talked to him, but I I I I I feel like I understand the type of person that David is because I I've worked with a lot of these sort of creative genius type people.

Play from 1:32:35 8:88:88 1:32:35 I just trust the process with people like him, you know, like he's gonna do what he wants to do on his timeline and I, you know, I just I if they haven't shipped it yet, I'm sure there's a reason for that. You know, again, like remember he is one of the OGs he's been doing, you know, i computer image stuff longer than almost everyone. 2010 is insanely fucking early for like computer vision. You know, he has a PhD in physics.

Play from 1:33:02 8:88:88 1:33:02 Like this guy is not your average person, you know. And if you look at him, he looks like some Tumblr art nerd, which I think he probably is, you know? And I mean that in a good way. So like that is a very I I always appreciate people in the world that have that sort of like really different particular point of view. And even if it isn't the one that even if I don't necessarily agree with it all the time, like I want more of that and I really trust and respect those people.

Play from 1:33:30 8:88:88 1:33:30 Mm. That's I think we tie a knot on that one. Tie a bow on it. Tie a knot on it. the, what's the, used the wrong phrase there. Put a bow on it. That's a good one. What is it? It wouldn't be a podcast without us screwing up something simple like that. You we can talk system engineering, but really put a bow on it. Tie it up, whatever. But yeah, you know, luckily, luckily David is a, luckily David has been innovating so that we can have a lot of fun and especially do things like this.

1:34:01Why does AI creative work need fun again?

Play from 1:34:01 8:88:88 1:34:01 Which I think is needed now in this AI space a little bit. This is my last my last little like side point here So it feels like some fun is needed again Like I felt like I don't know if you felt like this like a it feel you got real serious and like our jobs started focusing around it and like before it was like when we create crazy face portraits with paint and different textures for you know content and it was fun then it got like how do you build this, know production pipeline for my team and it became like alright now I have to write 50,000 word system prompts But look, swinging back.

Play from 1:34:33 8:88:88 1:34:33 Yeah, swing it back the other way. Let's have some fun. businesses like to mess stuff like that up. You know, like in marketing too, you know, when you get into these environments where it's like, you know, you look back at old marketing and advertising, it was so much based on feel and intuition and experimentation and creativity and novelty and you know, storytelling and this and that and the other. And then it kind of got to this business business. I don't know.

Play from 1:34:58 8:88:88 1:34:58 I'm gonna make that word right now, but. You know, where it's like, show me how many leads this is bringing. Yeah, everything has to boil down to direct attribution. Yeah. Show me the, you know, it's just like, yeah, but but like ROI doesn't have a direct connection to every single marketing decision that we're making. And we shouldn't be operating in that way because then we're really constraining ourselves, you know, like and and it is feels like a similar parallel in that sense where just businesses fuck it all up.

1:35:04Does direct attribution kill creativity?

Play from 1:35:29 8:88:88 1:35:29 That's the name of the episode. you gotta make money at the end of the day, so like I get that, but you know o oftentimes you end up being pennywise pound foolish, you know, 'cause you're chasing the dollar today, you know, and leaving a thousand dollars next year on the table. It's data ruined everything fun.

Play from 1:35:54 8:88:88 1:35:54 Anything that used to look good, it's because there was no data and it was just all vibe. Everyone was just vibe creating back then. It's like, why should we make this car brown? I don't know, because it looks awesome and it's a square, but then everyone likes it. And it's like, that's cool. you know, it's like, data to me is like, no. No, okay, then fucking make it white if you want. I don't care. let's put dolphin wings on the 59 Cadillac Eldorado and make it hot pink and make it 19 feet long.

Play from 1:36:26 8:88:88 1:36:26 And everyone's like, cool. But if there was data saying that that wouldn't work, you'd never do it. It's like, everyone likes the safe Kia. And and if it was and if if data, smart people, and compute resources were all it took, well then Meta, Google, Amazon, Microsoft wouldn't have any flops, and yet they have tons and tons and tons of flops, so clearly it's not that easy. Yeah, yeah, it's a...

Play from 1:36:54 8:88:88 1:36:54 I just hate designing. I probably just have to do a whole podcast on how much I hate designing with analytics in mind. Like, just... it's like I like it because it's like a guidepost, but then, like, I feel boxed in by that one specific sort of thing, and it feels like it takes all the fun out of it, and you're not creating anymore. You're just engineering, like, for something very specific. and if that worked, then I would feel I'd be like, Okay, well I mean I can't argue with if if if that's if you push this button and you make more money, well I can't argue with that.

Play from 1:37:22 8:88:88 1:37:22 But that's not how it works. yeah. Oftentimes it's got to differentiate, but I think, uh, think we gotta, we gotta, we gotta sew, you know, sew this one up. Is that a, is that the way we, is that, let's, let's sew this one up. Sew it up. Is that a new one? It's only because... So Roy's got a very small bladder over here. And also, it's just the way the week, weeks tend to work on Fridays now.

Play from 1:37:50 8:88:88 1:37:50 But we wanted to also just on a side note from Finn here, thank anyone who dropped us some hints last week on, you know, some guests and things like that, that we should be bringing on and spend time, you know, recommending people and also, you know, understanding that we are always looking for content ideas. Everything was noted. We took it all down. We're tracking it all. even though we shouldn't track data, is what we just said, but we have it.

Play from 1:38:17 8:88:88 1:38:17 We're getting organized. Damn it. What have we devolved into? But there's, out, shout out Finn, man. You know, great stuff here. I mean, an awesome guest. We got in some really good conversations. So shout out. I'll I'll I'll make sure that we include the links to all of his channels, website, et cetera, in the description so you guys can find it there. And um, Rory, who should we, who should we shout out for subs this week?

Play from 1:38:46 8:88:88 1:38:46 We gotta so Finn first off again. Thank you for coming on and getting getting through a lot of that so that stuff with us, but the To the the sort of like the tradition here at the end is we have to tell somebody, know If someone's made it this long, they made an hour and 30 minutes. It's got to do one thing I don't care what it is. Just like comment subscribe tell somebody typically we got to find someone to tell on my LinkedIn post during my like this Someone said we should ask our local quantum physicist or tell them about it.

Play from 1:39:13 8:88:88 1:39:13 It's a new one. Haven't heard that one before. I've gone through it's about every occupation here. So yeah, your local astrologer or your palm reader. That's a new one, I think. I don't think we've done that one before. So if you got one, ask them, tell them, whatever. Just tell them to come here. your moderator on Discord. Yes, tell them. Please tell them.

Play from 1:39:41 8:88:88 1:39:41 Awesome stuff, gentlemen. this was a great episode. Episode seventy five, wrapped this baby up. And um yeah, we'll catch you guys in the next one. Have a great weekend. See ya. Thanks guys.