Ep. 16: Midjourney at Scale The Superside Blueprint for Teams w/ Phil Maggs (Superside)

0:00:00Superside Story

Play from 0:00:00 8:88:88 0:00:00 What's up everybody? Welcome back to Midjourney Fast Hours podcast. I'm here, Roy Flynn is here, and we're also joined by another legend. We only invite legends on here. That's the rule. Legends only. And his name is Philip Maggs. He is over at Superside leading a team that's just doing some innovative [ __ ] I mean, we just can't wait to to dive into that today. So, Phil, why don't you give everybody a little bit of an introduction for yourself?

Play from 0:00:28 8:88:88 0:00:28 I mean, I'm I'm going to struggle to to match the amazing American energy 5:00 on a Friday that we were recording day for Phil. Yeah. I appreciate you guys having me. I'm Phil. I'm the director of generative AI at Superside, which is maybe a made-up job title, but is what I've been doing for the past 2 years is stealing Drew and Rory's PDFs from LinkedIn and then integrating them into our company and scaling it out. I think that's pretty much the the workflow that I've been doing for the past This guy's full [ __ ] This is the most humble introduction of all time.

Play from 0:01:01 8:88:88 0:01:01 I mean, I'm not good at this. You know I'm not. Rory, why don't you give him a proper intro? I don't even know where to start. Some of the stuff is some of the, you know, for anyone who doesn't know Phil, again, number one, very humble. Number two, probably has scaled AI at a capacity that no one else on this planet has, and I'm not saying that lightly. There's, you know, I'm talking about from a design and creative standpoint using the tools for some of the more, you know, creative projects and how to string them together and actually make, you know, they've built internal with multiple tools using APIs. Stuff just is some of the stuff blows my mind every time, but regardless, I think Phil is the next sort of generation of creative where you bring a sort of, I don't want to say a

Play from 0:01:46 8:88:88 0:01:46 developer background in, but like a technical background with a mixture of the creative prowess of a designer and animator, things of that nature, and you mold them together cuz that's kind of where it seems like everything's going. So, yeah, there is this new sort of landscape here and I I you know, you can you can choose to retort that Phil if you'd like. so you know, the way I like to generally describe it is my skillset is average engineer, average designer, average creative, which I actually think is the right skillset for this current era. So, knowing how to number one, I don't feel that bad that any part of my workflow is ever being replaced cuz I'm not good enough for any part of the workflow. So, like I feel genuinely excited like when something is like better than me cuz I've always got the thing in my head, but I'm just like slightly rate limited

Play from 0:02:32 8:88:88 0:02:32 by my ability to do it. And so the people that are really genuinely amazing, maybe we'll get into the whole world of 3D people later on in when as things go on, but like you know, they get pretty angry all the time about different things. They have this kind of you know, as we all know, this sunk cost fallacy. I didn't know if I wanted to drop that bomb straight away. But I just think personally like you know, I'm good enough to you go okay, cool. I can grab that and I can grab that and I can plug those things together and let's see how that works and yeah, we've been pretty fortunate at Superside that we've had some pretty like up for it founders and some pretty up for it customers that'd be like cool, let's try. And we've tried and now I think we're well over 3,000 different projects completed with generative AI pipelines. It's pretty much in every type of work that Superside does. So, Superside is

Play from 0:03:18 8:88:88 0:03:18 fortunate that we've got we we do every design discipline. So, it can be like marketing strategy and like where you should be doing your media spend and like all all you know, all of the kind of thinking stuff. Brand identity design, graphic design, 3D design, augmented reality, video production, live action, 3D, VFX. We have all of those people in house, which for me is like just having this incredible test bed. So, we go cool. Let's try out UI design and large language models this week and we will try that out or let's try out digital ads and localizing them in three different mark markets with Midjourney for the imagery this week. Like I And some of that stuff I learned from you guys is then, "Oh, cool. Let's do something that's, you know, American focused and then change it to being South Korean focused. And can we do that with Midjourney?" So, I've been really

Play from 0:04:03 8:88:88 0:04:03 lucky, right? So, we we we get to test on loads of stuff and then we've kind of passed that stage that threshold now where it's been like low-level testing, is this possible? And now it's just like continuous. Yeah. Lots and lots and lots of volume. And once we pass that volume, the learning rate is pretty good as well. So, we're like, "Oh, this doesn't work. That doesn't work. This doesn't work. That doesn't work." And like we then just move that forward really quickly. I I was just impressed at how fast you guys moved with that, too. And I'm sure like maybe behind the scenes it didn't necessarily feel like that, but it it didn't seem like you guys really hesitated when this became, you know, a resource or a tool that you could exploit and extrapolate across the company. I'm sure there were a lot of conversations that went into that. But for those of you that are not familiar with Superside, Superside is a creative as a service platform, right? So, they

Play from 0:04:50 8:88:88 0:04:50 offer subscription-based model for design and creative services. And so, this aims at really helping brands scale creative output. And that's why this is such an awesome topic and a and a just a legendary guest to have on because you put this into motion. And I think a lot of people were just like, "How do I do this?" Right? Like I know just as an individual I was working on my own, how do I properly do this? Let alone do that for an organization or for a team. And so, I think that's what really stood out to me just as a as an acquaintance and and later a a friend like of of what you've been able to accomplish there and your team's been able to accomplish and Superside just kind of leading the charge, too, on just those conversations with creativity and AI. Like you guys do a really good job with the content that you guys produce,

Play from 0:05:36 8:88:88 0:05:36 the webinars, the virtual events, all that good stuff. So, you know, shoutout to the entire team. It's just it can't be said enough like how important that was I I think to really set that precedent for the rest of I guess the future in the way that that some of these teams are going to think about this. Yeah, I mean I appreciate that, man. I can take zero credit for any of that. We've got an amazing marketing team that like really uh super up for it. Tell you what, fear is an amazing motivator, right? So, when you see the kind of meteor coming for the industry, like it it forces you to act. And I think you you know, that's the way we that's the way I saw it. Um I was like, "Okay, Midjourney 2, GPT-3, GPT-2." This was like just before I joined Superseed.

Play from 0:06:13 8:88:88 0:06:13 I was like, "Oh, you know, [ __ ] can do this." And if it could do that, and then suddenly like I joined Superseed basically at the in when ChatGPT launched. So, I joined the month before. And then suddenly that whole world blew up. And you know, if you can just roughly extrapolate out from oh, Midjourney's had this kind of improvement rate of, you know, every 3-4 months they were, you know, fidelity was going up or controllability or steerability was going up. And then, you know, you could just project a marginal improvement over time. You're like, "Okay, well, if between 6 to 12 months, it would be getting to this place." So, you just got to be slightly ahead of that. And that was the motivator was that well, a lot of the that kind of more production-focused work is most likely going to get commoditized. Maybe. I don't know. I

Play from 0:06:58 8:88:88 0:06:58 don't know if that's necessarily entirely true, but it you know, you want to be ahead of it at least. You do not want to be behind the curve. And that was really what I just went to the founders and said like, "Let's just be ahead even if it's 2 days ahead, right? Don't just try not to be like 2 months behind or 6 months behind because and I'm going to get into it today. You know, we're 18-20 months into like our internal transformation, upskilling, training, trained loads and loads of people. And we still I still feel like we're just getting started. Like I I genuinely do.

Play from 0:07:27 8:88:88 0:07:27 Like we have done this initial kind of big rollout. And now, you know, this week with the week we're recording, Runway decided to drop a bunch of new [ __ ] And Midjourney decided to jump sure you know, do their big one of their bigger updates recently. And Ideogram have launched their new canvas. And Craiyon have got every video model in the world in there. And like, you know, every like internal document or every little tutorial we've ever written then has to be rewritten. Or does Or does it have to be rewritten? I don't know, but like, you have to be in this thing of like, okay, just try and be like at least in the conversation. At least Or Or like, you know, ahead of it.

Play from 0:08:02 8:88:88 0:08:02 Because if you're If you haven't even started now, it must be the world's most daunting thing to look at everything. And we just started when it pretty much was. There was Stable Diffusion, there was Midjourney, and there was ChatGPT. And those were the things that we have. And now it's what? 300x that in terms of different choices and tools and opportunities out there. So, like, that's yeah, that's where we are today. Going back to the beginning a little bit with some context here. I think we're over a year now since we've been, you know, kind of collaborating, communicating, things of that nature.

Play from 0:08:33 8:88:88 0:08:33 Phil reached out to me via DMs, just like, "Do you mind if I share this stuff with every one of my employees?" I was like, "Go ahead." And then, before I know it, I get on the phone with Phil, and he's like, "Yeah, we're doing this internally for you know, 500 designers." I'm like, "Oh my god, okay." So, you know, the things that we talk about here a lot of the times is we're doing like individual sca- like individual projects, right? Like, we're working on like, how do we do SRF codes and you know, how do we, you know, really maximize the tool of Midjourney.

Play from 0:08:59 8:88:88 0:08:59 The difference here is that, you know, Phil and Superside are doing it at scale. Like, I don't think there's many people that are doing it at scale, training a team that big to use tools, but also working on the volume of projects that you guys did. Now, the one thing I will say, you guys two days ahead, you know, from what I've seen and what other businesses I've been in, you guys are two years ahead in that regard. But, Anything more than zero, I'm happy with. Yeah. But, operationalizing that seems like so daunting, you know? And I'm curious, like, when you went over to Superside, Phil, you're leading generative AI or Is it Is it leading generative AI in general right now?

0:09:00Scaling Creative Teams

Play from 0:09:34 8:88:88 0:09:34 Yeah, so that's what That wasn't what I joined. That's exactly was my question is like when you went over there, I guess what was the original role and what was your I guess foundational knowledge of AI or generative AI at that time. And obviously things have evolved quite a bit, but I'm just curious at where you were at in your career at that point. Good question. Like they they approached me in the summer like two years ago to essentially conceptualize and launch new services. So Super Saiyan was at a relative point of maturity but wanted to go into more front-end services. So it could be like augmented reality or it could be yeah, like immersive technologies or anything that blends tech the tech side and creativity. That's what like my background is. That's what I'm good at.

Play from 0:10:17 8:88:88 0:10:17 Very excited by new things. Will always jump on it. Will always want to learn about it. Yeah, as I said like just broad skillset. You would not hire me to be any one of those things individually. Like I just have the broad skillset. Particularly good at building teams of talented people that other people don't realize why they're talented. That's why I've I think that's my only real skill is to recognize and like nurture like talent going oh you know, especially people that are just not happy with the status quo. It's like the main thing that I look for is if someone is like continuously going no, we can do this better if we change this out and they can like replug these things in here and that takes a certain kind of person. Cuz some people are going to be like cool, you know.

Play from 0:10:57 8:88:88 0:10:57 Yeah. do do what I always do. And some people are like no, I want to burn it to the ground every time and start again. I like that. Yeah. That's the kind I like that. I look for, you know. Well, you do have a good team. I mean, some of the people on your team are insanely talented. I've seen their work. We see it on LinkedIn. We see it on you know, the other platforms and they make it look so easy, you know, like They're also insane as well. That's kind of part of it. Like so you have to kind of take the rough with the smooth. I love my team, but like you know, they're just kind of a band of crazy pirates that we kind of assembled together. So like you know, engineers, researchers, designers, technologists to fix weird problems in ways that haven't been fixed before. So they're not genuinely looking at like how other people have done it.

Play from 0:11:32 8:88:88 0:11:32 They will just sit there and plug stuff in and and figure it out. Um so I was hired to yeah, look launch new services. And within a month of being there, I basically said to my boss and the board, I was like, we should be doing nothing else apart from looking into generative AI. Let's just put all of our time and money and attention into doing that. And it took me a few months to convince them of that fact. We had to like, you know, run what we called them shadow projects to start off with. And so I would just we we again we're fortunate, right?

Play from 0:12:02 8:88:88 0:12:02 Superside has roughly 24,000 projects, creative projects that run through the business every year. And that all runs through one place. So we have a platform that we own and operate called Superspace. So that's basically briefing, so customers intent, they ask for the things they want. And if you abstract from that, we've basically got a bunch of series of text boxes that then store and log information, which is really good when it comes to later down the line if you want to automate these things. But all the customers come through one place. They will ask for creative services. We've got a then have, we're not Fiverr or Upwork or anything like that. We have a huge group of full-time designers and creatives there. And I can see pretty much all of the projects that going on. I can't, you know, be in the detail of all of them, but I can have quite some visibility. So what we would then just look for early on was like, okay, can I shadow a

Play from 0:12:48 8:88:88 0:12:48 project to see can me and Alexander, who was the kind of initial person that I was working with, um find a different way of doing this. So can we use Midjourney and can we use ChatGPT and can we use DALL-E? That was the three things that we were like doing early on to take the client's brief and their brand guidelines and get something similar back to what our designers who were working on the project were doing.

Play from 0:13:13 8:88:88 0:13:13 So they would be we'd be paralleling it. Like the designers would be doing the normal projects and we would be sat underneath it. And sometimes we would tell them we were doing this and sometimes we wouldn't tell them we were doing this. And sometimes we would send stuff maybe a little bit riskily to the client without telling people. Sometimes we would mislabel it as well, and some because we were getting this feedback like maybe when we would do focus groups internally going, "Do you like option A or option B?" And this is the AI option, and this is the non-AI option. And the people always look at that AI-labeled option going, "Oh, it feels a bit cold. I don't get much, you know, feeling from it." Yeah.

Play from 0:13:48 8:88:88 0:13:48 Then we just tried switching the labels. And then turns out if you just say it if it's AI, people then just have that feeling on the whatever image that you've sent us as the AI. at that? That's super interesting. look at that? That's That's interesting, dude. Cuz this is This is always, you know, I think there's a lot of people that are listening right now that are probably either one either just starting to work with clients or are working with clients. So, give us a give us a little insight into what those, you know, like the first time you're pitching AI to a client, like this is how this is going to work, this is how it's going to look.

0:14:15AI Client Projects

Play from 0:14:18 8:88:88 0:14:18 Give us a little idea, Phil. Oh, [ __ ] This is going to be good. Yeah, I mean, to be honest, it was pretty by the seats of our pants, to be honest. Like we would just really had no idea. And some of it was completely off the rails where we would then So, for first instances when we were pitching it, we were mainly listening. I got to be honest. We were just like listening. And sometimes we would just genuinely be shadowing. The second phase of it where it got is where it got a bit more interesting is then when we would just blind accept projects, right? So, we would We asked the the product team to put a flag into like the briefing system just to ask the client, "Are you cool with us trying AI on this project?" And so then they would click a button, and a proportion of them would click the button. And then they would just flag it into this little Slack channel that we had. And it was then our jobs to go in there, and yeah, try and grab that project before anyone

Play from 0:15:03 8:88:88 0:15:03 else would do it. We just would let them just be sat there like trying to do it. And so sometimes we would then be going back to them pitching it, and it would not work. Sometimes they just didn't even We would just go back and pitch it, and we just didn't tell them how we did it, and they were like just super happy. Sometimes we committed to doing a project, and this was like Was this Midjourney four then? This probably Midjourney four ish, I think. Most likely. I don't think so. Yeah.

Play from 0:15:29 8:88:88 0:15:29 Where we could not we were finding the edges, right? So like Midjourney and illustration like a lot of like maybe take one step back. We service quite a lot of like SAS companies, tech companies, great clients, but they generally have quite like this flat illustration style which is quite like present within like West Coast tech companies. We were not getting very good results with that flat illustrative like vector style illustration with Midjourney even like you know image to image back then and we would then have to do a lot of post-production. So a lot of times then we would then have to just sit there and draw it to make sure we then delivered on the client's request. And this is, you know, if you just take another step back from all those things, would I recommend anyone does it that way? Probably not because I

Play from 0:16:16 8:88:88 0:16:16 basically told the business this is how we're going to approach it. I put together a small group of people that could handle most situations. Like they're not just prompt engineers. They are like just all round design rockstars that can do most things. And we yeah, then we would like go and just try. And we would tell the client it's going to be half off before we even started. We didn't before we even knew how long it would take.

Play from 0:16:42 8:88:88 0:16:42 And so we just had to find those areas and and what it did is it backed us into a corner where we took away the the safety net of it being theoretical. And that's where I think most people struggle. And this is like I will just finish this one point on that is that I don't think anything's really real unless the client presses publish at the end of the day. If the client says, "Cool, I'm happy with it. Let's ship it out as our next ad or let's put it within our website or let's put it in the deck." I think it's just theoretical. Like it's so it's so within this realm of you may like it, Drew you may like it, Rory you may like it. I don't I think that matters, really. I think that matters when it comes to professional work, the client presses publish. They press publish, all of our opinions don't matter because that's the thing that we're trying to serve at the end of the day. And then if that

Play from 0:17:28 8:88:88 0:17:28 published thing performs, which is another thing later down the line, we can come to that maybe at the end like performance coming back into this is an interesting thing. That forcing yourself to be within the real pressures of a professional environment of it going there's a brief, they said they need it tomorrow, we said we're going to do it, and they need it for a real reason. That means that you have to get your chops. You have to go in. And then we like will come on to that maybe some systems that we built to help us like in those early days to like kind of get through a volume, but that kind of mantra of like move and then also, you know, just really be dictated by the client's request, not what Midjourney wants us to do, what the client wants us to do. Can we make Midjourney do it? That has some pretty interesting results.

Play from 0:18:09 8:88:88 0:18:09 Hey Rory, that reminds me of some of those conversations we've had about problem-solving, right? And unless you've been presented with certain problems and you try to solve them using these tools, like if you don't do that, it's just a choice. But if you've got, let's say a client that needs something done, you've got to figure it out. So, you're almost pushed into that mandatory space. And and you're right, like I think that's a really interesting sort of way to look at how you continue to roll the ball forward in that perspective.

Play from 0:18:38 8:88:88 0:18:38 I got a good I got a good question for Phil on this. Speaking about these same exact, you know, dovetailing right off that cuz you mentioned it and it's probably I thought one of the coolest ways that anyone sort of injected AI into the business organically. You mentioned your team, right? And like this little seal team that basically, you know, is a tactical unit that was, you know, just getting really good with the tools and doing this for client work. But how do how does it start with six people or five people, however many it was, and end up at 500? Like I think that's important for people who are trying to scale businesses, too.

Play from 0:19:09 8:88:88 0:19:09 So, candidly, not with me. And that's the thing, right? So, your innovation team, your seal team, your pirates, the tip of the spear, like however you want to call them, they are the ones that help they help you go from zero to one, and they're the ones you want there. And that's what we continue to do. You need then other skill sets to take So, like broadly, and I'm not saying this is the same in every business, but broadly, you if you're lucky, you've got like maybe 20% of your kind of demographic is early adopters, if you're lucky.

Play from 0:19:39 8:88:88 0:19:39 And those people you don't need to convince. You just need to give them the the psychological safety that it's okay to do. Hey, here's a credit card, go buy my journey, off you go, try not to screw us up too much. Those people those people don't need any real direction, they'll find their own. Then you've got maybe the 60% ish of people that are able to, but they don't necessarily um feel confident, or they not have the they don't have the early adopter mindset that um means that they'll push through that barrier themselves. They will need maybe a more of a structured approach, which is cool. And then you may have like then roughly 20% of people that will hate it. And that's a difficult challenge at the end of the day.

Play from 0:20:15 8:88:88 0:20:15 And that's not wrong, either. And maybe they shouldn't be doing the the AI work, because they are super good at the other stuff, and that's actually cool and fine. I actually don't think you need to make everyone into everything. But the direct answer to that is that there is then the different profiles, and there's Aki, who I work with, who's the director of learning at Superside. And me and him fought like cats and dogs early days, cuz I was just saying this is [ __ ] He probably listened to us, and like we were fighting. Cuz I was like, "Just throw them in." He's like, "That's not how people work." And I was like, "Just throw them in, throw them in, throw them in."

Play from 0:20:45 8:88:88 0:20:45 And he um found this way of doing essentially like a peer-to-peer coach-led learning program, which created a framework and a boot camp around it. So, we would, you know, for example, if especially if you've got a large group of people, if you've got 500 people sat in the room, and then you say, "Hey, can someone like share their like, you know, where it's all going wrong?" Like you're not going to have people like holding their hands up saying where it's all going wrong. But if you break them down into smaller groups, where they've got a coach that is listening to them, that they're willing within that community of eight to 10 people, then you you're creating an environment that they're just willing to learn a little bit more. And then we did it through stealing your content, you both of you, putting it into a different one.

Play from 0:21:26 8:88:88 0:21:26 Then we would put it though through the framing of practice projects. So client asked for this, we need to get to that, this is how it is. And so we did we staged those practice projects through there. And yeah, that 12-week boot camp thing is we're coming to the end of that process 18 months in, because we took people through only amount of cohort sizes, like the learning group sizes that we were willing to take on at the same time to keep the groups small. And that has taken people from zero to one, but that is not the end of the story. Like the we're about to roll out the next phase, which is much more of a workflow-based, like really targeted workflow-based, cuz now people have got the baseline information.

0:21:30Building Frameworks

Play from 0:22:03 8:88:88 0:22:03 They know how to prompt, they know how to do, you know, especially like some of the frameworks that you guys talk about, like how to think of composition and lens choice and subject choice and like, you know, all of the things that and not a lot of people know that. And I've got to be honest, photography styles, cinematography, like not a lot that is a it's a s- that for me was fairly intuitive, but for a lot of people wasn't. So my team collaborating with a transformation team, and you don't have necessarily have to be a massive business to to take those two things on, but the risk of aggressive people are not always the same personality types as the people that like are the kind coaches, supportive people. Like we're just like charging in with spears. Then you have to also have the people that like will then look after people's concerns,

Play from 0:22:50 8:88:88 0:22:50 psychological safety, just am I training my replacement? How am I, you know, getting compensated for this? Like all of those things are just totally fair and valid questions, which are difficult to answer all all time. And, you know, we're not training people as replacements. Like, we're just training them to be better. And that's the truth of it. We've hired more people due to AI than we've you know, that's it. It's grown because of it. So, that is the like cards on the table truth. That's just the way it is.

Play from 0:23:16 8:88:88 0:23:16 Love that. That's such a good thing, I think, for people to hear because it it is like this whole, you know, esoteric conversation of are we are we going to like is everyone going to lose their jobs to AI? And it's like you guys have figured out a way to hire more and to grow and to scale. And that's probably pretty rare at this point. I I don't know how many other businesses have, you know, figured it out in a way that you guys have in that sense. As you guys can hear a baby screaming in the background. Sorry about that. Um but there's It was interesting, right? Cuz when you guys when you first told me how you were doing, you know, it's like here I have like this little seal team. They're not allowed to use anything other than AI for some of these projects. Like, you know, we're hamstringing them. Like, no Photoshop, no, you know, Illustrator. I thought that was cool cuz I'm like Yeah.

Play from 0:23:57 8:88:88 0:23:57 That's forced Yeah, forced problem-solving, too. They didn't like me for it. I know. I would hate you, but I would also be like Now they love you, right? I mean, cuz I would be like, you know, regardless, maybe you guys don't love Phil. But seriously, if you told me to do that, I would hate you and like be like this is awesome at the same time cuz it's just like a challenge, right? softened that stance over time. Like, that was like pretty That was like That was so cool though. I was digging that. fun. And I think to be um that is forcing people to people out of habits, right? Especially, like, I've got these experts on my team. So, like, Julio and Manu, like, they like con- consistently make me feel bad about myself because I will do something and they will do something twice as quick and three times as good.

Play from 0:24:41 8:88:88 0:24:41 And I'm just feel like an old man. Like, I just like they just have just so much more talent. But you have to then sometimes force people with a lot of talent to be more constrained. And they're like, "No, no, no, I know you can draw this. I know you can animate it. You're wonderful. You are great. Please do it the new way cuz we need to find out where the edges are here and you are the only one that can judge the edges because every tool at the moment is a magic box that's going to turn out your productivity. That is like but it can't all be true.

Play from 0:25:08 8:88:88 0:25:08 No. Drew, this is also Oh, sorry. My bad. No, they're No, they're not. You know and that that's the truth. There there is probably 10 absolutely killer tools out there and then there's you know, a thousand. Yeah? Yeah. Tools? Yeah. Yeah. 100%. And Drew, we talk about this pretty often, right? Is like and it's just it's interesting to see at a more mass scale is like finding the finding the rough edges and the breaking points cuz the only way you do that is if you get creative with the tool and you try to like really and that starts with, you know, baseline understanding cuz you can't go and try to break something and figure it out if you haven't, you know, if you don't have like those baseline skills that we're talking about. I think it's just again, it's it's funny to watch this at a at a small scale and a large scale the same sort of process

Play from 0:25:54 8:88:88 0:25:54 mentality Mhm. goes into it. We definitely got to talk We definitely got to talk some Midjourney stuff too specifically cuz one of the cool things and not to not to take over and just ask this question cuz I when I first met Phil, this was also one of the coolest things that he told me. Back in the Discord days for everyone here, you guys were using it you guys were using it in a big in a big way from like a company standpoint, right? Like that was you guys were using it more so from a team perspective. You had people working in collaborative spaces. You had a bunch of things. I think that was, you know, still probably do do that still today?

Play from 0:26:25 8:88:88 0:26:25 Yeah. Is that still Okay. Like that's I mean, so I think we're at the tail end of that. We're losing some functionality because of it. So just to kind of add a bit of meat to that bone is like, you know, unless you are gamer nerds, some people were opening Discord for the first time, right? Right. started using Midjourney. And then we kind of go, "Okay, right. Let's How do we get, you know, people start to put out the simple tutorials of like how do you get the Discord bot into your own private me?" Yeah.

Play from 0:26:51 8:88:88 0:26:51 Yeah. the Midjourney bot in your own private Discord server. I was like, "Oh, [ __ ] Okay." And so then we just like started looking at that. And then we, you know, only we had a team of 20 initially. So it was quite easy to manage. You know, so like we went from 5 to 20 and then before we went to 200 and 300 and stuff. The 20 was relatively easy to manage in the first kind of in the early days. So we came up with this very simple system of like, "Okay, everyone's in this one Discord. We know we have up to 500 channels within that Discord. So let's have a channel per project." We We group them by that we were like, "Do we do it by customer?" We're like, "No, that's going to you know, like take loads of time." So we just did it by date. So it's like, you know, October, November, December. Like, you know, we would just have it by date.

Play from 0:27:32 8:88:88 0:27:32 And that was quite simple. It was in like have everyone within a private Discord space. That meant that you can just have visibility of of what's going on. And then we thought, "What's most likely better than visibility is like aggregation." So can we make our own Discord bot which will then take the prompts and the images out of Discord into another server. So we can start to actually look at a more global trend of what was going on. Because prompting is still, you know, still hard. Still hard because it changes all the time. This is one thing that made us stop doing it in the same way because each time Midjourney would come out with a new new coherence. Yeah.

Play from 0:28:08 8:88:88 0:28:08 Yeah. Yeah. Um but we just started aggregating a bunch of prompts. And we still got them. And um so there was one simple mechanic with that was that we would take, you know, we would say, "Midjourney bot, look at every single channel and then correlate that against our customers that we have and create um basically a aggregation or like an asset management system of per client per image created. And then store all of the images as images and store all of the texts as text. So like store all the prompts as prompts.

Play from 0:28:36 8:88:88 0:28:36 And then a simple um voting mechanic to see like, "Okay, if it was upscaled, that means it was interesting enough to be looked at. So the upscaled ones are like a plus one point essentially within that system because of the In fact, it was something I learned from you, Rory. Um permutations. Mhm. Still the thing that makes me go to to Discord more than I still go to Midjourney's um user interface is that that you know, like permutation of four four four by four basically is in like do a different amount of stylization and a different amount of image weight, for example. But like so you then you can then get 64 images per prompt. And then

Play from 0:29:22 8:88:88 0:29:22 slightly change that out. You know, you can be back creating 256 images very very quickly. Of 256 images, you are most likely guaranteed to get four. You're definitely going to get four killer images. Definitely. So, No typos, though. If you typo, you you screw that whole thing up. Yeah, but that was but like that's how I do it. Like I'm my my my my approach personal approach is like quantity over quality. Like I just will try and find the right framework of prompt right prompt framework and then use permutations to go to quality uh to to go to quantity.

Play from 0:29:56 8:88:88 0:29:56 Yes. Yes. Upscale to then go to go if upscale is like a plus one is an okay, that's interesting enough to be seen. And then the final um signal was, okay, add a thumbs up to that image to say if it's used. Mhm. And then you start to then get if you do that across thousands and thousands of images, what's the kind of prompt that is more likely to be upscaled? And what's the kind of prompt which is more likely to be used? And there's that's just an interesting data set over time. And then maybe you build a custom GPT out of those things because then you can start to get into some prompt formulas around that. But it's mainly centralize, aggregate, get it out of Discord, turn it into something that a large language model can look at plain text.

Play from 0:30:38 8:88:88 0:30:38 And then start to use like a ChatGPT or large language model to tell you what's good about these things. And that's a pretty Yeah. circle to get within. Midjourney's kind of broken that for us now, and we're looking for different ways to to maybe make that better for ourselves because the mid the user interface has increased in efficacy, but they've kind of taken away this like it being a good source of knowledge for us. It's now like a semi Photoshop replacement thing, especially from this week.

0:31:00Discord Pipeline

Play from 0:31:10 8:88:88 0:31:10 Phil, how many seats with Midjourney do you have? 250. Ooh. Midjourney, are you are you listening? Yeah. Might be number one customer here. $60 * 200 Yeah, we're going to do the math off that. We have to. $2,000 a month * 18 months on that. So, Midjourney, we spent quarter of a million dollars with you. Why don't you answer my emails, please? Like Ooh, brutal. Brutal. Well, you know, Phil, you know, the other that that was first of all, that was so interesting to just hear that laid out.

Play from 0:31:43 8:88:88 0:31:43 Do other people not do this, by the way? I don't think so. I think some of these thoughts come into play, right? But whether they you know, ideas are easy, it's the execution. So, I think there are pieces that probably people have thought of. There were pieces of that that I think happened organically in terms of like things like creating the Midjourney private channel or upscaling as a way to like filter down because they didn't have any organizational system at that time, right? And so, like some of those things felt like they would happen, but the way that you executed it and then kind of plus one that plus one did, I think that's the rarity, right? And again, that goes back to what we talked about earlier about almost being forced to move in that direction, and I think that that also has a lot to do with the space that you guys are in. It's creative as a service agent, you know, a company. So, of course, right? Like that could go hand in hand, but

Play from 0:32:28 8:88:88 0:32:28 one other thing that stood out to me that you just mentioned, when you're going through all those images and upscaling them or plus oneing them, curation comes into play. You know, like how do you how do you sort of control that or like what's the trust or the autonomy that takes place there because now it's easier to create good images, but if you've got 256 images, how do you pick the best one? Well, I want to be candid. Like we haven't entirely fixed this problem, right? In fact, I think we may be getting worse at it as the technology gets better because of we're we're we're more abstracted from Discord now. It was easier. You know, I'm fortunate because we just have really good creative people. Like we don't have to augment non-creatives to think

Play from 0:33:14 8:88:88 0:33:14 creatively. And what creatives are really good at is knowing like visual and having a good visual acuity, good visual understanding. They they are far better than than I am. In fact, they we've we've had to mainly teach them the opposite, right? Rather than picking curation they're good at. Perfectionism they're bad at. Right? And so do you want to post-produce anything that's not going to be final? Because people generally still want to do that because it before like so it's it's teaching people a new muscle to go right, you're used to having four options. Now you've got 40.

Play from 0:33:49 8:88:88 0:33:49 Curate down to eight and then only post-produce four. But don't but but if I'm not saying I actually think it's a bad comparison to compare Midjourney to stock because I don't believe in that comparison because stock I can't do anything I can I stock I get what I'm given. I have to search for it. Other people are using it. It doesn't quite fit my needs. Like I so you know, Midjourney is closer to photo shoot than it is to stock. In my opinion or it's closer to 3D design than it is to stock. Yeah. And I think that's a bad comparison actually is hurtful to the creative process. And so when people say it's like, "Yeah, but it's you know, I can get stock for cheaper." I'm like, "Cool. Stock still exists. Go do that." Like it is it's it's not it's not either or. Go go use

Play from 0:34:35 8:88:88 0:34:35 stock when you want to use stock. If you want to make something new, go use you know, go use an image generation tool. I don't know, I lost my train of thought there. Post-production, right. And then the perfectionism that then comes in will go into Figma. It's one of our It's our basically like collaborative tool of choice, you know, we can have hundreds of people bouncing around within Figma and you can be doing you know, adding comments and you can be adding it into you know, bigger artwork and Figma is the central hub of the company. And we've got a lot of automations to go in and out of Figma, you know, for that reason.

Play from 0:35:03 8:88:88 0:35:03 But, you know, I I go into some people's Figma files and there's like 32 different comments on one image going, "Yeah, but that's not quite that and that's not quite that." I'm like, "We're on the first round. What we want to do is be cool. It's reasonably good. It's not there. We want to add the extra 10% in post-production, which we always will do. We've never shipped anything direct from Midjourney. I don't think once ever." Important to note. Great point. Ever. Hundreds of thousands of images, not one time that we would just go, "Yeah, good enough."

Play from 0:35:30 8:88:88 0:35:30 I think if you look hard enough, there's always something, right? Depends on the screen you're looking at on there. Like, you know, if I'm looking at this if it's on you know, the side of a building, yes, I'm going to really scrutinize it. If it's you know, display ad, you know, midway through a website like it's going to be seen like that on you know, It's like 70 pixels. Yes. Yes, you're you're good. You're good. Um in perfectionism eye, but you want to say to them, "Do it in the right part of the process.

Play from 0:35:57 8:88:88 0:35:57 Do it in the part of the process where we're going to press publish. Don't do it on 400 images cuz you're going to drive yourself insane with doing that." And so that is that curation do like to answer your question is I think they're good at curation. I think most people um struggle with perfectionism because you're cuz you're used to most likely if you're brand if you're a designer working in marketing handling stock and stock is already post-produced.

Play from 0:36:22 8:88:88 0:36:22 You can't then compare it. You're comparing it more to raw files than what we're talking talk It's a great comparison. Yeah. to be, you know, retouched. It has to be you're in that stage of the process. The photographer has not done the post-production yet. Hasn't done the grading yet. So, you're choosing. You're selecting and you're finding what works best in the composition that you're doing. But you're not then going, "Hold on, that earlobe is slightly lower than the other one." Or what, you know, like none of that's not the stage of the process to do that yet.

Play from 0:36:51 8:88:88 0:36:51 I think that, man, that's such a great It's such a great point that I don't think Rory and I have talked about before, but it's definitely something that we've both felt, you know? Like I I know that's that's got to be a problem for a lot of people, you know? It's just like you're trying to perfect it, and then when do you take it out of the tool and turn it into, you know, when do you then transition it to Photoshop? Where do you draw the line in terms of, you know, repainting in Midjourney versus going into Photoshop and generative fill and adding all these other things, right? Because there is overlap there. So, you do need to make the decision. And then also, to your point, what's the what's the outcome or ultimate placement for this thing? Right? And And so, all those things come into play. Yeah.

Play from 0:37:28 8:88:88 0:37:28 Can I ask you guys a question, though? Of course. Sure. There's been some stuff coming out this week. I've done Maybe you can show the one image, Rory, that I sent you. There it is. Love it. So, you know, maybe I'll lose half the audience here. That's the one image Midjourney image I've made this week. So, I am not the the source of truth. But then, you know, that's how I see myself at From your guys' point of view, cuz my normal workflow is like, "Cool, got it. Let's go into Photoshop." Right?

Play from 0:37:55 8:88:88 0:37:55 Mhm. Let's just take it over to I'll do all of my post-production generally in there. Had, you know, over a decade of experience in using Photoshop. I know you guys have got that experience as well, but like, would you see what they've just Cuz people are saying this is the Photoshop killer. I'm like, I don't I don't feel that way. How do you feel about the new things that have been released this week? I'm like, "Awesome. Thanks." You know, like I'm not I'm not like overly hyped. I see I don't know if people are manufacturing hype for clicks on social media where like, this is the best thing I've ever seen.

Play from 0:38:24 8:88:88 0:38:24 Like I mean, well then someone would be like explain it because I, you know, realistically, the one feature I like is the, you know, if anyone's not familiar or hasn't seen this update yet, or you don't have access to it yet, um Midjourney released an external image editor, so you can now take your own images in and edit them in the same way that you can edit, you know, any Midjourney image. They also added a retexture feature, which is similar to ControlNet. I like it. I'm going to use it. I don't personally, I'm not a big tinkerer when it comes to editing. If I can't get something with a text prompt, like I'm typically going and just doing it a different way. Um for fine-tuning, right, or fine, you know, really going in with the scalpel, I like Midjourney for the Swiss Army knife type of thing, like you said before, and then Photoshop for the scalpel, because I can get in there in Midjourney, and like if I just need, you know, sometimes you generate like a 16x9

Play from 0:39:11 8:88:88 0:39:11 image, and it puts the black bars on top and bottom. Like, yeah, if I can just go in Midjourney, highlight that, you know, edit that piece, great. Then, you know, sometimes when it's really small details, like, oh, there's like a little earring that's kind of like off the side of the ear. It's not actually attached to the ear. Trying to get that to get fixed in the way that you want it to, it's really hard, but you can do it in Photoshop cuz you can like go down to the pixel. Now, we only have like this brushstroke that's kind of like dependent on that, and then also, the other thing that's really I've noticed about it even just testing it for a day or two, is that the one thing that I do like about it more, is you don't have like the combination of aesthetic. Like, when you generative fill with something new, Yep.

Play from 0:39:47 8:88:88 0:39:47 cuz that used to be Photoshop was just not nearly as good as Midjourney. They're like two different kinds of aesthetics. So, now we can keep that sort of consistency, which I will use it for way more there. But, if I want to go fix small pixels and small little artifacts, and be like very precise, going over to Photoshop. I guess if there's any other line I'm thinking about, and it's more subconsciously than anything, it's like, am I using this image for myself? Like, is this going to be used on like, let's just say LinkedIn for a post, or is this going to be something that I submit to a client, or like I'm trying to create artwork with.

Play from 0:40:18 8:88:88 0:40:18 You know, like if it's that, then it has to go into Photoshop. There are just too many I think benefits of doing that that it that way. One, you're not burning all these hours, right? And like you know, you if you're trying to do all this stuff in relax mode anyway, I mean it it can be painful cuz sometimes that [ __ ] takes forever. I mean and so how many Fast Hours do you have? I like to run a lot of permutations. I'm thinking about you know, how can I conserve some of that? I'm not sure external editor yet how fascinated I am with that. I think with a lot of these features it's it kind of unravels over a couple weeks in terms of like the full functionality and capability with it, but I do like the the retexture piece for a few reasons, right? Like perfect example, got a client that's doing like interior design. They have products that they put in those rooms.

Play from 0:41:04 8:88:88 0:41:04 If you wanted that same living room in different styles, Cape Cod, modern, masculine, farmhouse, I mean all of a sudden you can do that, right? And they can be exactly the same. So if you found a composition that really works well or a camera angle that really works well, that's just saved you a bunch of time, right? So like I I just that was probably the biggest value I got out of it in the first couple days. Same with like something like icons for example. If you got a set of icons that you've created, which I also just did this recently, I had those in a different texture and I wanted to switch gears and put that into I don't have to create a brand new asset and try to get that prompt exactly right. I can just go in there and refine that through retexture. So I think that that is interesting. Rory, I don't know I like that.

Play from 0:41:49 8:88:88 0:41:49 been playing with this, have you seen like with the retexturing, I just don't feel like it comes through as high res as it should. No. It depends on it's like going back to the old days of mid-journey where like if you have someone that's probably in the mid-ground of a photo or something that's in the mid-ground of the photo, seems like it loses a lot of fidelity when you retexture it. So if it's like a face Yes. here in the image, it's getting all morphed.

Play from 0:42:14 8:88:88 0:42:14 Right, for sure. That's how That's how I'm seeing it cuz it's The one thing that you just said the the the icon thing, super cool. Like I just that just totally sent me down a whole 'nother path of what could be done with that. Yeah, assets. Remember we were talking about assets like with with Billy a couple weeks ago. I mean, you could go a whole thing with that, right? Client to client even. Like if you're just like let's just say you're creating icons or illustrations and you've got certain ones that you've like all of a sudden you can quickly turn those around. You don't have to start from scratch. I mean, that's always been I think the hardest part about staying in tool with Midjourney is just the number of steps it would take to get to the final result. And you know, when they did change and update it to editor, that was I think probably the thing that you and I liked the most because it was like you can zoom, crop in,

0:42:15Midjourney Editor Update

Play from 0:42:59 8:88:88 0:42:59 change aspect ratio, paint all in one step essentially. And before you couldn't, you know, that would be like three or four steps. Right. Two things that I would really like from this tool, just to throw this out there into the ether. Number one, I submitted it on the on the ideas thing the other day. I don't think it got any votes. Sometimes I want to flip the picture. That's it. Like a quick flip. Oh, yeah. You know, flip horizontal, things like that. Like that that would be great. Or being able to take the image and tilt it vertically or however like you know, if I cuz I was doing this the other day, right? I wanted to make an image for I wanted to do a key frame for like a a fighter jet, right? I was going to do take us in a runway. I started with an image where the jet was, you know, flat plane, but I wanted to show it making a turn. So I wanted to tilt it, right? Now getting it getting it to do that with

Play from 0:43:46 8:88:88 0:43:46 just text prompting and editing was just not working. So I took it into Photoshop, right? Took it there, I rotated it, and then I did my generative fill and my cleaning up around the edges so that I had then two key frames. So simple things like that where you know, either cuz there you know, you get some images that you're like, that's cool, but I need it to flip the other way. Yeah, yeah, yeah. to Photoshop to do that. Well, I ran into that with the icon thing actually because it was like okay I had one icon that was kind of to the right you know that was kind of positioned at a 45 degree angle this way and the other one this way and I was like got to use those together but I can't do it in tools so I got to I got to you know use it and then take it out into preview or whatever you're using to flip it but You guys are fascinating with this cuz like I want just so much of the opposite of what you just said. I don't want Tell us tell us more I want them to focus on any product layer at all. I

Play from 0:44:31 8:88:88 0:44:31 don't I want them to spend zero of their engineering time on this UI and all of their time on the model and on the and then I just give me an API. I don't care. I just want to I I want to go between Midjourney's API and Runway's API as quickly as possible. Runway thankfully six weeks ago released that API. Lovely stuff. Now we can play around with things in their front end which is great you know just played around with act one today super cool loads of applications for that.

Play from 0:45:01 8:88:88 0:45:01 For those who don't know what act one is it and may know I've seen live portrait live portrait has been around for probably about the last four months in the kind of comfy UI community people have been doing things with live portrait. It will lip sync a human input to the output that you want it to be so I can take a an image of someone and that it would actually do realistic lip sync so I can have an input video of me talking it will then like transpose that onto whatever your video or image output you want it to be and lip sync has always been this really big problem when it comes to the generative video is that you know it's always been and it's they've got it into a pretty cool place but I want to use Midjourney's Midjourney's been great because their model has been the best right it's had the highest quality it's been probably the most it once you know

Play from 0:45:46 8:88:88 0:45:46 how to prompt the most coherent to prompts I think really maybe that could be argued it's not the most coherent to simplest prompts but if you want to go into like the way that you guys prompt it's very coherent to those prompts and it's got the best quality from it. I don't care about the product layer then like to make it into this kind of like I don't know what it where where is it going? Is it going to like into this Canva territory or like I don't want that. I want pure quality cuz that's the thing that really matters at the end of the day.

Play from 0:46:14 8:88:88 0:46:14 And ideally come on just release the API. Yeah. Release the API and then give me P ref, product ref. Yes. Rather than it like I just give me P ref and don't give me S ref. I've got C ref, I want P ref. I want P ref. Let's go. and prefix that would be like the thing that then you've got what we can down to with flux in Midjourney. Like that would be awesome cuz I want both. I want them all.

Play from 0:46:40 8:88:88 0:46:40 Yes. It's interesting because it you know if if Midjourney did have that API I don't even think we'd be talking about flux. I don't think flux would be coming up in conversations at all. No. Because then you've already got people that have been using Midjourney for a long time that have no reason to leave again. You've you've created the stickiness within it. I I just wonder if Midjourney's ever going to go the route of of API. Are they I mean maybe it's just like maybe it's just been this money printing machine or some I don't know what I don't understand. Like you know You can straddle two worlds, you know?

Play from 0:47:08 8:88:88 0:47:08 Right? I feel like Runway does a great job of doing that. Like you can be corporate and like enterprise but you can also be for the people. Like there's there's a way to do it. It's not it's not overly you know you see it. You work with it probably way more with the APIs that you guys are you know dialing into. You know there's there's a professional side opens up possibilities and then you start to then like listen you we well we mainly spoke about this like the more enterprise you go and the more that people ask about okay data sovereignty and where's it going and where's the data privacy and is there personally identifiable information in there and like am I is my am I images being used to train the model which they 100% are with Midjourney. Like I mean it's a pretty cool play from that point of view that they've just opened up a new way for them to have people upload model images to it Yep.

Play from 0:47:50 8:88:88 0:47:50 and then prompt on top of it which will then get used to train the next model so you know that's what they're most likely doing across the board there but you know there is the API side of things means that I want some if I'm going to put a lot of effort into it, I want to know that there's longevity with that effort or as much longevity with that effort as possible. What I really want is that to be chaining these things together because the real value for us as a business and where I think the real value for most people over time is not how to be a Midjourney power user, it is to how to go from what's in your head to what's on the paper as quickly as possible so you can do something with it. My expectation here is as everything grows in efficacy, people's expectation around what speed is will change. So my like my son, right?

Play from 0:48:36 8:88:88 0:48:36 You know, like he's iPad generation. Like his expectation around what a computer is is different to mine. And like, you know, when he is you know, when he's 20 years old, his expectation about what the speed of these things will be will be just like unmatched. So we have to be thinking of these things in like how do we get like and our clients will get there as well. Our clients are not going to be like, "Okay, we'll come back to you in 3 weeks." That is not going to be their expectation. No. That's just not going to be it within the next It's not even really it now, but it's not going to be it in 5 years time.

Play from 0:49:07 8:88:88 0:49:07 So what I want from Midjourney is to go, "Cool. My briefing thing is like an large language model powered. I know what a good prompt is and I know what a good prompt is for this brand. So then I want to be able to run batch process 1,000 prompts at once based off this campaign idea that my large language model has come up with. I want to batch process that. I want to then lay those out in Figma with another process. So take the design system data which is just code, use that code to write the layout cuz layout is just code. You know, I don't know if people think about it in those ways, but that's what it is.

Play from 0:49:42 8:88:88 0:49:42 Write the layouts, write the prompts, write the images all into one place, and then choose. Choose the ones that you want and go and go off and do that. If you are interested in having your mind blown, Phil did a presentation at Config, which is Figma's major conference. I think what was that June in June? And basically unveiled the design system, how it works, how the entire process looks, and it's how you know, basically how to go from brief to a final editable product in Figma without really touching anything other than the human input, which is insane from an image template text standpoint to get to that point. So, we might have to plug that one.

0:49:45Automation Strategy

Play from 0:50:20 8:88:88 0:50:20 Yeah, I mean that Thank you for that. I mean, I guess like the one thing I want people to know, and like I'm not I'm an open book when it comes to this stuff. Like don't use a diffusion Don't use a diffusion model to try and create layout. And we were just banging it like, you know, it sounds sounds so obvious once you know it, like layouts, typography, padding, kerning, like where images should go, the bounding boxes, you know, the like the stroke weight.

Play from 0:50:46 8:88:88 0:50:46 You are going to like kill yourself forever to try and get Midjourney to do that. Like that you are going to be, you know, really, really fighting against what it's good at. And even just like if you just, you know, fighting from what it's good at from which training data, there's not a lot of that within the training data when it's architecture, it's not very good at that within its architecture. Diffusion by its very nature is, you know, from, you know, Gaussian noise coming from things that are, you know, every part of the image will have something in it. It's really bad at white space generally, you know, all these different things that it's just not You're fighting against And you know, typography, like just What do you What do you really want typography to be? Editable. Because like Yeah. Yeah. Yes.

Play from 0:51:29 8:88:88 0:51:29 When ever Have you ever done a project ever like once where you've got sent it to the client and they haven't gone "Oh, can you please just change this one thing or move the logo up a little bit or just change that one word?" You know what's crazy? I always think that nothing's going to happen. Oh my god. Come on. of into an You know, it's that perfectionist in you that's like When's it going to happen? It doesn't. Never happens. Never happened for me once. I'm How would it look bold? How you know, how how how about this? Can you can you space the letters out a little bit? Ah, no, I don't like it. Go back.

Play from 0:52:00 8:88:88 0:52:00 In professional design, we are in service of our customers. That's it. Like that that's it. The reason that clients come to creative companies is that they want this additional injection of that special sauce, that cultural identity, that thing that they don't have. That's why they externalize this. They want that that 20% extra which creative companies bring and it's hard to do. But we're not doing it. We're not to create art most of the time. We're mainly doing it to create design and design is communication, right? Like design design is there to help and serve that business need. So, I just think some of these arguments around like why you would use AI in these processes are completely moot when it comes to like you know,

Play from 0:52:46 8:88:88 0:52:46 Honestly, Fiverr said it really well this last week. No one cares if you use AI. Yeah. Kind of they kind of nailed it with that one cuz it's kind of the truth. It's always at the same at the end of the day, whoever is whoever is the end user of this product, whoever's paying for it, right? A lot of times, you know, I'm going to be quite candid, they don't care how it gets done. They just need it done, right? And that's Fiverr nailed it. I think they went I think they tried to take like a weird shot with you know, with that at like AI creators, but also at the same time like it it is the perfect message There's people that buy art and they they look at art and they'll sit there and you know, sort of like theorize around all the different pieces of it like if you're in a gallery or whatever. Then there's other people that like I need this graphic design done tomorrow. Like I Who do whoever does it, what however it gets produced, I just need it. So, like

Play from 0:53:33 8:88:88 0:53:33 there's always going to be a market for that. And I don't This is why I don't understand why people are so They just I think they feel like there's going to be like some sort of moral evolution in like, you know, big business or any something like that where They're waiting for AI Jesus Christ to come down. Yeah, like when when is morality ever been the case for a lot of these companies? What's Well, you know, it's I feel like it's with every new art, you know, and I use that loosely here, but people are just trying to like put it they're trying to compartmentalize it somewhere. You know, and they don't they don't truly know where to put it in some ways. Some of those people that aren't sort of moving fast on it and then also we talked about this previously too, but it's almost like the the photorealism of people is the most polarizing image type that you could create using these tools,

Play from 0:54:20 8:88:88 0:54:20 right? I mean, you're not going to know whether somebody used AI to create an icon or not or an illustration in some ways, right? Or you've taken that into Like that to me is like, okay, like there's there's no argument there. It it's it's just kind of getting into the the people aspect and then this kind of loops back around to that that story that you told at the beginning with the label AI or not. It's like if you don't tell somebody it's AI, most of the time they're not going to know. It assuming you're choosing the right type of image, right? It does come down to that curation and what you're prompting for and how that output looks, but at the same time unless you're pointing it out we're getting to the point now where it's almost indistinguishable if you're selecting the right images.

Play from 0:54:56 8:88:88 0:54:56 There is a meme going around at the moment which is like you're starting to notice less AI images and then you're like, "Oh, I'm starting to notice less AI images." Like that's that's the worry about it and listen, we're not building an externally facing product that we need to kind of neuter and like put these kind of guardrails in place to to say like, "Okay, here hey general public who are, you know, just generally like and can do unexpected things.

Play from 0:55:23 8:88:88 0:55:23 Here's this thing, go for your life with it." Like that's a hard proposition. When you want to say to like, "Hey, you know, highly skilled, very professional, global globally elite design team, here's a bunch of new stuff. Go and make sure we're doing good stuff for our clients. You generally, you know, get to some pretty good results. You can definitely then say like, okay, well, I can clone a voice in 10 seconds, and I can clone a person in 10 seconds, and I can do live portrait, and all these different things. And it's like the applications for like, you know, people doing dodgy [ __ ] with it are tremendous.

Play from 0:55:56 8:88:88 0:55:56 But, also, like, I just generally trust That's why I don't want to ever take the human entirely out of the process. My The idea for me is, you know, brief to first draft means I've just got better Basically, just got better briefs. Like, basic This first draft means for, and we know this. And this is like maybe where we'll kind of like finish up. Is we just done like a bunch of projects we've measured. And I think probably the best thing that we we're proud of, and I'm proud of, and I keep forgetting that we've made it, is like this measurement system internally. So, we have a baseline of every single type of creative 120,000 projects have gone through Superside, you know, we we in general, not AI ones.

0:56:30Real Results Using AI

Play from 0:56:36 8:88:88 0:56:36 You know, we've got an amazing team that just looks at like pricing and productivity of that before AI. So, we are looking at how do we create the most competitive prices that are fair to creatives that we can get this job done reliably. How do we codify this and productize it? So, we don't have to treat every single creative project like it's net new, which is a lot of design companies do that. So, we've got this baseline of this is how long we think it, you know, roughly it takes based off like hundreds, you know, thousands and thousands of projects. And now we've got that against the the AI side of things.

Play from 0:57:06 8:88:88 0:57:06 And we we have built this system where we can pretty globally look at, okay, within presentation design, we know within this part of the process is really good, and this part of process is really bad. Within AI, you know, within ads design, we know it's good for key art, great for headlines, good for subtext, you know, sub sub headlines. We needed to build a system around doing layout, but it's good for translation. Great, we can like codify it in this way. And these things like measuring these things, truly measuring these things. Measuring like costs, speed, quality, feedback, and then finally performance. So, we have coming next month an announcement coming out about what we're doing in this space when it comes to performance. We've been running in stealth for a little while and we've been doing some interesting like annotation projects around like

Play from 0:57:52 8:88:88 0:57:52 what we've got with the data there. But then if you start to look at the If you start to look at the end result then it starts to take the subjectivity out of all of these conversations. So, like maybe in some dimensions AI is worse. But maybe in some dimensions it's better. And unless you have done it enough and unless you've measured it in a way that you feel like you is robust enough, then you are having opinions and opinions are cool. I like opinions. I have an opinion. Doesn't mean what I say is true at all. But if you can give me some corpus of information like this like body of work and and this evidence, then we can start having more sophisticated conversation around it.

Play from 0:58:32 8:88:88 0:58:32 It's That's the difference, man. That's the biggest difference. Hey, clip that right there. Like that was that's that's it. Um hey Rory, I'm going to let you wrap up with the final question. I do have one final question for you myself, Phil. Thinking about agencies or creative teams in general outside of Superside. A lot of these teams are at sort of infancy stages with how they want to adopt this. You know, is there any piece of advice from an operational standpoint or flattening the learning curve or getting people up to speed or anything like that that um you can leave folks with because I think a lot of these people aren't sure where to start, right? Is it Is it hiring somebody to to train their whole team? Is it um you know, them just outsourcing this in general, right? Like how should they think about it? You know, somebody that's kind of been through the trenches, is there any kind

Play from 0:59:18 8:88:88 0:59:18 of piece of advice that you can offer? Depends how big your business is. That's the general truth, right? Like mega businesses, okay, my advice would be come and use our AI consulting service because we've got a bunch of people there that are really good at that. If you're a big If you're a big company that is really regulated, we know it's difficult. And it those things are just difficult in general. If you're, you know, mid-size agency, don't over-regulate is my number one advice. Do not put like the absolute iron curtain down before you even understand what it means to your business. I still have loads of stuff running on credit cards. Procurement hate me. I have to have fights with them every every now and again. I've got a really great partner in our legal team.

Play from 0:59:59 8:88:88 0:59:59 Actually, procurement don't hate me. I'm doing them a disservice. I love you, procurement. You're very nice. Um but like, you know, don't over-regulate before you understand. So, set your risk tolerance to be higher than you probably would do if you were, I don't know, procuring a new data system or something where you would then go through like, is this SOC 2 compliant? Just understand before you regulate. We're getting to the point of being more regulated now we understand more because now we go, "Cool. This subset of people need regulation." So, that's then service them in that way.

Play from 1:00:28 8:88:88 1:00:28 But, we still I still have a bunch of stuff running on credit cards. I have a bunch of different pilots going on at different times. We are still in pilot with 2 years in. Still in pilot phases on on the next things. We go back to piloting constantly. So, my advice is don't over-regulate. Buy Midjourney. Buy Ideogram. Buy Runway. Buy all the things. And then apply it. Don't do the like Try and move to execution as quickly as you can cuz you'll learn way more from customers than you'll learn internally.

Play from 1:00:59 8:88:88 1:00:59 And then, you know, also create the psychological safety for people to say like, "It's cool. We are cool with this. We are cool with you screwing up. We will give the clients back the money. We You won't lose your job. We're good. Be sensible, ethical, lovely human beings that we know you are. That's why we hired you. We We We know We have trust in you because we trust in our own hiring processes. That's what That's what companies should be like cuz if they're not trusting their employees to do the right thing, you've screwed up somewhere along the lines cuz you haven't hired the wrong people. But don't over-regulate. And then when you find something that works like triple down on it. Really really go in and make it something that whether you want to be the AI company and you want it in your brand and you want it in like I can't advise on that cuz I don't know if that's good or bad or up or down, but like you know, give people the chance to

Play from 1:01:46 8:88:88 1:01:46 try. If you are just using Microsoft Copilot, you're screwed. Not because it's bad, because you're not thinking wide enough. If you're just using what's readily available and comes within a freemium model of what you already have, go wide before you go narrow. Like you have to understand these things. It's really the place like there's no shortcut. That was going to be my last question because I know you can get your hands on a lot of these tools and maybe some that we don't, you know, have the direct exposure to. Give us a Give us a couple ones that you're liking right now.

1:02:00New Tools

Play from 1:02:17 8:88:88 1:02:17 Just a few maybe Yeah, Flu Nugget. What? It's Flux. Just Flux? That's it? Yeah, yeah, it has No, no, I mean so Flux like for me is this step change. I haven't played with SD 3.5 yet, so I could be I could be wrong, but like Flux I I was I was starting to really worry about what we were doing about fine-tuning with models. So we do have this whole model fine-tuning service that we will you know, take a customer's data set and like we'll make it into like a codified brand style that you can then roll out. And I was just getting worried that like we were just going to hit this threshold of like we're tied to Stability AI's kind of [ __ ] of what they were doing before.

Play from 1:02:54 8:88:88 1:02:54 And Flux came out of stealth out of nowhere. And now I can train like product Loras and I can train like, you know, really like great aesthetics. He's kind of nearly there with text sometimes in certain things like so Flux is the the one for me that like get to know it and its ecosystem more. ComfyUI, get on the waitlist for the new um for the new native Oh, is that waitlisted?

Play from 1:03:22 8:88:88 1:03:22 I need to I need to get on that. It's coming out. All right. Okay. There we go. have they have they this is what version one of ComfyUI is coming out, basically. They have come to the first official release of ComfyUI. Everything up until this date has been this like kind of early market testers and stuff, but like ComfyUI, if you want to get into if you want true controllability, node based controls where you can start to build your own workflows is really a lot of the thin layer shitty AI companies are just built on top of They're just wrappers on top of ComfyUI.

Play from 1:03:52 8:88:88 1:03:52 Yeah. Like a lot of the thin layer, you know, chat to your PDF companies are just thin layers on top of of of chat GP you know, of large language models of GPT-4. So, try and know the source. At least as close to the source as you can get. I would just try and get as close to the source of the things that interest me as possible. Don't be too abstracted into like the final application layer. Try and find something that you can make and build yourself. The final part, I would say, follow the money, right? So, there is way more money going into large language models than there in there is into image models.

Play from 1:04:25 8:88:88 1:04:25 And so, and what kind of large language model do? A large language model can code, and what is Design? It's mainly code. So, you know, if you're going to build something, probably build it on top of something that all of the money is going. Like the absolute weight of cash is going to OpenAI and Anthropic. And so, you build something that's on top of them, or you're really interested in like Anthropic is doing some cool stuff.

1:04:45Future Vision

Play from 1:04:50 8:88:88 1:04:50 Yeah. Yeah. Sam Altman's dropping some cryptic messages about like the winter sky at night, which is meant to be this Orion model, which they're going to be dropping soon. So, he's you know, he's doing his same thing he's doing before. It's his whole playbook. Follow the money, right? So, if you're going to build stuff, or build things into your operations, I'd rather stand on top of Sam Altman's gazillion-dollar investment and do something cool on top of that. And then maybe what people are doing in the image space, which is you know, a hundred million dollar investment. So, if you're going to build up proper killer processes within your business, make sure you get some life like stuff in there as well.

Play from 1:05:26 8:88:88 1:05:26 That's great. That's great advice. You know, and and hey look, as we're wrapping this up, I think also if you are one of these agencies, you are one of these companies that is looking for creative services and you have that need, you don't have a design team, you need to move fast, you need firepower, you need juice like Super side is it. They've got massive massive logos, Salesforce, Meta, Google, Red Bull. Am I missing some? Amazon, I think, right?

Play from 1:05:54 8:88:88 1:05:54 there's a bunch of uh yeah, cool cool logos, a bunch of cool people and we're incredibly like ego free to the kind of work that we do. We will jump in at the small end, we'll jump in at the large end. Completely global distributed team. 64 different countries, 16 different time zones. Yeah. And they yeah, they nail it consistently and and really amaze me all the time. So, yeah, get in touch. We'll make sure to drop a link in the description for you guys. Rory, anything else you want to say before we wrap up, sir?

Play from 1:06:23 8:88:88 1:06:23 I need like six more hours with Phil here. I know. Come back. There's too much to talk about. We'll add Phil as a third third co-host here. I'm on I'm on relax mode right now. Like it's getting towards 6:00 o'clock. Yeah, man. We'll We'll let you go. No, it's cool with me. I'm just going to be It's diminishing returns after this point. I'm going to start really slating people and going off the rails soon, so like Love it. Well, thanks Phil for coming to hang out with us for a little bit here.

Play from 1:06:50 8:88:88 1:06:50 You know, we got a lot of cool stuff coming up. I think it's awesome to have this range of conversations, you know, from what we talked about with Billy a couple weeks ago to then you and I going into some of our favorite tokens the other day. Now back to Bigger at scale. So, we're trying to as we tell everyone every episode, we're trying to bring a variety of topics here. So, you know, we're not the same stale boring stuff every time. Yeah, man. I mean and we're bringing on the best here. Phil's great. We've had We've had Billy, we've had Chris. We've got some other guests that we've got lined up. So, thanks for joining us guys. As always, we'll be back next week with a new episode and we'll just try to come up with that on the fly, but we guarantee it'll be valuable and uh we'll catch you guys in the next episode.

Play from 1:07:31 8:88:88 1:07:31 Like and subscribe. Thank you. Thanks.