Ep. 83: 20 Year Creative Vet Sold Golden Wolf to Doodles, Then Went All-In on AI

0:00:00Who is Ingi Erlingsson?

Play from 0:00:00 8:88:88 0:00:00 What's up, everybody? Fast Hours Podcast, Episode 83. Drew Brucker, Rory Flynn, special guest today, Ingi Erlingsson. If you guys are not familiar with Ingi, I don't know, maybe you're living under a rock. You've seen this guy's work, especially on X lately. That's where I think I've been seeing most of it, but it's incredible. Icelandic -born creative entrepreneur and director who went from graffiti and design to building Golden Wolf, selling it to Doodles, became Doodles' chief content officer, then started a new company called Systems where he's building custom AI creative infrastructure and doing some really cool things there.

Play from 0:00:38 8:88:88 0:00:38 Worked with basically every brand that you can think of, Nike, Adidas, Jordan, Disney, Cartoon Network, Netflix, Coca -Cola, list goes on and on and on. So major, major accolades there, but tell the people that maybe are living under a rock a little bit more about you, and we're excited to have you here, man. So welcome to the show. Thanks for having me. Yeah. Yeah, I feel like you nailed it. That's my entire story.

Play from 0:01:04 8:88:88 0:01:04 Like that? Yeah. So I've done a lot of different things, as you mentioned. At the core of it, I'm a creative, kind of coming up from graffiti and just kind of making stuff my whole life and kind of ended up on the internet, making stuff for the internet largely and obviously done a lot of stuff for TV and movies. But a lot of the stuff I've done has been focused around making content for the internet and kind of giving people little hits of dopamine.

Play from 0:01:36 8:88:88 0:01:36 It's kind of in the summary of what I've been doing, whether it's ads where we, no one likes ads, right? And our thing was always like, if you're going to make someone watch something, then make it worth a while. And so even in a context like that, the sort of privilege of being able to make stuff that isn't just like, hey, buy this toothpaste. Actually, people coming to us and letting us do something fun.

Play from 0:02:02 8:88:88 0:02:02 And I think that's translated to a lot of the stuff that I put on socials now is like trying to make stuff that I enjoy making, but also kind of give someone something in return. So maybe inspiration or a little bit of dopamine. Can we just take a moment for brands to hear that last part about the advertising? Because you get so many, I feel like you get so many briefs over the last 10 years, where it's just like the safest, most boring thing you've ever heard.

Play from 0:02:32 8:88:88 0:02:32 And it's just like, now everyone's like, AI is going to, you know, it's not real art. And it's that you see like the brief and you're like, no shit. This is, this is art. You know what I mean? Like it's, that's not art. That's just like a, you know, a glorified sales script that is just like you said, buy my toothpaste. There's nothing interesting to it. So dude, I mean, some of the, some of those names on there are big hitters though. Like, uh, curious, like Rory, what do you like?

0:02:58How much creative freedom do top brands give?

Play from 0:02:59 8:88:88 0:02:59 I'm curious from an outsider's perspective, like how much creative freedom do these bigger brands give you? You know, kind of going back to Rory's point about maybe like sometimes there's like sort of this, you know, adverseness to risk or this, that, the other, but obviously like they chose you for a reason and your strengths, like, do they allow you to fully utilize those strengths? I mean, tell us a little bit about that when you get to working with sort of like those major players.

Play from 0:03:25 8:88:88 0:03:25 Yeah. Um, so I mean, it varies. So like, like, you know, no, no commercial brief, like regardless of what it's for is completely open. Like the only, the only open ones would be, um, we've done MTV idents back in the day when, when that used to be a thing where like, it was pretty open. They might, they might, might have been a theme, but like, you know, they, they really would let us do whatever we wanted. Um, but in terms of like commercial stuff, like largely it's down to, uh, whether or not you're hired for what you do, right.

Play from 0:04:00 8:88:88 0:04:00 Cause it can be, it can be executional or it can be creative. Right. So sometimes, uh, you know, uh, an ad agency will come to you and they'll be like, Hey, like, you know, so when we were doing gold, like when I was at Golden Wolf, like, it would be like, Hey, we want that Golden Wolf thing. Like we just like, we love what you do. You know, we did a, did a really, really cool, um, campaign for, for a, um, a soft drink. Um, and it was just like, go crazy.

Play from 0:04:27 8:88:88 0:04:27 Right. And, and, and they really wanted like what we did and we're referencing our stuff and stuff. And so, you know, those types of briefs, there's more creative freedom, but a lot of the times what will happen is, um, by the time you even get brought in, there's already a storyboard, there's already a script, you know, everything's kind of been approved by 20 different layers of, of, of, you know, all the, all the stakeholders in the company and stuff. So, um, sometimes it can be a little bit prescriptive.

Play from 0:04:55 8:88:88 0:04:55 And then the, the challenge for, for, for a creative is to, you know, is to, is to, to make it better. Right. Like, so, so, you know, you, there's a, there's a, there's a risk that you could take it and go, okay, well just execute what you've given us here, or you kind of give it your own spin. And so it's like finding, it's finding the, the areas you can inject your creativity and, and, and really make it something different. And I think that the people that are the most successful are the ones that can navigate that the best.

Play from 0:05:24 8:88:88 0:05:24 Right. So like, you kind of imagine that like, you know, best, best craves in the world, or like, you know, I guess even like music video directors and stuff like that, the, the best one never gets told no. And I don't actually think that's always the case. Like often the best ones are just the best to navigate the client. And, and kind of, uh, molding their brief and molding the thing into something really great without having to just do their own thing.

Play from 0:05:50 8:88:88 0:05:50 I have a, I have a studio question, right? This is something that's not very familiar to me in terms of like where I've operated in my career. Like, cause it feels like there's certain studios that have essentially like a brand. Like, like they bring like a look, a feel, a vibe, and it feels like they get hired for that. The other ones feel like they're just, they're there to execute. Like you said before, like, do you, do you find it that like, you have to build towards that? Like you have to start at the execution to build towards that, having like that creative freedom of like the, you know, like where they're hiring you for your look and your vibe, or is it like, you know, maybe if you're starting a new studio or you're doing something like, you know, a lot of people are probably starting AI studios right now.

0:05:56How do studios get hired for their style?

Play from 0:06:28 8:88:88 0:06:28 Right. And a lot of them are probably just like execution space. So the reason I ask is because I'm curious of like, if it takes equity to get to the point where you can have your vibes or if you just start there, if that's better, it might be a really random question, but it's something I've always like kind of thought about in my head. Yeah. No, I mean, I mean, we've kind of, um, uh, been, been through that journey. The truth of it is that like you, it's very difficult to become the agency that, or the, or the studio that people come to for that creative thing.

Play from 0:06:58 8:88:88 0:06:58 If you started as the executioner, like it's much harder to go into that than the other way around. It's very easy to, uh, you know, people at agencies are, or brands are like, you know, just like everyone else, like, you know, they, they are drawn to, uh, things that they like or, or, or, you know, or celebrities, they want to work with this, this and that, you know, actor or whatever. And so when it comes to creative, they want to work with the people that whose stuff they like and they admire.

Play from 0:07:28 8:88:88 0:07:28 And if, if they've kind of got their head around the idea that, oh, this is the execution company, it's very difficult for them to, in their minds to kind of like connect that then with some new cool, exciting thing. So it's much harder to go that way around. Not, not saying it's impossible, but I would say if you, if you want to, uh, be a studio that gets hired for your thing, which is, which is incredibly, it's incredibly hard thing to get.

Play from 0:07:55 8:88:88 0:07:55 Cause even like the biggest, the biggest studios, they might get it to a degree, but like probably 90% of the work isn't, they might just be the stuff that they show. Um, I think it's basically starting off and this is kind of what, what we've been doing and what I've been doing is like, you know, four months ago I was just doing like Breaking Bad memes or whatever. Cause I didn't, I didn't want to put style forward. I didn't want to like, I didn't, I didn't want to be responsible for creative anymore. Right. Like I just wanted to like, you know, make it work with AI or like figure out a system.

Play from 0:08:25 8:88:88 0:08:25 And, and, but when, when I started making stuff that was in my own style, all of a sudden flooded with inquiries, like flooded with people going like, can you do that for my brand? Um, and I think it's the same for, for, for most studios is like, you, you start by your, your work being the kind of the billboard for what you want to be doing. And that's what we did at Golden Wolf for years is that like, we, we went really hard into Instagram when that first came out.

Play from 0:08:53 8:88:88 0:08:53 And I think, I think we were the first animation studio in the world to get over a hundred K followers, like even before like Disney or whatever, cause it was, it was so new. Um, but we really invested in our social media and, and the idea with the social media and it was all kind of original work for the most part, it wasn't just showing the work that we've done for others. And the idea was that why, why have your future clients come to you having looked at your portfolio, asking you for something on the portfolio that might be two years old, why not like show them what you want to be doing?

Play from 0:09:23 8:88:88 0:09:23 And then eventually they'll come to you for that. And I think that that's the key is like, uh, you know, go out with the work that you want to be doing, but expect to have to do other stuff. And then eventually over time, people will start coming to you for that. But like, but that's the part that takes trust, right? Is that like, um, once you've proven that you can do the thing, then they'll be more, more, more likely to then let you, let you go crazy.

Play from 0:09:50 8:88:88 0:09:50 Thank you for, for answering my, my long winded, non sort of linear, not easy question there. It's just something that, you know, I've always thought about, because I got, you know, the studio world feels like it operates at a different, at a different sort of, there's a lot more like emotion attached to what it feels like a lot of times, a lot more subjective preference based on who's bringing people in. Um, cause you know, I've, I've been in those rooms on the other side, on the brand side where it's like, someone has a really good vision for what a production company can be and do for us.

Play from 0:10:22 8:88:88 0:10:22 And then it gets to like, you know, like you said, the 17 other layers and they're like, this is a little bit too edgy for us. Like, you know, we don't want the 24 look. We want the pharmacy look, you know, the pharmaceutical look where it's like, rom -com kind of, kind of vibe. So it's always, it's always sort of like, uh, been in my head. Cause I've been, I've seen a lot of stuff killed. That was awesome and epic because people just wanted to take a more like clear path where they controlled it versus like, let the creatives run wild.

Play from 0:10:53 8:88:88 0:10:53 So that was, uh, again, another long winded response by me. But I think, uh, I mean, there's a, like, I kind of saw a parallel with that as well when I, when I went over to Doodles, cause then essentially brand side, which I'd never been before. And so we were hiring studios and, and, and contractors to work with us in a similar way. And I also got to be on the client end of that. And what I realized was that I'd spent my entire career thinking that brands were brands, agencies were out there looking for something cool, brands are looking for something that sells Yeah. And so like, like I never once had a thought about that from a studio perspective, we were just trying to be the coolest, coolest ones out there thinking that, well, that will obviously sell or whatever, but really like what, what was more important is that, well, well, actually like,

0:10:59What did Doodles teach Ingi about brand work?

Play from 0:11:40 8:88:88 0:11:40 here are the stats like that, that, that's, that's a better selling point for a brand and look at this cool stuff. Do you think that's changed? Sorry. That was my last one. Do you think that's changed over the last, how, you know, you said you've been doing this for 20 years. Do you think it's more about the stats now than it is about the, the vibe? Um, I think, I think it's changing really, really rapidly. I think, I think that now it's about, it's everything's about attention right?

Play from 0:12:12 8:88:88 0:12:12 Like the, the, the big tech companies have been in the, the attention game for a long time. And I feel like everyone's in the attention game now. Um, and so I think what used to maybe, you know, maybe require a lot more, like, you know, uh, ad campaign might take six months, eight months to like make, you know, and now, so that's, you've got to do it in two weeks or whatever. Or like, and that, that kind of time span is, is, is getting shorter and shorter now.

Play from 0:12:39 8:88:88 0:12:39 And, and, um, um, so I think, I think in some ways that's shifting towards like, who can prove that they can get attention? Like who can prove that they get eyeballs? Like, you know, what's the, going to be the most effective thing? Because this ad will be forgotten the second it goes out. And so like, I think that probably is, is, is something that a lot of people will increasingly pay attention to that maybe hasn't, hasn't been such a big thing before.

0:13:05How do you turn attention into recognition?

Play from 0:13:05 8:88:88 0:13:05 What do you, what do you think the secret is to getting attention now? Right? Because I think in a lot of ways, like we're, we're kind of talking about this and alluding to it already, but it, it's the capability that, you know, any, anybody can create this sort of like cool, creative, novel, you know, eye catching thing in, you know, instantly, or at least in a lot less time than they could before. And so you're, you're getting flooded with that. Um, distribution feels harder now too like these algorithms are, you know, kind of tightening up in a lot of ways.

Play from 0:13:36 8:88:88 0:13:36 And it feels like the, the only true ways to really like, you know, break out of that is to already have a large following, or you're kind of doing these, like really, you know, you're playing the, the social games, like with comment this and get this right. And you're, you know, you're doing all that crap, but like, what are the things that, that drive attention and maybe even like mindshare to the point of people remembering you after the attention has faded a little bit?

Play from 0:14:04 8:88:88 0:14:04 Yeah. Yeah. I think, I mean, those are two, two, two different things, but that, you know, that can be the same thing, like, you know, being remembered and getting attention. Um, I did a video of, um, when Tim Cook resigned, I did a, I did a, like a meme thing where, where I, uh, made him, Leonardo DiCaprio and Wolf of Wall Street, where he's giving his big speech about not leaving Um, and it got like, it got a half a million views on, on, uh, on, on X and, um, you know, did fantastically well.

Play from 0:14:35 8:88:88 0:14:35 And I think it got about 20 followers from it. Um, so that was a lesson in itself, you know? And so that, there, there was, there was a lesson for me. I mean, I mean, um, I did a Breaking Bad thing that, you know, got millions and millions of views and, and did, did get quite, quite a bit from that, but because the, because of what the content is, because it's a meme essentially, um, you know, who knows who made what meme, you know, who cares? Like, like that, that, that isn't how memes work, right?

Play from 0:15:02 8:88:88 0:15:02 You can remember the thing, but you will not remember the person behind it unless they were in it, you know, unless they were the meme. Um, and so as far as like being remembered for something is definitely like, that's definitely not a great avenue, unless you're Beeple, right. And who can, who can do those kind of, you know, the, the meme style things, because obviously he's got his, his style and, and, and that. So there's a lot of things that kind of contribute towards, you know, this attention and, and, and, and, and blowing up.

Play from 0:15:30 8:88:88 0:15:30 Um, well, uh, Jack Dorsey liked that earlier. That was my, probably my best like ever. Yeah. Oh, really? That's a good one. That's dope. It's hardly got any likes, but Jack Dorsey liked it. I don't think he's ever liked it. Okay. So, so, so I'm giving up on the Tim Cook thing for now because it's buried here and I'm just scrolling past too many good, cool things here. I mean, like, so I'm just going to like, kind of, I'm just going to like, kind of scroll through this, feel free to like talk about it. But like, obviously you've got like this really kind of neat, I don't know.

Play from 0:15:59 8:88:88 0:15:59 I wouldn't even call it a singular style, but like these, they feel connected. There's definitely an artist's touch to these and a, and a parallel that drives between the different things. But I mean, I, I came across you on X, I don't know, several months back now, I didn't know your name, but I saw the things that you were doing. I was just like, wow. And he was at Doodles. And I, I had already been familiar with Doodles and I was like, this, this is, this is, this would be a great follow.

Play from 0:16:25 8:88:88 0:16:25 And I saw you start to put this stuff out and I'm just like, wow. Like, I don't even think people know that you're on here yet. And I, you know, just the, the, the work that you do is so polished. And so I just love every single thing that you're putting out here. It's so unique. It feels very thoughtful. And it's just like this really cool blend of AI being a part of the process, but not being this, this whole driver of the process either.

Play from 0:16:51 8:88:88 0:16:51 Right. You could tell that you've sort of put the pieces of the eras together, um, from all of your taste and your experience through client work and creative work as well. So what I love about it, to be totally honest, it feels like, it feels like I'm watching the making of, but in an artistic way where it feels like you're going through a number of different passes and phases. Like, you know, you have your standard traditional like photo realism, but it'll, it'll morph into, you know, depth maps.

Play from 0:17:18 8:88:88 0:17:18 It'll morph into skeleton rigs. It looks like some parts of it are gaussian splatting. Like it's, it's really cool. It looks like the creative process. Like if I was to like encapsulate, like what the digital creative process looks like for potentially something like, you know, either VFX or whatever. Like, I feel like I'm, I'm like watching that as like an art exhibit, if that makes sense. I don't know any other way to describe it, but I know it's your video. Cause every time I scroll and see it, I'm just like, I know that like, and I, and I like that. And I watch every single one probably through about three times. Like what's he using for keyframes?

Play from 0:17:52 8:88:88 0:17:52 What's he doing? How's he stringing this together? Definitely using ComfyUI. Maybe he's not using ComfyUI, you know, like I'm trying to like back engineer it myself, but dude, it's just really cool, man. Maybe, maybe he'll share a little bit of a secret sauce with us. Yeah. Yeah. But this is, it's got that unique look, you know, Rory, you bring this up a lot, but it's got that look where you, you, you could look at it just in a glance and you know who did it. Right. And I think that's like one of those signature things. So yeah. And, and on the, on the, the kind of the, the sort of a behind the scenes look thing.

0:18:13How does Ingi build his signature AI style?

Play from 0:18:20 8:88:88 0:18:20 So what there's a couple of things behind that. So one is, um, a lot of these things are me testing out tools that I'm making. So like, whether it's like custom nodes and ComfyUI or whatever, like, and it's just like playing with them and seeing what I can do with them. Uh, and then they just become like shorthand or, or, or really easy for me to grab because I've already made them. Um, but then the other thing is like, uh, well, it's a little twofold. Like, so I think one inspiration is like what, like dealing with these AI models day in, day out and like getting really familiar with them.

Play from 0:18:52 8:88:88 0:18:52 And like, like I don't really like using like new models cause like I really want to know the models. Like I use Wan for the most part, like Wan 2.2 was like 2.2. Okay. Like, like pixel quality wise, best model there's ever been like, like better than Seedance, better than all those but not great for prompting. Um, but, um, so getting to know these models, like really, really well, like you start to get a sense of like, what's going to happen if I do this, like what will happen if I do that?

Play from 0:19:21 8:88:88 0:19:21 And so a lot of, a lot of these are kind of inspired by the idea of like, how does AI see us, right? Like what's the LLM view of a human? Like, because it's piecing together all this stuff in the latent space is picking like all these different references from like all over and, and combining them into this thing. And then it presents you that thing. And so like, what does that, I mean, the latent space is just like math, math, mathematics, whatever, but like, it's like the idea that you can visualize that space and, and, and you look at a human, let's say you're, you're looking at a person and it will look at information about the person instead of looking at the human first.

Play from 0:19:57 8:88:88 0:19:57 Right. So that's kind of where a lot of the overlays come from, where the, every, every single overlay is extracted from the video. And so that it's all tracking data. It's all like, you know, points that in this one, it's like, you know, like tracking, like where the shoes are, where there's a phone box where like, so those are the things that flash up, like they all track things. So everything is derived from the video rather than just being stuck on top.

Play from 0:20:22 8:88:88 0:20:22 And so that that's part of it. And then this sort of the bigger, the bigger part is that I want to make stuff like, I'm kind of, kind of sick of the debate of like, you know, is, is AI or whatever. Like I, I, I don't care about that label. Like people call me whatever that, whatever they want. But, um, but what I, what I think the, the partly the issue is, and like, you know, I've been in art or creative for a long time and I can understand the frustration that people have towards it.

Play from 0:20:51 8:88:88 0:20:51 But what I want to make clear is that like, it's possible to make stuff that takes effort and like, and, but also can be done in a way where it doesn't matter, like what it's made with. And so I'm trying to almost like put the AI in this, like in front. So like, I don't want people to think that it's just, I don't want it to look like I'm managed to make a really believable version of a real thing. I want people to know that it's AI as soon as they see it so that they like it either like, despite it being AI or because it's AI.

Play from 0:21:21 8:88:88 0:21:21 And so like, rather putting that forward than like, look at this, look at this like five minute fake movie trailer I did that almost looks believable. I'm like, well, that, you know, I don't really care about that. Well, you know what's, what, what's also interesting too is right. The perception that exists once you mentioned AI was involved in the process. I think that a lot of the things that you're posting, you're not, every post doesn't say, Hey, I did, you know, this part of this was in AI. You're just like, you've got a name for it here and it's posted.

Play from 0:21:47 8:88:88 0:21:47 Right. So, so people probably aren't initially thinking this was entirely created by AI, or they're not really thinking deeply enough as to where it was used in the process. Maybe people like us were like, I wonder how he set this up and did this. But you know, to that point, it's, it seems like as soon as you throw it out there, right, it's all, it's already polarizing. And so a lot of people will just always kind of put out, Hey, look at this trailer I did with C dance. You know, like, and then it's like, okay, well you're, yeah, somebody is going to come and try to take a shit on you because of that.

Play from 0:22:18 8:88:88 0:22:18 Right. But if maybe you just posted something and to your point, maybe even the style is not photo realism. It's got its own sort of characteristic, just becomes a little bit more in the background. Right. And so it's not so polarizing in that aspect too. Well, I, I did like, I mean, I only started, you know, I've started doing this like title thing recently, but like before that, you know, it feels like months ago, it was probably like three weeks ago or something like time goes so fast or slowly.

Play from 0:22:49 8:88:88 0:22:49 But I used to list every single tool. And it just like, you know, it, the conversation became too much about that, you know? So I just, I was just like, well, cause what, what I didn't want to do, like, like I also, I'm not trying to hide anything, right? Like I'll post up like all the layers that go into it, whatever, like I open source the nodes when they're ready. Like, you know, I try and share this stuff, but I just didn't really like, like, I'm not trying to hide that it's, that it's AI, but like the conversation became only about the tools.

Play from 0:23:21 8:88:88 0:23:21 Whereas like the point is that like, you can use all these tools together to make something really, really interesting. But you know, if people ask like, what's the prompt? It's like, I don't even know. Like the prompt was, Hey, LLM. You're asking the wrong question. Yeah. Your asking doesn't matter. Yeah. Like the prompts, the prompts are like, Hey, LLM, describe this image. You know, like that, that, that's how I prompt, right? It's like, I don't really, don't really write anything. It's just like, here are some images that describe emotion.

Play from 0:23:48 8:88:88 0:23:48 Just like describe it for me. Um, but, but the point should be more like, how did you get the idea? You know, or like, what's the meaning of this? Cause I think that's what I see here. People aren't talking enough about that. You know, where, where did you source the idea? What was the process of the ideation standpoint? And I think like even AI could come into that, but you still have to contextualize where it's going to get the idea from or how you're going to approach a situation in the first place.

Play from 0:24:14 8:88:88 0:24:14 Right. And you know, even times from taking raw materials and then having that be enough context to build novel and creative ideas to then settle on the idea, right. Before you even move the mouse or type something in the keyboard. Right. I mean, like there's a lot that goes into it before even the creation part of it. Go ahead, Rory. Sorry. No, I was just going to say, cause I feel like like this conversation and I, and I look at your stuff and I feel the same way.

Play from 0:24:41 8:88:88 0:24:41 I feel like this might be a bad analogy, but I feel like in the AI space, it's like the same as like, if you gave someone like a piece of paper, like there's, you know, 99% of people are just going to like use it like a piece of paper. They're going to write on it with like a pencil. And then there's someone that's going to make origami, you know, like it's, it's like, is that a new analogy? I just made that one up in my head, but it feels like, you know what I mean? Yeah. There's like people that use it for what it's supposed to be used for. And there's a limitation to it.

Play from 0:25:06 8:88:88 0:25:06 Cause there's not beyond, like, they don't think beyond what it can do. Like what I see here is like, you're probably doing stuff. You're finding things. Maybe you messed up on a couple of things. Oh, let me wrap that into the whole creative pie. Like I, you can see it when you look at this stuff. It's not like this. I can, I'm generalizing. You're probably going to tell me I'm wrong, but I see stuff in here where I'm like, there's no way that was the intent, but you ran with it and made it into something else. Like, I mean, this is, this, this piece is a really, really good example of that.

Play from 0:25:33 8:88:88 0:25:33 Right. I mean, and I never really, I never got this into a place I was happy with, but like posted anyway, but like, but basically, uh, so you can see, like you see the graphics going around. So like, um, you know, everything was initially meant to be like masked inside that shape. And then I think I just forgot to turn the masking off or whatever. I was like, Oh, now it's created this like depth where it kind of goes inside and outside.

Play from 0:25:59 8:88:88 0:25:59 I'm like, okay. So, and then, but that's, that's exactly how, what my process is like. It's like all of it's trial and error. Yeah. And then for the next one, I'm like, well, that thing worked pretty well last time. So I'm going to use that again and then try something new on top of that. It's, it's cool. I was just going to say, that's the beauty of, that's the real beauty of working with AI for me, with AI for me is that, is that, you know, I can't think of any other medium where you can iterate and experiment so quickly.

0:26:13Why does AI make experimentation cheaper?

Play from 0:26:28 8:88:88 0:26:28 Um, you know, the, you know, the cost of iteration and in traditional media, I mean, like we used to spend 25 K on pitches, right? Like, and not win you know, whatever. And like, and, and we, you know, you'd spend three weeks on a pitch and you'd end up with like four images and some writing. And it's like, it just, it's so slow and expensive. And most of the time you're just panicking towards the deadline. Right. And like, you're not really able to go like, but what, what if, right, what if we did this?

Play from 0:26:56 8:88:88 0:26:56 And like, that's, that's the thing that's so exciting about it is like, you know, I had to look up earlier, like, like when I made this spot, for example, and I was like, it must have been like four months ago. And it was like six weeks ago. And just the amount of stuff that I've made in the, in, in the meantime, and like, you know, and, and this is just stuff I do outside of work. Right. And, and, you know, the, the, the amount of learning I've been able to do through being able to make that is, is, is what, what makes it so amazing for me.

Play from 0:27:24 8:88:88 0:27:24 Um, you know, it's not, it's not the, it's not that it's easy to make. It's just that you can, you can make things and learn, learn much quicker. It, it's sick first. I mean, I'm like, I'm listening, but I'm also just like mesmerized. I'm just staring at the, they're all, they're all, they're all so good. I can't take my eyes off it. I don't know what it is, man. I just, what do you, what do you call, what do you call like this, this style, um, that you have going on here?

Play from 0:27:50 8:88:88 0:27:50 I mean, what, what do you, how do you, how do you describe this? I don't know. I mean, like, it's just motion graphics to me. Um, yeah, I don't really have a name for, I mean, I think one thing that I do maybe do differently to, to a lot of other people when, when they, when they approach their, their, um, AI stuff is that I, I look at people like, like they are like objects or elements.

Play from 0:28:16 8:88:88 0:28:16 So I don't like kind of like an emotion graphics animation where you might have like block of text that, and you know, animates like this like, I think of the person and how they move as if they were like a shape. So like, rather than, okay. Right. So I'm not trying to tell stories here, really. Like I'm trying to, I'm trying to, I'm trying to convey an emotion through movement, almost like dance or whatever, a lot of the time, um, where like, I think about, so, and everything's mostly, mostly is done with like very short bursts of like first frame, last frame, um, where I, I, it's like almost like key frame animation where like every key frame is planned out.

Play from 0:28:51 8:88:88 0:28:51 Um, and like the pose and the, and the distance from camera and all that is, is designed so that it creates a certain type of movement when it goes between. Mm -hmm. And what, how much of the, like something, you know, I'm sure all of these are different, but like for a lot of this process, like how much of this is, is AI versus how much is like post -production or After Effects or things like that? Um, so every, every piece is, well, not every piece, but a lot of the pieces that have a really similar, similar structure. So, um, the first thing I do is so that there's, there'll be a clean version for everything, right?

0:29:01How do AI and After Effects work together?

Play from 0:29:26 8:88:88 0:29:26 So there'll be like a, a version that's just like a perfectly smooth morphing thing that just goes like loops basically. And then, and then everything on top of that is, is added in After Effects, but like all the tracking and, and, uh, normals and everything else is all, all generated and ComfyUI. Um, everything's extracted with AI, but then it's put together in After Effects. I was going to say, I don't I don't even, I don't even want to dream up how to get some of this stuff in, in a, because it looks, it's very controlled.

Play from 0:29:57 8:88:88 0:29:57 You can see like how controlled it is. Um, I think when I look at it, at least like I, that's what I see. Um, so I was always curious about like how much of it is actually being done. If you've like gone that deep on the ComfyUI side to get it to that point, because I'll ask you later when we get into more technical stuff, but I have, I have, I have questions about the post and the, and, and After Effects, but we'll, for the, for the purposes of this, I think it's, it's really cool. Like if you, were you always ComfyUI?

0:30:20Why did Ingi move into ComfyUI?

Play from 0:30:23 8:88:88 0:30:23 Was ComfyUI like the place you started, you evolve into it? Like where, how did we end up there? Uh, well, I started, started Midjourney. No, I started in DALL-E 2, actually. All right. Early 2022. Um, and then Midjourney for a while. And then, uh, AUTOMATIC1111 um, and then, and then ComfyUI. Um, and when, when like, just, just after that come out.

Play from 0:30:50 8:88:88 0:30:50 Um, yeah, so been, been, been very deep in, in ComfyUI since then. And then, um, uh, I've started branching out a bit now. I got, got really into, um, uh, Astra's really brought me into the agentic thing. I'm absolutely in love with, with working with agents. Um, I used to be really pessimistic about it. I, I, I was just like, what's the point. I can just do this. I could just create an agentic flow in, in ComfyUI through like lots of nodes or whatever.

Play from 0:31:20 8:88:88 0:31:20 But, um, you know, now I'm, you know, building, building systems and, you know, have like sub agents for things that I have to do constantly. And, and, you know, now it's just like, it's one action to do something that might've been 20 before. So like, yeah, I'm getting really, really into that, but yeah, ComfyUI, ComfyUI is like, is like home base. Okay. And what kind of, what kind of hardware are you running for this?

0:31:42What hardware does Ingi use for ComfyUI?

Play from 0:31:45 8:88:88 0:31:45 It feels like this might be, we need some superpowers here. RTX 6000. Okay. Um, and, um, only 128 gig of RAM, which I realized isn't enough for that, that GPU. Um, okay. Yeah. And I, I, I, I found a note this week, um, that I'd written somewhere where I was like looking at, uh, prices for like upgrading it.

Play from 0:32:11 8:88:88 0:32:11 And, and I don't know how old the note was. I don't think it was older than like three months old. And it was like 1500 bucks or something to upgrade. Now it's like eight grand. Oh my God. It's crazy. It's insane. It's not easy to go up from 128. You got to go like 256 or whatever. So, um, but yeah, that is just, it's gone crazy. So I'm, I'm constantly restarting ComfyUI just to like clear my, my RAM. Oh man.

Play from 0:32:37 8:88:88 0:32:37 For, for anyone who's not familiar with ComfyUI, you're running it, you're running it locally. So you need some power to generate this thing. You know, a lot of times, I mean, I remember going into some businesses who were like, can I, can we run this? Can we do this on ComfyUI? I'm like, you can, I'm like, what hardware do you have? And they were like, you know, like Mac, like, uh, you know, iMac or, uh, what's it called? The MacBook Pro. MacBook Pro. Yeah. And I'm like, what chip They're like the M1.

Play from 0:33:02 8:88:88 0:33:02 I'm like, M1. No, you're toast. It's not going to work. They're like, why not? I'm like, go ahead and try. I don't know. I don't know what to tell you. Just like, it's, you talk to your 3D, talk to your 3D team. I think weren't we on the M1? Like when, when it first started, like AI was still, and then we were like, yeah, this is, this ain't going to last long. Like we both, I just, we both had to upgrade. It was so incredibly hot. I mean, that's, that's when I, I switched to, I finally made the switch to a PC.

Play from 0:33:28 8:88:88 0:33:28 This must've been like two or three years ago. I've got a PC laptop and I was like, oh, this is great. Or whatever. Like, and compared to the Mac, it was, it was brilliant. And then when I got the desktop, I was like, oh God, the, the laptop's never coming out again. So I'm just like, now I've just got this absolute beast of a, of a PC is like, I bought up the box is like, I didn't, I just didn't see on the picture how big it was, but it's like twice as, I could fit two PCs in that easily, maybe three.

Play from 0:33:56 8:88:88 0:33:56 It's so for like, let's, let's maybe talk ComfyUI for a little bit. I think probably a lot of our audience is not, not as like overly familiar. I personally have used it. I have not engineered anything of significance on there. I find it to be very technical if you're a technical person. And that's something that I think if you, if you love it, you're going to absolutely love it. If you have, if this is like, looks like hieroglyphics to you, you're never going to use it.

Play from 0:34:22 8:88:88 0:34:22 But, you know, from like a, maybe from like someone who, for someone who hasn't been in there and like what you can do and like how you can sort of like the, the granular level of control, you know, maybe just give someone a little taste of that. Yeah. I mean, it's, it's an interesting one. Like, um, uh, it, people often, I often see people talking about like comparing, comparing ComfyUI to like node -based workflows on, you know, on various AI platforms and, and kind of, you know, figuring out which one's better.

0:34:34What does ComfyUI let you control?

Play from 0:34:55 8:88:88 0:34:55 And I'm like, you're, you're, you're comparing like autocomplete with an LLM. Like it's, it's, it's like the, and you don't need like, like for a lot of things that people are doing they don't need ComfyUI, but the difference is pretty, it's pretty staggering in terms of like the control you can have. I mean, obviously there's a lot of models that aren't open source and therefore you can't really do anything apart from just prompt them or, or give them references. But, um, for those that you, that you can control within ComfyUI and, and, and also, um, you know, that there's a lot of, there's a lot of functions within ComfyUI that, that, that, that are really helpful for this, but like, you know, you can, you can take a model and, and, um, change the way it functions, like, you know, change like the parameters and, and, and, um, you know,

Play from 0:35:42 8:88:88 0:35:42 you, you know, let's say you generate something and you're like, oh, it's like the motion blur is too high or the movements a bit slow or whatever it might be is like, yeah, just change some parameters and that'll fix that. So it's not even a prompt thing. It's a, it's like a, you know, almost like CGI or very much like CGI. It's like, it's, it's really granular level, um, control, but then also what it does is, um, pretty much everything you can do in After Effects, you can do in ComfyUI apart from the, like, obviously there's no timeline that there are some timeline nodes, but, um, you know, a lot of the nodes that I've built have been just me making an easier way to do things that I would have done in After Effects anyway.

Play from 0:36:25 8:88:88 0:36:25 Um, just that like took forever to do in After Effects. And I was like, well, I could probably build a node for that. And so, uh, um, yeah, so like, so like, sorry, I'm just laughing at that. I hadn't seen this one before. It was great. Oh man. Yeah. It's a, that was a, that was a fast follow after the, my Breaking Bad one. That's good. Yeah, man. Hey, hey, Ingi, do you want to, do you want to maybe take us into something on, on, um, your side, like with ComfyUI or After Effects, just kind of show us a few things.

Play from 0:36:55 8:88:88 0:36:55 And yeah, to Rory's point, I think like for both of us, like we, we haven't gone super deep and ComfyUI and maybe we haven't had to, right. But like, I think for a lot of people and, and maybe even including myself, like I'm, I'm really interested in the additional capability that this, this unlocks and maybe like what those things are and why somebody would want to be in here versus another node based system, right? Like what is the difference? Yeah. Yeah. Cause I think like a lot of businesses want ComfyUI, but they are not, they are not prepared for ComfyUI, They don't have anybody in here that, that knows it.

Play from 0:37:30 8:88:88 0:37:30 And then it's like, how long is this going to take to learn it and activate it? Yeah. Operate it even just like fix it if something breaks. So it's like, I mean, then there, then there's companies that are highly technical and it's just like, they slot right in, you know, like that's the thing. It's like, if you're highly technical, probably a lot of people who worked in CGI 3D, anything node based, you know, before flame nuke, whatever those, those programs were that a lot of people have sort of come up with, um, you know, Houdini, things like that. This looks like, this is like regular life, you know?

Play from 0:38:01 8:88:88 0:38:01 So it's like, you had to jump right in. But like, again, I think people get intimidated by the nodes and the, you know, the parameter sets and what's a seed and like, they get really confused, but it's also a way to, you know, you don't have to re -prompt or re -generate things. You're just, you're, you're dialing in a lot of instances, I'm assuming, which is, which is cool if you can engineer it, but I'll let you take over from here. Yeah. So I mean, I can, I can show you a couple of things in here. I mean, like I totally get why people are intimidated by, because it, it looks like an absolute nightmare or it can do, especially if you're, if you're disorganized like me, but I mean, this is, so this is essentially, this is where I make all my kind of graphics.

0:38:33How does Ingi create tracked AI graphics?

Play from 0:38:38 8:88:88 0:38:38 So like, this is a, you know, dots tracked onto from, or tracked from this person's face. So that's like my latest, the latest post. And, and I've got a node here where I just type in what I want to track. Okay. And then, and then that gets put into the, I've got another node that can control the colors of them. So like, you can either got presets or I can just put in like the hex codes for every color that I want to use. Um, and then, these are custom.

Play from 0:39:05 8:88:88 0:39:05 I am, I'm releasing it, but like, it's so difficult to use that. I don't want to release it until I figure out how to, how to give it to people in a simpler way. But essentially, it creates this thing called, called regions, which essentially is just the, like the dots in space, basically bboxes, which are, which are essentially the box around the thing that you're tracking. Mm -hmm. So here we've got like, you know, dots where the eyes and the nose and the mouth is and the ears.

Play from 0:39:35 8:88:88 0:39:35 But if you, if I, if let's say if I use this node here, which is, which is the box one. And so I can create corner squares, which are kind of like the sort of bounding box squares, like in Photoshop or whatever, or like you used to, or sort of like the stereotypical bounding box, rounded ones. You've got a regular box. You've got corners, which are like, kind of, almost like crosshairs. Um, so if I was to make a box and, you know, make the opacity one, then, then that would allow me to not just put the dots there, but I'd put a box around the thing that is tracking.

Play from 0:40:07 8:88:88 0:40:07 So then the mouth would have an entire box around it and the, the nose would have an entire box around it. Um, I'll show you just so here. It's a few. And then these are like all these little elements, like you've created these over time, right? Like, so like all these, and then basically you're, you're kind of using them as the raw materials for new things that you're building. Mm. Yeah. So like, so this, this shows like those layers that come out of that one. Right. So like, this is all tracked from that.

Play from 0:40:35 8:88:88 0:40:35 Turn that music off. So this was the base video and then all of the stuff on top basically is tracked. Yeah. Yeah. Adding on top. So sick. Yeah. I love this. And then this is your motion design background coming in here. Yeah. You're dreaming it up from that, from that skillset. Yeah. So then like, you know, I just, I just use as many tools as I possibly can get my hands on that can create images from another thing. And so like, you know, things like pose, pose rigs and like various other things.

Play from 0:41:03 8:88:88 0:41:03 So how can I create things from this that can then apply it to it? And then I mean that the, the, so the videos themselves, like, so this is an example of like, these are the start and end images and, and they're just little sections like this, just loads of little, little videos like that, where they go from one to the other. And that's another one. Okay. So I just, I just make all the images and then do little, they're like, you know, maybe like two seconds long and then they're all stitched together into one long one.

Play from 0:41:32 8:88:88 0:41:32 And then, and then I work on top, like I work on top of it, which I can show you here. So this is, this is it in After Effects. Look at that beauty. Oh, it's crashing. Too big. Yeah. Yeah. Riverside might be like, too much going on here. That's all right. It's a, it's just, it's surprised. I opened up. That's, that's all. On the ComfyUI side. So first off the videos you generated, you said that, that was Wan 2.2. That's the, that's the keyframe animations that you're using.

Play from 0:42:00 8:88:88 0:42:00 And then a lot of this stuff is, is all custom nodes that you have built for yourself. Now, let's just say Rory over here wanted to go build a custom node. Is this something you're doing with, you know, with Claude, Astra, things like that. I've built nodes for Replicate so I could bring them into tools like Weave. It's like with Docker process and things like that. Is it, is it similar or is it easier with ComfyUI? It's super, I mean, it's not probably not as easy as like, as doing it in like a, a, a, an app that maybe like has, has a built in.

0:42:05How do you build custom ComfyUI nodes?

Play from 0:42:33 8:88:88 0:42:33 Right. So you've got to, it's all like it's, it's code and like you've got to add it from somewhere. Right. Like you've got to add the folder and like, it's not quite kind of like drag and drop. I mean, it kind of is, but it is still pretty easy. Right. So if you want to, if you want to build a node for a ComfyUI, you really just ask Claude, just like, Hey, I want to build a custom node that can do this. You just got to know like what you want it to do. So I made one like yesterday that gives things like a glass refraction because like I saw some images on.

Play from 0:43:01 8:88:88 0:43:01 It's like a shader. Uh, kind of, I was, I was looking at images on Pinterest and like, they had like some cool like refractions on, on graphic stuff. I think that might be an image. That's why it's not a video. There you go. Oh yeah. So that node I made in like 10 minutes. Okay. Um, just by, by saying, Hey, like I want to make a node that, that does glass refractions. And it gave me like, there's like 15 options on there.

Play from 0:43:27 8:88:88 0:43:27 Uh, they're super easy to make. Um, you just gotta, you gotta really like, you gotta understand how ComfyUI works and, and be able to, uh, wield that. And obviously like, you know, manage a folder full of, full of Python scripts, but it's a lot scarier than what it actually is. Once you get into it. This is making me want to go in there now and play with, I never, I never really went down the custom node route. And now that I'm thinking about it, like this is now with, you know, Claude and Astra can do this stuff in like, he said like five minutes.

Play from 0:43:59 8:88:88 0:43:59 So we just have to like, really, that looks sick. I love that. Yeah. And then, but, but this is the, this is the thing of change the strength. I mean, you can create any parameters, right? So normally with, with sort of normal tools that you're using, you really only have the options that they give you. But with this is like, you can literally just decide what the options are. Or you could like, with this one, you could like combine two, right? So you could just copy, copy this and run it back through another one.

Play from 0:44:27 8:88:88 0:44:27 And you change the angle. So like make it 90 degrees instead of, instead of zero. And then you're going to have two on top of each other. It's just going to create some sort of absolute. Right. But like, you can, you could connect 15 of them all at different angles. Right. It's just really easy to make stuff. But like once you get your head around it and, and the fact that like, this is a note that I made yesterday with an LLM. And then this is the first time I tried this actually, it's kind of cool.

Play from 0:44:52 8:88:88 0:44:52 But it's like, I love it. It's the freedom to be able to do anything you want because it's open source, because it's like just sitting on a computer. Like no one, no one's controlling the backend of ComfyUI for you. Like that, that's, that's yours alone. So that if they update, you don't have to do the update, you know, like you can, you can freeze it. Like there's no kind of, uh, there's no big ComfyUI behind the scenes, like controlling what's on your computer because it's, it's all, it's all separate and you just downloaded it.

Play from 0:45:19 8:88:88 0:45:19 What about when you're, when you, when you're creating this, like, do you feel like you have a system now, like with these custom nodes where if you're using an LLM to help with this, like, you've got a really good way to contextualize what you're wanting to get the quality level that you're striving for? Um, yeah. So usually it's like, I mean, because I've made so many, like, I understand like what, what parameters I might need or whatever. But to be honest with you, like those, those, like those LLMs are really good at it.

Play from 0:45:47 8:88:88 0:45:47 Right. So like, um, you know, I didn't really have to specify anything on this. I got pretty much everything I wanted. Um, but then I could go back and go, look, I want to, I want an option where, you know, only every, every other one is refracted or like, you know, or there's colors in it or whatever. Like, um, but largely it's just like, you need to have a goal. Right. And my goal was like, I'm sure you've seen them. It was like, it was like black and white typography that had this like really cool kind of refracted, like slice thing through it.

Play from 0:46:16 8:88:88 0:46:16 And I just didn't know how you made it. So like, I just made a node that, that did a similar thing. Um, yeah. And then are you like, are you providing that as like a reference to then have it describe and further contextualize it? So then you have that raw material of what to, you know, kind of, uh, describe it both textually and from an image standpoint for data and context. Yeah. Yeah. So I'll, um, yeah. I mean, now with, with the agent stuff, like I'll, I'll just make a folder and put all the references of the thing that I'm trying to achieve into that.

Play from 0:46:46 8:88:88 0:46:46 And then I'll just, I'll just, I'll just, I'll just say, Hey, like, let's make a custom ComfyUI that, that can, that can generate that using this, this and this. Um, but yeah, they're really good. I mean, like they, they, I guess the, the beauty of open source code is that, um, the models are now already, already trained on it. So they understand all of the bits because they've even got like the sort of underlying ComfyUI code built in. Huh. Have you, have you ever dabbled in the fine tuning of models and things like that? Is that a.

0:47:11How does Ingi combine fine-tuned AI models?

Play from 0:47:17 8:88:88 0:47:17 Loads. Yeah. So all, all of this stuff is, is, um, so these, so I'll show you up here. This is like, so these things, so these are like, for example, two LoRAs that I trained. Um, one on the left is using like a sort of not, not full strength, um, um, what they call thermal imaging LoRA. So like some of the characters coming through and then this one that the other one is like, uh, is more of a, like a studio photography LoRA, but then combine it.

Play from 0:47:47 8:88:88 0:47:47 And you just get this like weird kind of otherworldly glow. Um, which is what then went into like making these ones, right? Like, like they're kind of, they're like 25% thermal imaging, but that's why they've got that weird lighting and the, and the weird kind of almost no shadows on the face. Is this, is this flux or stable diffusion or. This is, uh, this is, uh, Krea 2. Krea 2. Oh, okay.

Play from 0:48:14 8:88:88 0:48:14 Interesting. Love to get into that conversation a little bit too, because we were, we were talking about this about two weeks ago that we hadn't really did gone deep into Krea 2, but I've loved that they open source stuff. So you could do. When, and Krea 3 is coming out. Yeah. Uh, so it's going to be an edit model. So that that's, that's going to be. Oh, I didn't know that. Did they, did they say anything about a rough timeline on that? When would you expect that?

Play from 0:48:40 8:88:88 0:48:40 Um, no, I don't think so. Um, yeah. Yeah. I was at that. I was at their launch event, launch event in New York, um, a couple of weeks ago for it. But, um, they, I don't think they put a timeline on it. No, I, I'm guessing, you know, as, as with all, as with all model companies, they're probably still training it. Yeah. I feel like we're going to get a, uh, influx of stuff here. Like now through November, you know, before Q4, like we're just going to get like just bombarded with all these updates, you know, image 2.5 from GPT, Krea.

Play from 0:49:15 8:88:88 0:49:15 But yeah, that thing is interesting. So prior, how long have you been using Krea as sort of like that, that tool in that sense, maybe as opposed to something like Midjourney, you said you were using Midjourney. Like, is that still a place that you would go to maybe at least create sort of like initial creative concept thing or images to start from or? Um, it depends. Like, like, I mean, mostly recently I've been using, um, GPT 2.5. It's just, it's just so unbelievably good.

Play from 0:49:45 8:88:88 0:49:45 Um, it, I mean, I, I bounce around it's you know, it's kind of, kind of like the, um, the styles or the tools or whatever. When I hone in on something really interesting, then I'll, I'll use that for a while. And then I'll, something new will come out and I'll, I'll test that out. Um, I want to have a, a decent amount of understanding of like most models, but yeah, GPT 2.5 at the moment. I mean, it's, it's so incredibly good at like, you know, human anatomy and like, I've really been pushing.

Play from 0:50:11 8:88:88 0:50:11 into things where like, I'm trying to put like, you know, a hundred, a hundred people into an image and they're all known. You know, they're all people that I've given the model to use. Um, and can we do lots of images where they're, they're, you know, doing different things like the one where the, the woman's being like pushed up by like a whole, like flood of people. Like every single person that came from character sheet. And so, uh, when something new comes out, is like pushing like the, some, like whatever the, the, whatever the new skill of that model, I really want to try and push.

Play from 0:50:45 8:88:88 0:50:45 Like with Krea 2, it's like, you know, it handles style like better than any model, right? Like, and, and I, I wouldn't necessarily go to GPT for like a specific, like illustration style necessarily. But for Krea, it's great for like evolving new styles, kind of like the, the, you know, the thermal imaging and, and, and studio ones mixed. Um, so really good for that. And the training is, is, is incredible. Whereas obviously you can't, you can't train GPT. That's what I was going to ask next was on the training side.

0:51:09How should you train a style LoRA?

Play from 0:51:12 8:88:88 0:51:12 Cause I think there's probably a lot of people in here who have either one trained LoRAs and failed or two, like don't necessarily know the right strategy to do it. Like for someone like Krea, where are you, where are you training? And then like, if you're doing, let's just isolate it to a style. Like how are you, how are you going about that in your mind? And like, like pre -work before you even get into it? Um, I mean, first is like deciding what the style is. Um, I mean, usually for me it would be, it will come out of a need, right?

Play from 0:51:39 8:88:88 0:51:39 Like if I'm, if I'm making a lot of something at that time and like, and I'm, I'm constantly having to do this thing. Like, like for example, the, uh, you know, I've done a lot of the less sort of like 3D scanning, Gaussian splat looking stuff. Um, I did try and train a LoRA on it and like, it just, it came out spectacularly wrong, but, um, but I don't think I made enough, um, made enough training data, but like, that's normally kind of where the, where the, the need for it comes, comes from.

Play from 0:52:09 8:88:88 0:52:09 Or, um, when, um, when Qwen Edit came out, which was like the first sort of open source edit model, um, I just needed to just like try and make like every, every like different effect that I could think of. Like I, I, I'd made a, um, I'd made an inflation LoRA for, um, for SDXL back in the day. And, um, so that was the first thing I did with Wan. I did an inflation thing with that.

Play from 0:52:34 8:88:88 0:52:34 And then when Qwen Edit came out, I did an inflation thing on that one. And then I did that Breaking Bad clip. And then the, and then the, uh, this is America one using that, but like, yeah, it's normally like, uh, the first step is obviously coming up with, with what you want it to be and, and being really specific about it. And then it's like, you know, having as much good training data as possible. And then usually a really big part of, especially for video models is, is really nailing down on like what the best captioning strategy is.

Play from 0:53:06 8:88:88 0:53:06 So you've got to like, uh, cause you're working with what the model already knows and what the model has already done. Um, so, um, with like a lot of the newer models, like the, you know, SDXL or SD 1.5, like they were probably trained with like, you know, man walking down the street, cinematic. Yeah. Like that was probably the extent of like the, the, the captions that they were being trained on. But now it's like essays, you know?

Play from 0:53:33 8:88:88 0:53:33 So you're kind of, you're battling that a little bit where like you have to, you have to come up with the captioning strategy that there is no rule book for. You know, so sometimes, um, like I've trained LoRAs where I didn't even caption them at all. Right. Just didn't include anything. So like, and that worked okay. Or I've done some where, you know, you've, you've given them an, given an essay for some images and then like two words for other images. And, and sometimes you train them several times.

Play from 0:53:59 8:88:88 0:53:59 So like you'll train one batch on, so for example, with video models, there's a trick where you, where you train for, you know, 2000 steps on really low res videos. And then you train for like 500 steps on really high res ones. And, uh, you get like much higher resolution results, um, from doing that mix. Um, so yeah, this is like loads of, loads of things to consider.

Play from 0:54:24 8:88:88 0:54:24 And, and, um, you know, there's, there's, there's easy ways of doing it. Like kind of using the sort of built -in ones like Krea's, Krea's, um, LoRA trainers and absolutely second to none. Like I, I, I've trained that LoRA, like locally on my machine and I've trained the same, same LoRA on their website. And it was like a hundred times better at like a quarter of the steps. I do not know what, what they did differently on their side, but obviously it's their model.

Play from 0:54:49 8:88:88 0:54:49 So they know, they know it better than anyone else. Can you, can you take the models off of Krea now when you do it on their actual, that was going to say, cause I remember that was the, I was like, wait, well, hold on. You can, you can train it on their site, then, then, then it's theirs. Yeah. Um, but as I say, like the quality is, is that much better on that, but probably because I haven't trained enough on them. On like locally. This is a, this is a technical question for me because I've always been interested in the, in the video training on how that could work, would work. And I feel like what you're doing here, it makes a lot of sense for like very specific task is like what I need the video LoRA for.

0:55:13What should a video LoRA learn?

Play from 0:55:26 8:88:88 0:55:26 Like how, if I, like, if I wanted to do something like, let's just say the Gaussian splatting look side of things. Like how many videos are you, are you looking at like as a data set? It does it matter if the quality is good enough or you want it to be that specific or do you need more range to like push it? Cause I understand, I feel like I understand what you need from an image side, the video side, like my head just like, oh, it makes no sense. Well, first of all, I wouldn't, I would never train, I would not advise you to train, uh, today in today's world, I wouldn't advise you to train a video LoRA on a style.

Play from 0:56:02 8:88:88 0:56:02 Okay. Um, or, you know, unless it's like a style that the model really struggles with. But, um, as I said, that that's why I use Wan so much because, um, uh, the, the combination of a, of a capable video model and a brilliant image model gives you one, you know, gives you the best video model in the world, essentially. Because the better the input that you give to something like Wan that, you know, the, it's going to keep that level of level of, um, detail.

Play from 0:56:35 8:88:88 0:56:35 And so with like GPT 2.5 and Wan put together, like the results are incredible, meaning that like you, you don't, you wouldn't ever have to have to train anything to get that look. But what you do train for is like the movement. So like, or for example, like I, I've spent the last, must be like nearly a year and a half now, like training transition models. So like all of the transitions you see, like they're all using my own models.

Play from 0:57:00 8:88:88 0:57:00 And there is, and there, yeah, they're, they're like they're Wan video models, um, trained to do transitions, which is incredibly hard thing to train because, um, newer models allow you to do context training. So you go, if I give you these two images and these videos resulting videos, that's the thing you need to do. And so learn from, from those examples. So like, oh, okay. So when I go between this image and this image, it should result in something like this, but you can't do that with Wan.

Play from 0:57:29 8:88:88 0:57:29 Wan can only have one video, like one video as an input or like one at a time, meaning that you have to teach it to look out for the transition in the video through the caption. Got it. That's quite like, oh, hey, this video, and you can't just tell it like an LLM. You've got to be like, okay, you've got to be really simple on the description of what's going on in it. You've got to be really detailed on the description of the thing that happens in between.

Play from 0:57:55 8:88:88 0:57:55 Or you don't describe that at all, and you describe everything else in a really detailed way, so that the model learns that when this trigger word is used, it means everything that isn't named in the captions. So it's, it's like, it's, it's, it's super kind of tricky to get right. And it did take me like six months to get a working version of it. I can't even remember what the, what the, what the working version actually used, but, um, I think I might just use really simple captions or something, or maybe I trained it a few times, but, um, yeah, it works pretty well now.

Play from 0:58:31 8:88:88 0:58:31 Cool. I love this stuff, man. I just, I love when, I love when you get to go and really put your own time. I mean, if you weren't doing looping as a style or this more so keyframe transition, would you even have a need for that transition, LoRA? You know what I mean? Like it's because I feel like it's because you, you, that's like your vision. Now you need to go build that. Right. Yeah. I, I, I, I mean, the other thing that, um, the other reason why I work in the way that I do is that, um, I just really like having control.

0:58:50How much creative control should you keep?

Play from 0:59:00 8:88:88 0:59:00 And so I can go and prompt a 30 second video, but I'm going to be like, oh, I didn't quite like that bit. Or like, you know, I'm not, that wasn't the right angle or whatever it might be. Um, and also like, I think it's almost like too, it's too long a duration for my like brain to feel like I made all the decisions or like, or, or maybe like consolidated what came out because like, you know, I could generate something while we're chatting now or whatever.

Play from 0:59:29 8:88:88 0:59:29 I wouldn't post it because no matter how good it is or whatever, like it's, it wasn't mine. Right. Or I wouldn't feel like it was mine because it wasn't my idea. And so, and so with, with the way that I'm working there and especially with the transitions, it means that I'm always thinking in terms of images. So like, uh, everything I do is a sequence of images, whether it's like different person in every frame and then it morphs between them, or it's like the different poses of the thing, the person's doing.

Play from 0:59:55 8:88:88 0:59:55 Um, so I'm thinking, thinking it through, I guess, almost like a storyboard or like, or like how a choreographer might think. Um, which yeah, makes it make more sense to me, but that's probably because I come from animation where that's kind of how you have to work. Yeah. It's cool. It's very cool. What, what stuck out to me is just the creative control thing. Like the, the thread that we've been kind of talking about in terms of the tech that continues to evolve and get better.

Play from 1:00:21 8:88:88 1:00:21 And then there's this sort of threshold that exists where it's like, I'm kind of giving AI too much now where it kind of feels like I'm losing. I'm losing either control or direction or input from me that makes me feel more connected to it and more valuable. I think intrinsically, but also like that you'd want to share it and we're, we're not slowing down here. So it's like, there's going to be things that AI, right?

Play from 1:00:50 8:88:88 1:00:50 Like we were sitting here a year and a half ago, year ago, we were like the anatomy of these videos was still fucked. You know what I mean? Like you still couldn't get a lot of the physics to, you know, to operate the way that they would in real life. But we, we saw the promise of it. And so I think like that, that piece about like 30 second length, right? Like you're going to continue to see that push for longer videos, but is that really what the majority of people want?

1:01:04Do AI videos need longer generations?

Play from 1:01:16 8:88:88 1:01:16 And I think like, yeah, in some ways, right, depending on what your role is in the creative element, filmmaker versus designer, whatever, right? Like that's more attractive. But at the same time of like creating in general, you brought up that really good point about like, if it's a certain length and there's so many decisions that have made, did I really like impact that decision at a level that I feel responsible for, but also like I did something.

Play from 1:01:41 8:88:88 1:01:41 And I think like we have this internal need to feel satisfied that we helped build it. We got so used to building the whole thing and now we've, we've sort of like pushed the tech forward of like, how far can this go? And then where's the middle ground that we actually really like and would want to do, even if that means at the expense of speed or cost, like where do we take these decisions back?

Play from 1:02:07 8:88:88 1:02:07 You know, like, I don't know. That's been a theme I think we've, we've kind of been talking about here over the last few months. And I think it's especially true here and something I'm thinking about basically daily now, you know, with anything like that. Yeah, it's a really interesting one. I think, I think there's, I think there's different schools of thought and I think they all exist in different vacuums, right? Like some, I think some people are looking at, like, I think a lot of people were excited about longer durations.

Play from 1:02:33 8:88:88 1:02:33 Like, you know, like Wan is basically restricted to 81 frames. You can make it longer, but like, you know, there's kind of tricks you have to do or, or, or, or it doesn't kind of hold context very well or whatever. And for a while that was like, well, that's like, I always said, that's all you need. Like, you know, how many, how many movies do you see with shots longer than five seconds? Like people don't have the attention span for that. People are like, I want to make 10 minute long videos. Like with the single shot.

Play from 1:02:59 8:88:88 1:02:59 I'm like, no one will watch it. I promise you like Birdman was like a cinematic event. It wasn't the norm, right? Like no one wants to watch a shot that's that long on a, in one go, unless you're watching a podcast maybe. Right. Um, but there's no need for like those types of durations. But I, I, I think the other thing there is like, it's a little bit like, if you think of it in terms of stills, right? If you ask a photographer, why just take one photo when your, your camera could take a million photos at once?

Play from 1:03:29 8:88:88 1:03:29 And it's like, why would you want that? And then, so it's a little bit like, uh, I think there's, there's a school of thought there where I think a lot of people that were excited about the longer durations are potentially people who are looking to automate things. And like, and looking to automate like entire YouTube series or like, you know, looking at it from like the very like technical standpoint. Which I think is perfectly valid. Right. And it's like, I think a lot of the people that came into AI from crypto would be probably be thinking about that.

Play from 1:03:56 8:88:88 1:03:56 Whereas like, how can we use this technology to like build something and make something with it? But from an artist perspective is like, you're just going to make it all up from that first image I gave you. You know what I mean? Like I need control of every bit. And so there's less of an urgency to have that longer duration, but I think the use cases are there for like the different ones. But I think maybe the issue is that the model makers and platforms, um, I think they're building for the, the, the sort of the, the social media, uh, uh, content makers that, you know, uh, love using the technology to use the technology.

Play from 1:04:36 8:88:88 1:04:36 I don't think they're necessarily thinking about the professionals too much because I think it's just a smaller market as well. Like, you know, it's not, it's like the, you think about, you know, like the amount of money I spend on like AI just to like make my own stuff. Like, I think I spent a hundred pounds yesterday, just, just on Astra, just to make some images. Like it's, it's, you know, it, there's an unreasonable amount of money to spend that people are doing that for like vibe coding and all these other things So you think actually the market for that.

Play from 1:05:07 8:88:88 1:05:07 Like, you know, you're selling to designers, like people pay Adobe 50 bucks a month. Like that's as much as they want to pay. And so like the, the, the size of the market and the fresh professional world just isn't anything compared to, to what, what it is for people who just play with it and post stuff on socials and have fun with it. Interesting. I mean, it's, it's an interesting sort of like conversation around that. Cause I feel like we've, we've also said this many times. There's a much different, much different perception between hobbyists and the people that have to use it professionally.

Play from 1:05:36 8:88:88 1:05:36 Like it's just a totally different vibe, especially ComfyUI. I got, I mean, do you know many like hobbyist ComfyUI users? Yeah, I do. I do. Yeah. Sorry. I meant to say like, that feels like the hobbyist platform. I guess. I'm sorry. I mean like hobbyist, like, you know, Magnific or something like that. I feel like you use it for work and maybe use it to generate some images of your friends. If you want to generate images, you use, you know, Gemini or ChatGPT, whatever your subscription, like ComfyUI feels like there's hobbyists.

Play from 1:06:07 8:88:88 1:06:07 Other AI tools feel like it's just, you have to use it or you don't use it at all. If that makes sense. But yeah, I think I kind of see it the opposite way around where like ComfyUI is like, you know, ComfyUI is the, like the tool that like ILM should be using. It probably is using, right? Like Netflix is using that, like, you know, the, the, the, the, the people in like the depths of the most technical like VFX and, and, and all that, like it's absolutely the tool for them.

Play from 1:06:36 8:88:88 1:06:36 Um, and then all of the stuff in between or a lot of the stuff in between, I think like it's probably overkill, right? Or might be overkill, um, for certain things. So like all of those tools are very, very capable. Um, but, um, it does attract a lot of people who just do it for fun, like, because it's hard. Right. And I like, and that, that, you know, you go on, you go on Reddit. Or like, there's lots of, lots of discords and stuff where like people are not trying to do this for work and they don't want to do it for work.

Play from 1:07:06 8:88:88 1:07:06 They just love it. Right. Like they just like, it's just, it was, it was a, it was a Lego car, you know, like it's the hardest Lego you can get, you know, and like no one else can do this sort of, and like, oh, I would never use mid -gen because it's too easy. You know, like they want it to be hard, you know? Yeah. And I love that. That's why I got into it. Intentionally challenged. Yeah. Yeah. Why do something? That's true. Do you think there's a, do you think the future with open source is different than like, do you think there's like a different, like, what do you see as like the future of open source, like moving forward? I feel like there's been a lot of discourse around it and sort of like seeing a little bit of a narrative pop up as of late.

1:07:28How can open-source AI sustain itself?

Play from 1:07:44 8:88:88 1:07:44 So I'm curious, what do you think on the open source side of stuff? So I think my view, my view has been on open source has probably always been a little bit different to like a lot of the open source community. Maybe like, I think part of that because, um, you know, I was on the, I was on the company side and in, in the NFT space, you know, in the NFT space being, you know, a very like, uh, a very vibrant space with a lot of people involved where the companies are kind of seen in a very specific way because people have spent their money on them. Right. And like, and, and then coming into the AI space where there was a similar kind of world where like someone, like people will spend like 10 grand on a PC so they don't have to pay a 29.99 subscription.

Play from 1:08:31 8:88:88 1:08:31 Like, it's insane. Right. Like, like there'd be like, like, be like, yeah, I would never, I would never like, like, you know, or like people stop, people stop open sourcing and they're just like that company's the devil, they're greedy or whatever. I'm like, if they got to make money, you know? And so my, my, my, my view on it has always been, I want companies to open source, but I want them to have a plan. Like, and, and I, I think a lot of it was kind of, a lot of the open sourcing has been done kind of because people felt like they should, which is great, but there needs to be like a financial plan at the end of it because otherwise you can't continue to open source and it to make sense.

Play from 1:09:11 8:88:88 1:09:11 And so I've always been very keen to work with these companies to like figure out like, well, okay. So when we, when we then want to use your model commercially, like what's an easy way of doing that? You know, what's a, what's a, what's a better system to, to, to make it easy for everyone so that we're incentivized to use your models to build things. But then when it actually comes to like making a professional, we just turn on the tap and it goes through your inference, you know? And I think, I think that, that is where the, that is what the future of, of, of open source, I think would be or should be is that, you know, you, you give away the kind of like the light version, like, like, um, my first copy of Photoshop came on the front of a magazine, right? Like it was on a DVD or CD on a magazine or whatever.

Play from 1:09:53 8:88:88 1:09:53 And it's like, I got this free version of Photoshop and like, now I pay like, you know, God knows how much every month. And it's like that, that's what open sourcing should be is like, yeah, open source the thing. No one's got a data center to put it on. Right. And, and put some, put some commercial terms on it. So, you know, Amazon doesn't steal it or whatever. And then, and then, uh, incentivize people to, to do a partnership with you when they want to use it, use it.

Play from 1:10:18 8:88:88 1:10:18 Maybe not even commercially, maybe, maybe, you know, there's ways of letting people use it commercially, but if you want to use it at a larger scale, just do it through them. Right. Like, and, and, and to have that, to have that kind of two way street, right? Because I think the problem, the problem for a lot of the sort of early open sourcing was just like, we're open sourcing because, you know, down with the system, you know, and like that, that's not a financial plan.

Play from 1:10:44 8:88:88 1:10:44 And, and, and, and therefore like a lot of companies went under or like, or decided that it wasn't, or their sort of higher ups decided it wasn't worth it. Um, and so really hope, hoping to see better business models around that in the future. I, do you guys remember when they used to mail those CDs like to you, like in the mail? I remember like AOL, you got free minutes or something like that. You remember that shit? That was wild, wild times.

Play from 1:11:11 8:88:88 1:11:11 We've come a long way. I, yeah. Okay. And DVDs in the mail? before Netflix. Yeah. Yes. I mean, yeah, let's, I mean, we were, we were, we were kind of talking about team blockbuster a few episodes back, you know, like, yeah, but I, okay. Ingi, I want to get it. I want to get into a little bit of a pivot. I want to talk about Systems for a second too. And I think like you brought up the Doodles thing. I don't, I don't think we went deep there, but like, I'm really curious about that transition from sort of like NFT world into AI and like, were there, are you seeing, any lessons or things that you've applied by being in that space early and then kind of being in this space early?

1:11:31What did Doodles and Web3 teach him about AI?

Play from 1:11:53 8:88:88 1:11:53 And it looks like maybe you've even did some work here with Doodles on since you've been at Systems too, right? Is this, were you part of that? Like, I guess that toy, the toy thing that they're doing too with that? Is that sort of what I'm gathering here with this project? Like you kind of built the, the model or the LoRA for that to produce those images in the doodle style? Well, I don't know exactly how that, so they did a, they did a partnership with Krea as well.

Play from 1:12:21 8:88:88 1:12:21 So they, they've got a Krea 2 model, um, trained on it as well. So I trained the, the first version of the, of the Doodles model. So the one that's shown in that, um, which I think is still the one you can use on that platform. Um, but yeah, the, the, the, the, the toys are kind of like a permutation of that, of that project in general. Like it might, it might be, yeah, I'm not, cause the, the toys themselves needed to, needed to be a very, like needed specific things like, um, you know, couldn't have floating eyes, things like that.

Play from 1:12:49 8:88:88 1:12:49 Um, so obviously they need to be kind of suited to being able to, um, 3D print. But yeah, I mean like, yeah, so back to your question on the, on the, on the transition and the kind of the two spaces. So like, I mean, I was kind of starting out in the AI space before the Doodles acquisition of, of Golden Wolf, um, and was very close to setting up an AI company in 2022 that just didn't end up happening because of, because of the acquisition probably would have been way too early for what we were doing, but it was very similar to what Systems does now.

Play from 1:13:21 8:88:88 1:13:21 Um, I feel like the experience of being in the, in the Web3 space, um, I mean, absolutely helped the transition into AI. Like first of all, a lot of people just moved over, you know, like, like a huge amount of people in AI, you know, were in, in, in, in web three before. Um, I think that like, just understanding like how like the communities work and like, you know, how, how, like, you know, this huge, you know, big transition from like, I guess like COVID, you know, create this kind of like, you know, the more of this online community and like people, people really valuing like who they interact with every day and all that.

Play from 1:14:02 8:88:88 1:14:02 Like, and you know, that, that's one thing that Web3 has in spades, right? It's like, you know, it's an echo chamber, but it really does feel like you're just there with your friends. Right. And like grassroots, like feels very organic. Yeah. Yeah. Yeah. So the, the, and there's, and there, and, and the AI space feels similar to that, but I think had I not, had I not worked in the, in, in the NFT space, like I wouldn't, wouldn't really have had a taste of that. Um, you know, I really understood that the value of that and the power of that.

Play from 1:14:29 8:88:88 1:14:29 Um, and so, yeah, I mean, but yeah, and to be honest, there's a lot of parallels, I think between the two, like, you know, it's very, very, very much focused on technology. There's, there's kind of the, you know, the, there's like the hardcore technologists and there's the people that are there for like, you know, the money, you know, it's exactly the same basically on both ends. Right. It's like people that are just trying to, trying to, trying to make that, get that X payment every two weeks.

Play from 1:14:54 8:88:88 1:14:54 Right. I don't know. I don't know anything about those. Yeah. No, no, never. I've, I think I got to the X payment for the five, for like the 5 million views or whatever in the three months span. I think I got to like 4,999,998 or something like that, whatever it was, it was like two, it was like two impressions off. And I was like, you know, I'm never trying for this again. Like, I don't know. I, I, I, I can't do it. I, I got on it like, like a month ago, I think.

1:15:01How does viral content connect to X payouts?

Play from 1:15:22 8:88:88 1:15:22 Really? And it, it was like, it was like, because one video went viral, I think it got like 3 million views or something. And then I just posted like every fucking day until. You gotta go. That's what it's like. Yeah. You need that. You need that big one. Yeah. And, and yeah, yeah. I wouldn't, I wouldn't have, wouldn't have got it without that, but, and they don't count like, I mean, I, I don't spend a whole lot of time like commenting or anything like that, but, um, they don't count that anymore. Like it's all, and it's all, um, verified views now.

Play from 1:15:50 8:88:88 1:15:50 Yeah. I saw people used to game the system just by like turning on post notifications for Elon and trying to comment like in the first five seconds so that you could get just like automatic impressions from him. And then it was like, that was their way of monetizing. Like I saw people, it's like their whole strategy was that they like, they would like put it out there too. It's funny. I just, I, I feel bad. I feel bad for those. What is it? Is there a 3D model guy? Cause he, he won't be getting anything now because all of his comments, it's all comments.

Play from 1:16:19 8:88:88 1:16:19 I've been, I've been waiting for him to like, I've been waiting for this moment for him to get like, to realize that like, Hey, it is a 3D model now. It just wasn't modeled by hand. So do you know this guy? I don't, who are you talking about? It's, it's this troll. He's the, he's like the quarterback of the AI trolls. Yeah. It's called, we don't have to put him, pull him up on here.

Play from 1:16:47 8:88:88 1:16:47 Yeah. Yeah. I was going to say, let's not give him the, uh, satisfaction. It's like, is this a 3D model? And it's basically any AI video from the last two years he shows up on in some way, shape or form. And then just like spits it. But he's got like these little minions now that are like, that like tag him and stuff. And like, and he's clearly bossed, it clearly bought it as well because the second he comments has got like 20,000 likes and it's like, post has seven likes.

Play from 1:17:14 8:88:88 1:17:14 It's like, it doesn't make any sense. No way. Um, what's funny to me about that is that like, I've, I've had a few people, I don't, I don't get many, I don't get many of those comments anymore, but, um, I've had a few people point out the sort of Gaussian splatting 3D scan stuff and they're like, Oh, you made this with AI. You're not, you should just learn how to do Gaussian splats. I'm like, you know, that's AI, right? Yes. That's literally what that's what it is. But like, I just, I just did a shortcut to it.

1:17:42AI critics keep moving the goalposts

Play from 1:17:43 8:88:88 1:17:43 I mean, like you can find something to complain about when you saw it with like the transition of everything with the video over time where it was like, Oh, like everything is slow motion. It doesn't even look real. The skin looks plastic. The hands are bad. Then it's like, Oh, well, it's not long enough. Oh, well, they can't talk. Oh, you know, it's like every iteration, it just gets fixed. They have to find something else. Yeah. They evolve their excuses. They just keep moving the goalposts. It's the same thing. Like these people cannot be real. I mean, who, who's just getting on there to just be that miserable every day?

Play from 1:18:11 8:88:88 1:18:11 There's a bunch. LinkedIn, definitely. They, they didn't, didn't, didn't X figure out that like that, like a substantial portion of it was like foreign propaganda. Yeah. Um. Big portion. Yeah. Which like, which, which makes sense. A lot. I mean, to be honest, a lot of it was like, was like, um, engagement farming for money as well. I think there are a lot of genuine people in it, but I think that they're obviously, they feel like they're part of a movement because there are so many bots agreeing with them.

Play from 1:18:44 8:88:88 1:18:44 Dude. It'd be, that'd be so like disappointing. I mean, dead internet theory, but like at the same time, it'd be so disappointing if you found out that like everyone you interacted with, like actually didn't exist. You were just like there by yourself creating to a void. I feel, I feel, I feel like I'm relatively good at spotting the AI comments, but obviously they're getting more and more sophisticated, but like, and they're, they're always nice these days. Right. Like they're like, Oh, I really liked that. And then they just like repeat your description or something. Oh yeah.

Play from 1:19:12 8:88:88 1:19:12 That's it. If you go on their feeds, we go on their replies, they'll just have a reply every second. And so it's, it's, it's basically a bot like, yeah. And it's gotta be, but, um, I mean, that's, that's kind of what I like, what I like about having just like, like my, the, the titles of my videos are always like artsy, whatever, but like that, that they're like, they relate to something, whether it's the theme and the thing, or maybe like a current event or whatever.

1:19:12How do bots expose themselves in replies?

Play from 1:19:38 8:88:88 1:19:38 Like, uh, there was one I did the other day that was like, not so fast or whatever. And it was like this woman, like wearing like a duvet, like, like basically trying to stay in bed and she keeps trying to get up and the duvet gets her back or whatever. And the caption was not so fast as like, don't get out of bed. But it was, it was like the day they announced they were going to do the slowdown in AI or whatever. Like, uh, I always wanted to have like, like, you know, not, not I don't really care if people like notice it or not, or get it, but like, I try and kind of tie it in with something current.

Play from 1:20:05 8:88:88 1:20:05 But the funny thing about those and what I kind of like is that then I get the bots replying to what I said when it doesn't have a direct, like link to the thing. Right. And that it was one I said, like, prove you're human or something, you know, just like this sort of the CAPTCHA thing. And there was a few that were like, I'm definitely a human. That's great. Yeah.

Play from 1:20:31 8:88:88 1:20:31 That's good. It's like, yeah, you put like, you know, not so fast that it's probably like, well, how do you think it could get faster? You know, like, yeah, it's always a question most of the time. But yeah, it's funny. I've seen, I've seen some TikToks where people go, um, okay, so no one reply and all the comments are going to be bots. And then like, I mean, look at the comments. They're like, oh, I love this video. Yeah. There's, there was like that story a while back.

Play from 1:20:57 8:88:88 1:20:57 Cause dude was putting in his resumes, like hiding it in like, I don't know, like four, like point four font. It was like, yeah, it was like, if you're a bot reading this, I'm like the perfect candidate or whatever. Just like if, if it was LLMs reading it, like, I love people, you know, just love it. People are putting that in like LinkedIn, like, uh, about me sections too. So like when people scrape it, it has some weird thing that comes in it. So anytime they get an email, they're like, oh yeah, like already know. Did you, there was a thing, there was a thing not too long ago where someone tweeted something or no did a LinkedIn thing and it was like, I'm just posting this here.

Play from 1:21:30 8:88:88 1:21:30 Like, and he just wrote like a one sentence thing about himself. And it's like, what's that? And then he came back a month later and it was like, go on Google. That's like the top, top hit. And it was like, he just did it on purpose. It's like, like, I'm just posting this for this. And then it just became the top hit on Google or whatever. It's, it's crazy now. Like, I don't, I still don't necessarily technically understand how the AIEO or AI, AIO. AEO.

1:21:48How does AI search surface a local business?

Play from 1:21:57 8:88:88 1:21:57 Yeah. You sound like, what is that? The like old McDonald? I sing the ABCs. Yeah. Yeah. Do it on McDonald. Do I sing it every night when I put my daughter to bed? So it's constantly in my head. But the, I found out we had to get a notary a public notary for something. Maybe I mentioned this on you. Maybe I haven't. It was like, I typed in, you know, need a notary that comes to my house in Brooklyn. And this guy came up and he, he came over. I was like, dude, good job on the, good job on like the AI sort of like optimization for your business.

Play from 1:22:27 8:88:88 1:22:27 He goes, what do you mean? I was like, I searched in GPT or Claude and you were the first one that came up. I obviously just clicked on your link and hired you. And he goes, no wonder I've been getting so much business lately. No clue as to why. Oblivious. It's just, I think, I think his headline is just like traveling notary in Williamsburg and Greenpoint. Welcome to your house. Like, that's it. And it's like nothing, it's just like a schedule page.

Play from 1:22:54 8:88:88 1:22:54 Nothing else. It's like people that there's a thing, isn't there a thing where like you, you put reviews on the bottom of your website or something where people were saying things about your company and then they feed that into the thing. So people say this about these guys. Oh, that's a good one. Yeah. I like that. I like any sort of hack like this. This all goes back to like some sort of like micro optimization that feels like you're winning. It feels like you know something that no one else knows and you want to exploit.

Play from 1:23:22 8:88:88 1:23:22 Yeah, absolutely. I keep trying to, I have tried to do that like with Instagram where I'm just like, sometimes I get like absolutely no traction. Sometimes it just blows up for no reason. I remember I was like, what is it? The description, whatever. And, and, and, you know, LLMs keep telling me like, well, you know, you just list your tools. You don't really like say anything or whatever. Like, so I'm like, okay, like write me like a really good caption for this thing. It makes no, maybe it makes absolutely no difference.

Play from 1:23:49 8:88:88 1:23:49 And then I'd be like, it's gotta, there's gotta be a secret to this. And then I like, and then there's the account I'll follow that doesn't write anything and like blows up every time. I'm like, nah, I give up. It's just, it's just a time of day thing or something. Oh, I got, I got something that would blow you up immediately. Share, say that you're going to share your workflow and ComfyUI, have them comment something. And I guarantee, I guarantee that will go. Yeah. Because everybody, you know, like the Instagram play is just ManyChat.

1:24:04Why do comment-for-workflow posts spread?

Play from 1:24:17 8:88:88 1:24:17 It's, it's the automation of comment this, you get this. If somebody deems it valuable enough, it blows up every time. And it's just like, I don't know, man, I can't do it. That's why I could never be like this content creator that just does this shit because it's like, that feels so disgusting to me to do over and over and over again. But I get it. I get it. I, you know, because once you also have the following. Yeah. Yeah. Yeah. I respect it, but I, I, I personally couldn't do it.

Play from 1:24:46 8:88:88 1:24:46 Like I couldn't be a phone sales, sales person either. Like I can't be calling people up all the time. I can't be posting things like that. I just cringe me out too much. But absolutely respect people that can, because that's exactly how you should do it. And to be honest, like I really respect people who can trigger the, the, the AI hate in that, in that post when they're like, literally will say something like this is going to destroy this industry or whatever. Like, you know, like, cause you know, it works because it triggers people.

Play from 1:25:14 8:88:88 1:25:14 Suck it. I can't bring myself to do that. Yeah. You got to have a thick skin for that too. You like got to know what you're inviting and mentally prepare your, yourself and your armor for that. Cause you already know, but you're right. Like social, these algorithms love the discrepancy and the polarization of people going back and forth. Okay. Rory, we should probably wrap this. I want, maybe we'll, let's, let's do one more question each here.

Play from 1:25:39 8:88:88 1:25:39 Let's, let's just kind of dig in. Let me ask you this. You've been doing this for a while, 20 something years. I mean, obviously like evolved, adapted, it's a long span of time doing creative work and all different sort of surfaces and models and tools and this and that and the other. But if you were 20 years old again, assuming you're 40 something, um, what would you spend the next few years doing? Doing. If you had the tools that you had now.

1:25:42What should young creatives learn now?

Play from 1:26:06 8:88:88 1:26:06 I'll be doing. I'll be spending, spend the next few years doing. Um, I think now more than ever, like the, the key is not what to do, but to do something right. Like, you know, to make something, to make thing like to constantly explore and try, try things out. When I was growing up, it was like, you know, which, which one's the right degree to do or like, you know, which one's the right software to learn or whatever. And, and now it just feels like it's finding a voice, right?

Play from 1:26:32 8:88:88 1:26:32 Like and finding figure it, figuring out like a, a thing, a thing that's yours or a thing that you're interested in. And like, and really, yeah, really, really trying to double down on the thing that you're good at and, and interested in because like, you know, if you're going to be working with and alongside AI, if you're doing something you're bored of or aren't super interested in, obviously that's not going to be very fun. And the technology obviously rewards that reward loop, right?

Play from 1:26:59 8:88:88 1:26:59 Of, of the more you get into it, the better that you handle it. Like the, the more you get out of it. I think also like, you know, with, with this technology kind of being there already, right? Like I often wonder like what just, what it's going to be like growing up with AI. I didn't grow up with cell phones, right? Like I didn't, I didn't have the, like the internet, I got the internet when I was 14 or whatever, like, you know, it was, it was not there when I was a kid. And now a lot of people have grown up and they've always had these things and it's, it's a very different world for it and they're different people for it.

Play from 1:27:28 8:88:88 1:27:28 And, and I wonder what AI always being around will do for people to people whether like, will they even bother learning to draw? Is it just like, I'm going to dream up this thing now, or is there going to be a, you know, a revolt against it from kids or parents or whatever. But I think that kind of, I think highlights the fact that like, it's, it's incredibly important to understand what you're bringing to the table and understand what you want to bring to the table because there's so much conversation around this AI stuff in terms of the anti side is like AI does the whole thing.

Play from 1:28:00 8:88:88 1:28:00 You didn't do it. AI did it or whatever. It's like, you know, you didn't ride the bike, the bike, the bike rode you or whatever. Like, it's not like, it's not quite as straightforward as that, but it can be like, you can use AI in a way where it just does all the work and A, it's not very rewarding and B just isn't like, it's just going to be the average, right? It's going to be the most, the most average thing you can get out of it. What do you, what do you think other people should be spending more time becoming, you know, really unusually good at from this point forward? Because I think like what, what I also notice is, right?

1:28:23What should AI creatives get good at?

Play from 1:28:36 8:88:88 1:28:36 Like we're, we're all three here. We're deep in this stuff. We're in, you know, in different ways, but overlapping ways. We're sort of in our own echo chamber. Same can be said when we get into some of these social media channels and who we follow and who follows us so forth. But people that are maybe just getting into this or have laid the foundation for this, like how should they frame their thinking in terms of what they're getting good at? And what I mean by that is like, maybe an example is like, Hey, like maybe don't attach yourself to a singular tool, right?

Play from 1:29:07 8:88:88 1:29:07 Like the tool is not maybe the right way to think about it. It's almost the outcome of what that tool does. Right. Or like creating sort of leverage for yourself. And it's like, you know, ad hoc things versus like creating things that can scale or operationalize or you can use as leverage. Like what are, you know, like, what are some of those things that you might call out? Um, so I saw, I saw a thing recently, or maybe it was, maybe ChatGPT told me, um, I was asking about, I was asking about agents.

Play from 1:29:34 8:88:88 1:29:34 Right. And I was like, how do I, you know, how do I best set up an agent or whatever? And it basically said like, um, uh, you know, speak in natural language. Don't try and like prompt or whatever. Um, but then it said, it said the most important thing is give it a goal, not a task. And I honestly think like that, like, and that, that's where LLMs are going, right? Like, it's like you, if you define your goal well enough, it will figure out how to get there.

Play from 1:30:00 8:88:88 1:30:00 And I think honestly, like that is exactly how people should think about it. Right. It's like, I think a lot of people, there's a lot of people in like, especially in open source AI that are like trapped in this thing where they're just like, they're just desperately trying to like make like the longest possible video or whatever. They picked like one thing in one model and they're just trying to like get that thing working and they don't have a goal. Right. Or the goal is this thing that just doesn't really have an end point.

Play from 1:30:28 8:88:88 1:30:28 Um, and I think the, the really important thing, and as you say, like to not be completely wedded to anything, right, because everything's going to change that the only, the only thing we can rely on is that things will change. And, and the key is to have, uh, understand what your goal is and, and to, if you can point all your energy towards that, figure out how to make it along the way. And then, you know, the tools can evolve and you can still, you'll still have the same goal.

Play from 1:30:56 8:88:88 1:30:56 But I think, I think that's the key is to, is to get really good at knowing what you want to do and where you want to head. That doesn't mean like, I want to be this in five years time or whatever, but like, at least when you're using AI to have a point to doing it or like have a goal that you're reaching, not just like, what does this prompt do? You know? Um, because, because people, when people are like out there hunting workflows or prompts, they're looking for someone else's goal.

Play from 1:31:24 8:88:88 1:31:24 They're just trying to make someone else's goal. They're not trying to like, they're not trying to make their, they're not trying to make their short film better because it's someone else's prompt doing something totally different. And so that, I think that's the key is like, you can't, you can't use someone else's prompt to achieve your own thing. You're just going to make their thing. Kind of reduces the, the value to you too, because it's like you, you kind of know internally whether you're thinking about it consciously or not. I didn't create all of this too.

Play from 1:31:51 8:88:88 1:31:51 So it doesn't resonate and then feel as valuable to me. It's almost like if someone gives you something for free versus you paid a hundred dollars for it, you treat the thing that you paid a hundred dollars for it differently than you got for free. Yeah. You know, I think, I think that there's also, I mean, like, obviously like, like let's say with ComfyUI workflows, like that is absolutely the best way to learn. It's like you find someone's workflow, you learn what they did and you pick it apart and you make your own. I think with the prompts though, like you don't need to know prompts anymore at all.

1:32:15Do creatives still need to collect prompts?

Play from 1:32:21 8:88:88 1:32:21 Like an LLM knows prompts better than anyone. I think a lot of the prompt hunting thing is that I think there's this idea that there's this secret thing that you just haven't quite figured out yet. And once you know that secret, then you'll get it. And I think that's, that's what people are looking for in those prompts is like, there's that secret line at the bottom of it that you just didn't know about. Um, which just isn't the case anymore. It's just, it's just an LLM. I like that. Yeah. That's right though.

Play from 1:32:46 8:88:88 1:32:46 Yeah. It's, it's crazy. But the other last thing on, on my end is, uh, popped open the IMDb. So there's some pretty notable franchises there, especially for any of our nineties fans on here there's maybe some DuckTales and some power, Powerpuff Girls. You weren't going to let that go the whole episode without mentioning that. Yeah. Yeah. DuckTales, man. Like, I feel like the childhood selves of us watching this right now, everyone, someone loved DuckTales.

1:32:54How did Ingi end up working on DuckTales?

Play from 1:33:15 8:88:88 1:33:15 Who's watching this right now. Come on. Don't lie to me. Everyone, which I was, uh, that was a, and you said this was the, was this the original or the remake on that one? This is the remake. Yeah. And that was, it's still 20. When was it out? 2021. It was later than 2021. Oh, maybe. Um, yeah. I mean, the crazy thing, I mean, like I grew up on DuckTales. It was like, it was like my Saturday morning was like, would not be a Saturday morning without DuckTales. And the craziest part for me is that like, we did the intro, the song, right?

Play from 1:33:45 8:88:88 1:33:45 Like the, the like DuckTales. Woo. Like we did that. I think actually that video there we did as well. Yeah. That's that. Play it. Yeah. We did this. Yeah. I won't have sound on it though. That's one of ours as well. So that's not from the show. That's an ID for, for Disney or for Disney. How cool is that moment for you? Like that's, that's like this full circle, like moment, dude. I mean, like working on something that you grew up watching.

Play from 1:34:11 8:88:88 1:34:11 I don't know that I have anything quite like that, but I would imagine that would be such a cool feeling. Yeah. I feel like at some point I need to work on a Teenage Mutant Ninja Turtles. That was, if you touch that, if you touch that, you gotta, you gotta, you gotta bring us in. I don't know what, but you gotta bring us in. I thought it's funny because like, I've got, I've got five -year -old girls, I've got twin girls and I brought them into that nineties teenage mutant.

Play from 1:34:38 8:88:88 1:34:38 Like to me, that's the only version I, I even talk about. Oh yeah. It's that original, you know, cartoon series. And it's that, those two nineties movies, you know, like I don't acknowledge any other Teenage Mutant Ninja Turtles outside of those. Like just don't. Those weird Michael Bay ones. No. No, those, those guys don't even look like turtles to me. No, no, no. I was, yeah, I had all, I'd like all, all I wanted it.

Play from 1:35:06 8:88:88 1:35:06 Like the only thing I wanted in the world, I don't think I ever got it. It was the, uh, was the camper van. That, that. Oh dude. And I had a friend that had that. He had all the stuff and I like just had the turtles. He had the, the van. He had the, what are they called? The agridome. What is it? What am I? The big dome where Shredder was at and Krang. Oh, yeah. Yeah. Yeah. Yeah. Yeah.

Play from 1:35:32 8:88:88 1:35:32 Yeah. Yeah. Yeah. There's a, there's an Airbnb themed in New York. I don't know if it's still there, but it was, it was like they're what they made it as the Teenage Mutant Ninja Turtles lair and you could like rent it and sleep in it, but it was like exceptionally well done. I don't know if it still exists. I know that was like a, that was like a novelty Airbnb back in the day. Let's see if it exists. Shut up. Shut up. Is that it? Oh my gosh. Come on, man. Come on.

Play from 1:35:58 8:88:88 1:35:58 It was awesome. Oh, wow. I feel like we tried to shoot something there one time. Like, or we tried to have an event there or something, but it was like, it's so cool that someone did this. That's incredible. Oh my God. Well, should we wrap this, wrap this up? Rory, who do you want to shout out, man? Who are we shouting out for subscribing? Well, first we'll, first we'll say this, right? Ingi, number one, killer.

1:36:12Who gets the Fast Hours sign-off?

Play from 1:36:25 8:88:88 1:36:25 Thanks for coming on. You know, love your stuff. We've always been sort of Drew and I send your stuff back and forth. Just like what? But, um, it was good having you on, but we always do this at the end of every show. We just got to tell everyone something while my dog's barking in the background. Um, you know, if you've been here for an hour and 40 minutes, just got to do what one thing it's like, comment, subscribe, whatever, but really just tell somebody. Just got to tell somebody. Maybe, uh, like with the most obscure demographic possible, you know?

Play from 1:36:52 8:88:88 1:36:52 Yeah. Let's, let's see. Who cleans, who cleans? Snowplow driver. Okay. Snowplow driver. Who cleans the, the sewers in New York? I mean, I don't know who, I don't think they get cleaned. Sewer cleaners? I don't know. Okay. Well, all right. There we go. I don't think they ever get cleaned. Your, your local parks department worker. Um, you know, he might, might do that.

Play from 1:37:18 8:88:88 1:37:18 Park ranger. Yeah. There you go. Have we ever called out local like Uber or delivery driver like that? Have we ever called that out? But that would be a good one. Dudes on mopeds with the little like hand, you know, the little hand muffs. Now, I don't know if they got those down in Atlanta. They got like a, they like these makeshift like on the scooters in New York, like hand like warmers, but they make them out of like, because like in the winter it gets, you know, obviously freezing and there's a million deliveries, but they, it's like never like a, it's never like gloves on there.

Play from 1:37:46 8:88:88 1:37:46 It's like, you know, paper mache, like wrapped with like plastic, like double wrapped with something else. It's ingenuity, it's ingenuity at its finest. But, uh, you know, scooters. These are scooters that they're riding? Yeah. Well, New York, everyone drives scooters for delivery for, you know, DoorDash, Uber, whatever, because you can't have a car and park. You just got to drive scooters. So it's scooters everywhere. But, you know, God bless those. God bless those, those first responders.

Play from 1:38:11 8:88:88 1:38:11 You know, they, they are essential to us. And when it's negative 10 degrees outside in a snowstorm and you you know, someone needs a cheeseburger. So, uh, that's, that's how it works, but tell your local one. Do it. Do it. Well, uh, yeah. Ingi, thanks for joining us, man. It was, it was a blast having you on. Thanks for sharing all that with us. And guys, thanks for joining us. We'll see you guys next week. Have a great weekend. See ya.