Host 2 (2:53)

Yeah, tools have come a long way. What's actually one of the most recent tools you've been blown away by in terms of its ability to create creative assets.

Rory Flynn (3:02)

So we to me has been like the first like we. Aha.

Host 1 (3:06)

How do you spell that?

Rory Flynn (3:07)

W, E, A V, Y, AI. It's very similar to Comfy ui. It's a little bit more user friendly, it's node based, it's a workflow building tool, but it also has access to API access to like a hundred different tools. So it's like you can use any of them. It's just how you creatively sort of string them together and build stuff. So I love it. It's simple, it's clean, it does what I like and it's easy to share and it's scalable. So someone builds a workflow once the whole team can have it, whole team can use it. So you don't need the entire team working on it, figuring it out. You just need one good person who's totally interested and is like just laser focused and you can build a whole team around that. So I love this tool. Not a paid shill for them. I just, I just really like it.

Host 2 (3:55)

Very cool. It's like a zapier for creative workflows.

Rory Flynn (3:59)

Yeah, yeah, it's very similar. The other thing that is, that's pretty interesting about it is to me it's like a precursor to agents. So if you can go in there and test and build workflows, you know, to me it's a little bit easier to go figure out in weave than it is in like N8N or something like that to just like get the flow down, get the process down and then that can be translated. You know, we've vibe coded a few tools off of weavy workflows and things like that. So it's like a great testing ground. It's a great team tool. It's a great creative tool. I just think it's super versatile right now.

Host 1 (4:28)

So you're going to walk us through a bunch of this, which I'm excited about, but kind of my question before we get into that is like, what's the Aha. In all of this, like, you talked about, like, you were up for eight hours straight. You've been building a lot. Like, what's the moment for you where you were like, whether it's set, a creative, a campaign, whatever. You were like, oh, like, this is. The game has changed.

Rory Flynn (4:48)

For me, it was legitimately the first time I got asked to speak on stage. It was about four months into doing this stuff. I was so green and like, you know, I'm talking about building, like, mood boards with Mid Journey and stuff like that. It wasn't revolutionary, but I did it at a performance marketing conference. They asked me. They. They begged me, and I was like, I, you know, begged, maybe too hard, but they asked me about two weeks ahead of time. They wanted to slot something in A.I. you know, I take that back. And I was like, guys, I have never spoken on stage, let alone talked about this stuff out loud. You know, like, I've only posted about, like, fun stuff on social media. Never done, like, the actual business side of it, the operational side, because we had started building, like a business case around it to make sure that it was actually valuable right at the time, which is 2023 ish, early 2023, which sounds crazy now, but I got up on stage and first off, I was about to die. You know, like, I was a really. It's really. That was not my thing. They were like, they're like, oh, you're gonna be on the side stage. And this. I'm like, okay, cool. How many people's out there? Like 200. I'm like, I think I. I think I can, like, maybe handle that. Then the day before, they're like, oh, you're gonna be on the main stage. I'm like, how many people is that? They're like 2, 000. I'm like, that. That was not what we agreed to.

Host 1 (5:59)

That's not part of the agreement.

Rory Flynn (6:00)

That was now. So I'm having, like a whole meltdown. But I got up there and just. I just kind of like, let it rip, you know, I didn't really have any public speaking experience, just sort of like talked off the slides, just like rapid fire, one thing after the next. And I got hounded when I got off stage. Like, I had no business at the time. I had no sort of, like, I was sort of on my own, just kind of figuring out what I was going to do with it. I had no offer. I had nothing. And the minute I got off that stage, it was just like the doors open to everything. So I was like, okay, there's something here.

Host 2 (6:30)

What were you showing? You were showing, like, prompts to create graphics on midjourney.

Rory Flynn (6:35)

Yeah. It was basically walking through the process of how we operationalized, like, prompting at scale. So it was like, how did we take an image and break it down into its core parts to then build basically, prompt formulas so that we could scale that out to. Anyone could use it?

Host 2 (6:51)

Like templates.

Rory Flynn (6:52)

Yeah. And that was. There was sort of like that core idea of deconstruction, and then that. That led to everything. It was like, oh, this is how everything works with AI. Sort of like the. The gatekeeping sort of option there.

Host 2 (7:03)

Well, its ability to reverse engineer assets and create templates that then can be reusable by people is one of the most powerful skills it has and probably the most widely underused. Like, custom GPTs are widely underused because I don't think people realize you can create onboarding docs for them and then just have them be experts in niche sets of skills. You can just have a team of experts. But that's the other one. Something Kip and I have talked about is it's a time of, like, great change, and everyone is like, oh, my God, everything is going to change, and my role is going to be obsolete, and what do I even do with my life? Whereas it's also a great time for just doing. Because you're just like, I'm going to do stuff. And then I got asked to talk about that stuff and now have an entire business around that stuff.

Rory Flynn (7:45)

Right.

Host 2 (7:45)

Because you were curious. You were like, this is super interesting. I'm just going to go and, like, figure it out. And I think it's a great example for our listeners is like, now is the time. If you have curiosity and you have tenacity and you have persistence, you can actually just do stuff. And that stuff might blow up pretty quickly.

Rory Flynn (8:00)

It's like, if you like to create, how do you not like this?

Host 2 (8:04)

How do you not like it? 100. You know what I mean?

Rory Flynn (8:07)

Exactly. To make stuff. Like, you've never been able to make more stuff. Whether, you know, it's digital, but whatever, it's making things.

Host 1 (8:13)

It's the golden age of making things.

Rory Flynn (8:15)

Yeah.

Host 2 (8:16)

Crafts.

Host 1 (8:16)

Okay, so speaking of that, you've been grinding away and making stuff for a few years now on this topic. Right.

Host 2 (8:22)

And.

Host 1 (8:23)

And you're going to show us kind of the best of the best, and you're going to show us some of, like, the most important things you've learned. And so I'd love for you to kind of Take us behind the curtain and do a little bit of that.

Rory Flynn (8:33)

Totally. I can show you sort of like this slide and these things that I've had have not changed in three years. They still work, right? Like this is this. They still make sense. Like I use them almost every day. So that's where I think like something like the deconstruction, right? Like I talk about sort of the three main skills a lot of times that I think are super important in this era of AI. It's like not. You don't have to know every tool, need to have like core competencies in a few areas and you can use any tool. It's like being able to drive any car, you know, it's like the same thing. So that's sort of how I like to look at it. But really like, you know, going back to again, tying into that story. If we're trying to create photorealistic assets, right? Like that was our main focus, especially on the marketing side is we need things to look real, right? Like it was cool to make the futuristic, like spaceships and stuff like that. But like we needed stuff to put in our emails. So, you know, we started using this type of thinking. It's like if we want to make a photo, like what is a photo? This sounds so like intrinsic and out there. But it's like, what is a photo? Like what actually are the core categories? Like what's actually controllable within a photo. So you know, this basic little structure here was sort of how we looked at it and it was like, okay, you know, these are also essentially like what I call like the non negotiables. Like in AI, if, if you don't prompt for it, it's going to be provided for you anyway. So this is sort of like the core elements and it's. This doesn't have to be how you prompt every time. You can do more, you can do less. It's just like if you wanted to control the main aspects of an image, like this is what you'd look for, right? So we always start with like that shot or photo type, depending on, you know, the perspective, where it is, how it's captured, right? So it's like the difference between like a close up image and a drone shot is like radically different. So again tells a totally different story. Your subject and your action. Like who's in it? What are they doing? Again, like the core sort of idea of the image. You have your environment where it takes place, the color scheme. It's always going to be colors in an image of Course, even if it's black and white, those are colors, right? This is where we sort of hit pay dirt early on, was this utilizing different cameras and lenses, because, like, if you use different cameras, you get different visual esthetics. So depending on what the brand was, sort of like what their vibe was and their visual aesthetic was the difference between prompting for an iPhone photo and a Polaroid. Like, it's a totally different visual signature. So being able to control that, look at things like film stock, things of that nature, to sort of like guide the output was really important to us. And then lensing, of course, you know, the difference between using a 6 millimeter lens and a 200 millimeter lens, you know, you're going super wide and tight. If you're going super narrow and in, like, it's very different. So, like, those were really helpful for us. And then of course, we look at things like composition, how things are constructed. There is a visual hierarchy in sort of every single image. If it's professional, at least, you know, the mood and emotion. Like, that's always been big to us because it was the vibe. Like, how do we set the vibe? How do we, like, sort of bring that out? You know, we need to give it a little bit more energy than what was original. The original, like, sort of AI idea of that. You know, lighting, of course, every single image has lighting, whether you like it or not. Like, if you didn't have lighting, there'd just be a black image. So, like, those are like the core sort of things again, of course. Textures. There's texture in every single image. Details and modifiers. I put this in here because, of course, there's always other additional elements when you look at things like props and, you know, cars and whatever things might be in the background. So it was understanding that this is what's happening in every single image, no matter what. So why don't we build structure around that? So when we did that, we sort of started thinking about it like visual building blocks, where it's like, okay, each one of these things makes up the image, right? But if we're going to prompt and we sort of isolate each variable in this equation, which is creating an image, then it's also, you know, from the iteration standpoint, which we all know happens all the time, then if we have, you know, lighting represented in our prompt and we don't like the lighting, we know it's in there, we know what to go change. We can do it quickly, we can iterate faster. So we're not endlessly re prompting right?

Host 2 (12:32)

Can I ask a quick question, Rory? Think if I was the listener. I think this is really cool because what you're going to show is the amount of time and effort you put into actually creating context for the prompt. Like this is all context of what good looks like. And is your background photography or did you use AI in of itself to actually craft these building blocks? Like how did you know these are the most important parts of a photo?

Rory Flynn (12:58)

It was definitely a lot of trial and error in the beginning. I have a graphic design background, so I have probably half of the technical knowledge of a photographer because also working with a lot of brands in this space, we would always running photos, photo shoots, things of that nature. But you know, learning more of like the technical side of photography helped a lot. You know, again, like I didn't know what each specific lens did or what a focal length was prior to AI, but it's like forced learning because then I was like, I know how this looks from looking at it all the time and working with it as a designer. Just didn't know like what the term was for it. So starting to understand what like bokeh was or things like that. I'm like, okay, now I know what this is. I've been looking at it my whole life basically, but didn't know that there was a term for it. So definitely brought my interest in photography way up now. Yeah, now you can sort of. I don't want to say I can speak the language, but like if someone says something to me, I have an idea of what they're talking about.

Host 2 (13:52)

Yeah, makes sense because you can also your interest goes up when you can like do real things much quicker. Yeah, like the payoff and learning anything is like doing the thing that's like the diplomat. But it usually takes you a long time to, to get to do that. And whereas in this you can just sit in your room and do a thousand photos overnight.

Rory Flynn (14:09)

A hundred percent. And I think that's what sort of got me early on. It wasn't just I was interested in the business, I was like interested for myself. Like I use this as like a decompression sort of thing early on. Just like let me dump all of my day's frustrations into an image generator and like let's just create, you know, and it was just, it felt very therapeutic early on. Like this is fun. Like I'm not doom scrolling on Twitter. Like I'm like building little worlds in mid journey and like having a lot of fun with it. So yeah, you know, underrated thing. I Think of all this AI stuff is just like, if you find a tool that you like, just go have fun with it.

Host 2 (14:42)

Yeah. Just go create it.

Rory Flynn (14:43)

It doesn't have to be for a purpose.

Host 2 (14:44)

Yeah, I agree.

Host 1 (14:46)

Okay, keep going. I love what you're really telling people is, like, whatever you're going to. In this case, images could be anything. You gotta, like, understand how it works, reverse engineer it, get the building blocks. Right. Because you need to know what variables you can control. Because the variables you can control are the things that are ultimately leading to the output and whether you're getting what you're trying to make or not.

Rory Flynn (15:06)

Correct. 100%. And that's this sort of thinking right here. This will work for video, this will work for music, this will work for system prompt. This will work for anything. It's just like, what needs to be present and what can be controlled. And then from there, even if you're asking an LLM to do this for you, it doesn't have to be like, off the top of your head, you know, you can sort of like build the core building blocks and then roll from there. Because once you have these, then everything becomes a lot easier. Right.

Host 1 (15:34)

What's funny, in that slide you were just on, literally somebody watching on YouTube could have just screenshotted that slide, right? And then had some image and put it in chat GPT or anything and say, hey, tell me what the building blocks are for this image and learn that way. Yes, within five or 10 minutes, you would have a pretty good understanding of what this all actually means and how it works.

Rory Flynn (15:56)

100%. I think that's the best part of it. It's relatively simple. Once you have the formula correct, that's just getting the formula.

Host 1 (16:01)

I love that.

Rory Flynn (16:02)

That's the hard part, because that's what we did was like, oh, okay, if we. If we have the building blocks right now, we can build a prompt formula which becomes essentially like Mad Libs for, you know, using AI. Like, I don't have to. I can fill this in myself. Or I can ask at the time, chatgpt to do it for the team who might have not had as much experience in the design field or in the photography field, like people that just wanted to create as well. So basically just took those building blocks, threw them into a prompt formula, separated them by commas, and let it run from there. Because it was like, all right, well, if this is like the operating structure and AI, you know, if we were using ChatGPT at the time, could help us fill this in. We could also ask for 10 prompts, we could have asked for 50 prompts for it to just fill in the blanks.

Host 1 (16:46)

Totally.

Rory Flynn (16:47)

So it was like, okay, we can start to scale this now. So this little piece here basically started an entire business. This, this little prompt formula, which is kind of crazy that it was that, you know, at the time I wasn't looking at it like it was that important. But, you know, I see how much I use it now in just about everything. It's obviously updated and been tweaked and manipulated a million times. But this is sort of like the distilled version. Like, I also like to think about this as like prompting for, especially for images, doesn't always have to be first shot stuff. Like, I like to give the condensed version. So I don't want to go overboard and write these like five paragraph prompts. Because isolating what's wrong majority of the time and troubleshooting is way harder than you think because it's just like one little word here could screw the whole thing up out of three paragraphs. So, like, give it the condensed version, you know, which is like, you know, if you boil a pot of water, like all the way down to the bottom, like, that's what I want to give it first. And then I'll keep adding water to it to sort of like, you know, make it fit.

Host 1 (17:40)

So I think that's a really important point that you're being start simple and add versus starting complex and taking away.

Rory Flynn (17:46)

Because also, you know, it's much easier to start simple and then you see what you can do again. Back then too, it was, you were not getting the type of hit rate that you are now, which is everything is a lot closer to getting it like 100% on that first shot. Back then it was like it was slot machine, like, oh, maybe we'll get it. And you know, now it's definitely more. So the thing was like, you'd throw in these prompts like motorsport photography, Red Bull 1 F1 car on a racetrack. You have the color schemes here. You'd have the warm tones, the 35 millimeter shallow depth of field, dramatic sunset, backlighting, center framing, motion blur, which just seemed like this random collection of words with no rhyme or reason, right? But you'd run it and then you get something that was like, oh, okay, like this is actually listening. And you realize that you have, you know, motorsport, photo, racetrack. Like that's represented, right? Warm tones, you have that yellow orange hue. You obviously have the car driving. You get the radial blur on the tires. It's Moving, you can tell that you have the blue, red and yellow obviously represented 35 millimeter. Shallow depth of field, if you're not familiar with that, it's basically a wider angle storytelling lens. But shallow depth of field is like portrait mode on your iPhone. So you have the subject in focus in the foreground and the background blurred center framing. Right. We have our motion blur in the background. You can see the stuff's moving from left to right. And then we have sunset and then we have backlight, which means it's behind the subject. So like seemingly just random words, but you have control. It's like we controlled the non negotiables so it listened to us. And then we can build upon this if we wanted to. Like, now we know that it listens, right? Now we can go iterate it. So some of this stuff, it was like, this was sort of super, super helpful because then the team could be like, oh, okay, I see how this stuff works. And now it's easy to iterate. There's so much more stuff like that. It's kind of crazy at the end of the day when you start going with it.

Host 1 (19:39)

Look, Rory just gave you the playbook on how he uses AI. Now he's giving you his exact workflow so you can copy him. You give AI one brief, it spits out 100 assets, social posts, email copy, blog drafts, you name it. This is literally Rory's process packaged up. So you can start using it today. Scan the QR code or click the link in the description. Now let's get back to the show.

Host 2 (20:02)

One quick question before we get into some of the other things you're going to show. I'm curious. Right. So that actual prompt took you a long time to figure out and refine, I assume, Right? That is kind of like your usp. That kind of drives a big part of your business. How do you think about giving prompts away? Like, if I was watching this, I've got a huge amount of value. Next, I could just clip your prompt. And now I'm kind of as good as whatever version of you was when you were doing that prompt. I'm sure you have, like, more sophisticated ones now, but I shortcut my learning pretty rapidly because it's not the same as, like, when you create an ebook. To teach someone how to do imagery, they have to like, still consume the ebook. They have to like, learn those skills and they have to get better at that craft. Now, prompts, I can literally basically take your skill. It's like the, you know, I always loved, like, The Matrix, where he's like, I want to learn how to fly the helicopter goes. And he just, like, learned to fly the helicopter. Like, prompting is quite similar in that I can be as good as you. Where you were then by just taking your prompt. So how do you feel about that? Because I think about that a lot in terms of some of the prompts that I give away.

Rory Flynn (21:09)

Yeah. I'm honestly sort of have, like, a very open nature about this. I feel like everyone should be learning this. I feel like everyone should be on board. I feel like, a sense of responsibility also, which sounds weird to, like, especially to my audience, who've been following me for a long time. And I feel like I have to give a lot because as a graphic designer, like, I was replaced by Fiverr. Right. You know, like, that was. I felt this before, like, once Fiverr came along and I was no longer necessary because I could not compete with Fiverr. So I'm mad at the people that worked at Fiverr. I was mad at the people who are making decisions, you know, just going with the cheap option, because that's just the cheap option. Right. Maybe I wasn't that good. That could be my fault as well. But I feel like a lot of people are feeling that same way right now. They're feeling that same sort of, like, burden of, like, this is coming and, like, I might not be needed. So I feel like if we can get everyone at least to a good level of understanding of how to do this, then they can be sort of. I don't want to say irreplaceable, but they're valuable, you know, valuable to a company, valuable to themselves. They can go and build something if they want to. Like, you know, I seemingly just dreamed up out of nowhere, but I want to help. It sounds cheesy, but I really do because I feel like everyone is overly stressed about this. And if we can alleviate and make people see. See, like, see the Matrix, you know, like, when Neo actually sees it for the first time. Like, I love getting people to that moment. Once they get there, I know they're fine. It's just like, you have to get to that moment of, like, oh, I see what's going on here. Yeah.

Host 2 (22:39)

So makes sense.

Rory Flynn (22:40)

Hope that answered the question. Yeah.

Host 2 (22:43)

Yeah. Okay. It sounds like you've kind of progressed onto some other workflows.

Rory Flynn (22:48)

Yeah. Weavy now is, like, my playground. Right. Like, Weavy's taking that sort of idea and then blowing it out. It's like, how do we take that idea of, like, oh, now we Have a formula and then how do we blow it into a number of different sort of realms and how do we sort of create the multiverse? Because I think that's the place that we're at now, which is really interesting to me. An image isn't just an image anymore. Like an image to me is like the foundation of AI. And that's my own opinion. But like an image can now become a video, an image can now become a 3D asset. An image can become a piece of text, it can become code, it can become a gif. It's sort of like branching, right? Like you can go into a video, extract a frame and then take that in a million different directions. It's all sort of like this now infinite branching of reality which can be overwhelming. So it's kind of hard to, you gotta like narrow it down a little bit every once in a while. But the weavy stuff was really, really important. Like I'll show you one workflow to start which is where basically again like solving common problems. Like we know, I know a lot of people in the E commerce industry, especially in the fast fashion space, right? Like they have, you know, constant turnover of product. But a lot of times to get that product photographed, uploaded into the website, you know, live and ready to go is not easy because it's expensive. And if you're doing that with 4,000 SKUs, like it's not necessarily the most time efficient process if it's only going to be live for a season or a couple months. So it's something that we did early on. This was sort of like the test case to be like, oh, does this get us somewhere? Does this get us into a place of like okay, we can start to solve problems like this? Which met with a lot of good response. So it was something that we started to build upon like this was the building block for everything else. So this was sort of like the origin of it where it was like, okay, we can take a way to sort of art, direct your model, control your clothing, put them together and then iterate the photo shoot, right? So this was a very at the time, well, I wouldn't say simple. This was just like an easy way to show it. Definitely a more efficient way that we've sort of reverse engineered from this. But it was sort of the building blocks, right? So this thing looks really crazy. It's not necessarily that crazy when you get into it because it's just duplicated stuff. But. But if I was going to show you what was going on here, this is basically Modular prompting just so people.

Host 1 (25:06)

Who are not maybe marketing dorks or AI dorks. What you have here is essentially you're using wevee to build a workflow. You've got some workflows here, but that workflow is basically to replicate a traditional physical photo shoot using AI.

Rory Flynn (25:21)

Yes.

Host 1 (25:21)

And it's to try and to solve that exact problem. You were saying just a few minutes ago, it's like, oh, how do you shoot 4,000 items in a cost effective way that looks good for those items that are only gonna be on a website for like 90 days?

Rory Flynn (25:33)

Exactly.

Host 1 (25:34)

And that becomes really painful. And you were like, oh, there has to be a better way. And so you built this as kind of a foundational workflow. And you're doing this for shorts, but you could wear clothing or shoes or whatever, but you could do it for literally anything. It's just like this is a workflow to replicate a photo shoot. If you were gonna do a photo shoot for something, you could then now replace that photo shoot with AI. Now you're going to walk us through, I think, a little bit of like the building blocks of the workflow.

Rory Flynn (25:59)

Totally. That's a great way to sum it up. That's exactly the thought process behind it. So, like, again, this looks super confusing with all the spaghetti strings, but when you put it together, this is really just modular prompting. So it's using essentially sort of the same building blocks that we talked about here. You know, our photo type, our shot type, you know, we have our gender perspective, age, ethnicity, body types, or doing like a, almost like a video game style, like create a character right where we're running it into this thing called a concoctionator, which will just combine all these pieces together and then use that as a prompt to generate the image. Which is cool now because we can also take our clothing that we want to upload into here so that he's wearing it. Every time we have some prompt, we have a system prompt here, which if anyone's not familiar with system prompts, it's like the prompt that guides a custom GPT so it works in the background, but it basically says do this every time. It's like custom instructions. So this one's not even that crazy. It's just describe each article of clothing essentially. So we have in text and visual form as well, to help sort of provide it. Now this was done before the Nano banana, the Flux Context craze, the C dream, everything like that. So had to be a little bit more intentional with what we were doing, is the foundations still work. Which is why I want to show it. Because sometimes the best way to get it when you're troubleshooting.

Host 2 (27:17)

Can I just touch on that to kind of walk our users through what's happening. So the first workflow you showed us through allows you to create the quote unquote model you would usually have. And then the one below that is, I assume you're dynamically like wherever you use this workflow, you're giving it a bunch of imagery. And because you're going to take that imagery and use in another prompt, you need a system prompt to convert each image into a text based description that you can then feed into a master prompt that's going to combine the model with the clothing.

Rory Flynn (27:50)

That's 100% correct. That's a way better way of describing it than I just.

Host 1 (27:54)

That's a great. I actually didn't even understand it that way. Kieran, that was a really good summary. That was awesome.

Rory Flynn (27:59)

That's perfect. Yeah. So we're using like LLMs here in Weave you have the option to do, you know, basically use any of the ones that are available between Claude, Gemini, Llama. But it's just a great way to utilize this. Instead of me describing it via text, just giving it a same idea like a system prompt, just describe this stuff and we'll use this one over and over again. So we run that to again to combine it, which is the prompt of the model and then the prompt of the close plus the image of the model and the images of the close and we put them all into. In this case it was ChatGPT. Again, this is built like about five months ago. So it's just the core idea is what works here because I'll show you on the next one sort of how that expanded from here. So it was like, okay, we have our one image, we upscale it. So now we have, you know, we have our shirt, our, our shorts, our shoes. It's pretty cool. Local. Yeah. And then that's how we started to build batch. It was like, okay, once we have this, it's the one image again. Like once you have again one prompt equals a thousand. One image equals a thousand. Same thing. We'd branch it into the realities. Like this one is like a building block of batch processing here. What we're doing essentially we're creating a prompt saying describe six different camera angles for this model, focusing on creating a 360 degree view. We give it the idea of the perspectives. We want a front view, side profile facing left, side profile right three quarter angle, etcetera make sure it's a full body shot. We want head and feet and keeping the environment exactly the same with. Let's keep these prompts shorter than 30 words and separate them with a star. This is like the whole batch processing thing. So the goal here is to take this, create six more prompts so they come out in one single output from an LLM. This might sound confusing and try to, try to keep it simple, but the reason we want to have it end with a star is so we can use this thing called array, which will split every prompt by the star. So every time you see a star in here, that means separate new prompt, right? So then we're able to split them out and then we can run them to these different list selector views. So it's basically each individual one is a prompt and then we can go run them simultaneously.

Host 2 (30:22)

You know what's hilarious about this? I have a couple of friends actually who are like real photographers and go and do fashion shoots and I, I suspect if I had told them like a year ago, I think the core skill you're going to need to really learn is like, you're going to really understand automation arrays. Technical.

Rory Flynn (30:39)

Yeah.

Host 2 (30:40)

Like, what are you talking about? That's nothing to do with me. I'm like doing all the art and you're like, I'm splitting these into arrays so I can use them and auto create different camera shots. It's. It really is an example of like how every single field because of AI becomes so technical.

Rory Flynn (30:57)

100%.

Host 2 (30:58)

It's like art, but there's a lot of science, right? There's a lot of science integrated in all of this.

Rory Flynn (31:02)

It's a weird blend. Like I, I kind of love it where you can be creative but technical at the same time. Which, you know, I think there's probably limited fields and where you can do that but like foundationally then, right? Trying to teach something like this, right? Like this is basically like a batch processing engine. So like this will work. You just change the prompt, right? Like, then I just duplicate that three times and just change what I'm asking for in the prompt, which is like, you know, I want lower half shots closer up. I want, you know, more action poses or I want more lower half shots. Or this one might be, I want him like in action, doing something, walking, not looking at the camera, more like B roll style. So this overall is cool, but like people don't want to look at that most of the time. So having these outputs here, which is great, just attaching an output. So it's telling the system that I want to make this sort of like an app. Right. Like we can make this an app. We're just. It's really good where we're putting in. Put in like the actual description of the model. We do the description of the close and then we push enter and then.

Host 2 (32:07)

It'S kind of mind blowing.

Rory Flynn (32:08)

It all comes out at once.

Host 2 (32:09)

Oh my God. Like every single retail company who does these kind of like asos and all these. That do all these large amounts of photography for every single piece of clothing they have. I mean like their world is just sizably different.

Rory Flynn (32:26)

Definitely.

Host 1 (32:26)

And hardly any of them are doing it.

Rory Flynn (32:28)

Right. I mean there's some. There's definitely some.

Host 1 (32:31)

You're doing it for some. Yeah, but that's it.

Rory Flynn (32:33)

This is like the watered down version of what that is. Because that's obviously so much more quality control than is here. This is just a sort of like an example piece of like taking them run with it. If I'm. Like I said I'm. If I'm giving people stuff, like, please go ahead and use this. I mean, I'll definitely share these workflows with everyone. You know, hopefully if you guys drop it in the comments section or something so everyone can take a look at it.

Host 1 (32:51)

Yeah, we'll put it all in the comments and everything so that they can find you and find these.

Rory Flynn (32:55)

Totally.

Host 2 (32:56)

And to be clear, that kind of workflow was generate the model, generate the clothes, and then you use this batching process to generate that person doing different actions and different shots of their. The clothes on the body. That person is like all AI generated.

Rory Flynn (33:09)

Right.

Host 2 (33:09)

Like you haven't uploaded a photo of a model. No, that has been created.

Rory Flynn (33:13)

Yeah, controlling the character. Right. Like creating the character. Clothes are the only thing that might exist in this.

Host 1 (33:19)

The only non AI generated thing is the clothing that you're trying to, you know, demonstrate, basically.

Host 2 (33:24)

Have you done anything like this with video? Because the tool I am obsessed with right now and we're trying to build out some cool stuff is V3.

Rory Flynn (33:31)

Yeah.

Host 2 (33:32)

Now, you know, you still have to keep character consistency across eight second clips, which you can do using flow. But I'm wondering, like, can you give us a little quick demo of anything you've done? Video wise?

Rory Flynn (33:42)

Video wise. So a lot of this stuff what I'm doing here is precursor to video. So all of these. And I would use, I would take this essentially these stills and then I would use them as keyframes. So as a way for us to. As a little tip for anyone who's doing fashion stuff, right? When you have like a shot like this, you know, up top and you want to scroll down to the bottom here. If you were just to say like, you know, go down to this guy's shoes, right? Like, you know, camera travels down to his shoes. Problem is like he's not gonna translate the logo on the pants or the shoes. Cause it doesn't have any sort of context of that. So you'd want to go from like keyframes, which is start here, end here. Tools like Cling. Luma Runway I believe has them as well. Like you want to go here, this image to here so that you get the shirt, you get the branding on the shirt, you get the branding on the shorts and you get the branding on the shoes. Right. So that's sort of like the idea.

Host 2 (34:35)

Okay, so just to quickly say something because I think you've taught me something here as well. So you actually create an imagery first and then the video tools to actually animate those images.

Rory Flynn (34:44)

Oh yeah.

Host 2 (34:44)

Is that what you're doing? Whereas I'm actually just trying to create everything in the video tools.

Rory Flynn (34:48)

Should I create images first for control purposes only? If you like really need precision. Okay. Like I think text to video is more capable than image to video in that case. But when you're looking at precision and you need to have like all these fine details that are exact across all of the, you know, the images. Again, on the brand side, you work with a lot of these companies. It's, you know, it's very nitpicked, of course, especially cause it's AI. So if I can control the images, I can control the video.

Host 2 (35:15)

Got it.

Rory Flynn (35:15)

That's sort of how I look at everything. But I do love text to video as an option for just about everything.

Host 1 (35:20)

Just as a quick aside. Cause I want you to show us like how you get the files and everything else that you were about to go into. You can do this for literally anything. What blows my mind, we have a lot of startup founders and everything watching this. Like if I was into like design and ux, I would just upload my product design language, any competitor's design language, all of those things that you could literally have a workflow where you could put a feature idea in and see it in every possible design style and user experience. Exceptionally quickly.

Host 2 (35:50)

Wireframe it in minutes, very quick.

Host 1 (35:52)

It's insane.

Rory Flynn (35:53)

And like, you know, with tools like Nano now, Nano Banana, it's like a, it's like, I don't know, a 15 second gen time. Like it takes nothing so crazy. It's super quick. But, like, stuff blows my mind because it's also endless. Like, this is where I have to. This is where I have to focus myself because I could just keep going here I was like, why not have 75 other replications of this and, you know, do one click, 800 images? You know, it's like, all right, like, reel yourself in, Rory. Because this is sort of where, you know, it gets refined. Right. If I look into something more like on the product image shot side of things, where I'm using a product to get to an end result here.

Host 1 (36:30)

These are ad creative, it looks like.

Rory Flynn (36:31)

Right? Yeah.

Host 1 (36:32)

This is like a big use case, I think, for most companies, regardless of what your company is.

Rory Flynn (36:36)

Exactly. So, again, you'll notice a lot of the same sort of theory here. The batch processing engine, we're not doing that much different. You know, we're basically attaching an image, which is you input an image, we're attaching a user prompt, which is just saying kind of what you want. We have custom instructions here, which is basically like your system prompt for any of your prompt generators on your custom GPTs. But really the idea here is to focus on, like, I want to be able to produce again, 10 prompts.

Host 1 (37:06)

Yes.

Rory Flynn (37:07)

And I want them to be very specifically formatted and I want them to go and just go directly to, you know, the image generators without stopping. So this can all be one click and then the other piece here that we've started to add a lot more, which is the Visual brand profile. Right. So taking about 20 images and dumping them into ChatGPT or, you know, any of the LLM of choice, and asking to extract like a visual pre profile. So we have like, like a style guide for each image generator. Right. So we're using the custom instructions, which are the basic rules saying, like, this is how you create, this is how to format, this is how to output. We have our visual brand profile, which is saying, this is how it should all look. You're combining that as a system prompt. This sounds so much more confusing than it is. I promise to anyone who's listening that meets with your prompt and the picture of your image. Then it goes and splits. And then we can run these to any sort of different variation of this. There's, you know, I've run millions of them here, so it's like anything here can work. And we're doing it again. We've done it one for 10. We just split it and do 10 more. Thing is, this can be 40, 100 scales, infinite. You Know, a thousand. Because when I open here, it's like, you see, like, everything that's getting created in this. This stuff takes me, like, 20 minutes. I ran a lot of.

Host 1 (38:25)

These are literally making thousands of ads in 20 minutes.

Rory Flynn (38:28)

Yeah, right? And it's curating. It's curating versus creating. At this point, the curation portion is going into the system prompt and the. The actual building of the tool. And then, like, that's the creative piece to me at this point. This is. Now, how do I run this at scale and just, like, get to a place where I'm looking through all of this and being like, oh, I like this one, or I like this one, or I like this, this one.

Host 1 (38:51)

Have you built an agent yet that ranks and, like, basically gives you the best revs?

Rory Flynn (38:56)

That's the next piece. We keep creating newer problems, right? It's like.

Host 1 (39:00)

Exactly. That's why I was asking scale.

Rory Flynn (39:02)

You know, we need a visual analyzer.

Host 2 (39:04)

We're less than an hour in, and you've showed us a use case that basically has built a entire, you know, photography business and a whole creative design business through, like, a single tool and a bunch of prompts just so people understand what's happening here. Like, this is a creative design agency.

Rory Flynn (39:19)

Yeah.

Host 2 (39:20)

You would have to pay a frigging fortune to get a creative design agency to come back with this amount of. Usually you get, like, you pay in thousands and thousands. You get, like, three choices. Yeah. You want to pick A, B, or C. And you're like.

Host 1 (39:33)

And the outputs.

Host 2 (39:33)

And you're like, hey, do you want to pick any of these thousand images?

Rory Flynn (39:38)

Yeah, it's just creating custom libraries. Right. Like, and this is sort of like, even the crazy thing about these tools. Right? This is. Some of this stuff blows my mind where even as the user prompt, I'm saying dumb things. Like, this is where I like to have fun. Where I'm saying dumb things like, give me a. Give me a lo fi meme. And like, it's coming up with something like this. Like, all hell, the candy. Like, this is so stupid but awesome at the same time.

Host 2 (40:02)

So good.

Host 1 (40:02)

That ad will work. That.

Rory Flynn (40:04)

This was genius. This was lo fi meme. Right? It's like, busy office at. I didn't do anything here like that. And I hate to say that, but it's just I put so much work into that system prompt that now when I use a user prompt, it's just like, whoa, okay. Because there's some smart stuff, like these memes that I was asking it to make. Like, this is Smart. Just asking it by saying to make me lo fi meme, you know, athletic greens. This is my personality now. No prompting of that. Just came up with it. Self care mode with a bottle, you know, putting. It's so absurd. That's so good, right? Like it's really good.

Host 2 (40:38)

I want to get in here and build paid advertising energy for direct response because these are all like really great brand ads and for B2C. But in B2B, like what's really weird about B2B is the worse quality the image, the better it performs. I would love to just like take some of the.

Rory Flynn (40:53)

Would you like to see this? So this is.

Host 2 (40:56)

Oh, here we go. This is a B2B advertising.

Rory Flynn (40:59)

My campaigns are so good. Your spam folder wants an autograph, right? Like these are. These are we. We prompted in here. Like give me like a used car salesman vibe. This is my old business partner Phil. Still, you know, play around a lot together with this stuff. So it's. How do we take. We had a winning ad for B2B and the email marketing services. That was basically, we called him the King of Klaviyo. We put him on the Game of Thrones throne and that was it. And that sold us clients for months. It was insane actually how many people responded to that. So we're like, let's just replicate and do the dumbest stuff we could possibly imagine and see what works again. So this stuff is pretty funny in regards to respect the CTR King. These are just taking that one image of him and running it through saying give me a meme. So, you know, to the B2B side you can also do really weird stuff that sometimes catches the eye.

Host 2 (41:49)

I love that. By the way. Kip and I are like whatsapping frantically but tell all the I want to do with this. Part of the big takeaway is, you know, in companies you have these creative teams who own all the kind of creative asset generation. But this to me means that actually teams who are relying on imagery should be self sufficient, 100% right? Like there's no reason that you kind of have to have different teams who are skilled. You should have people who can just use these tools wherever they use images. They should just have a seat, a license and figure out how to use these tools. And I hadn't really thought about that before until I've seen how much you are able to do now. You have to master foundational elements, but how much you're able to do when you actually really understand how to do prompting and workflows. It's pretty, pretty mind blowing.

Host 1 (42:29)

We should do a follow up episode to this. Kieran. I don't know. I don't know if it's a priority or if it's who, but it's like, oh, in 2024, the average paid marketer or paid marketing team spent their time this way and now they need to spend their time in this very different way. Like how you spend time on asset creation, segmentation. Data reporting has changed dramatically huge. And the number of campaigns essentially we're saying is like at least 10x. Oh, easy, right?

Rory Flynn (42:59)

I mean it can be, it's going to be infinite.

Host 1 (43:02)

I guess that's a question for you, Rory, before we go. It's like you're working with this technology, you're scaling up a bunch of creative. Like what has been the delta in terms of like number of campaign and amount of creative before doing this to like now? How big is that?

Rory Flynn (43:16)

We just created more work for ourselves, to be honest. Like it's. Of course, that's what I'm asking. It's like I think there was this crazy promise of AI, as if it's going to get rid of a lot of the work. It's like, no, we're still going to fill the time, we're just going to do more with the time. Right? So from a, from a standpoint of being able to test, it's been a lot more effective because if you just look at like the way the game is played a little bit more now, it's always been, it always evolves and changes, right? Like what works right now in the performance space won't work in, you know, a year. But it's like if we have more that we can test, that means we can optimize better, which means we can find winners better, which means we can iterate better. Right? Like it all sort of feeds itself. It's just having optionality and having freedom. Whereas if to make that library that I just showed you on Figma graphic design, that would have taken me a year maybe to create all of those. I'm not even kidding. And there's 20 minutes now. So I don't always want to be like, it's the time, it's the time, it's the time. It's like the fact that it's also a little bit intuitive when you put some intuitive backend into it that, you know, again, not everyone can just go and build that thing in five minutes. That took me three weeks to build that tool. But that tool, you know, after three weeks, it's just a Basically, did I get the system prompt right? And like once the system prompt is right, it's like you strike gold. And then it's like you just hit go and it works.

Host 2 (44:35)

Yeah, it's creative design team.

Rory Flynn (44:36)

Yeah.

Host 2 (44:37)

Took you three weeks to build a creative design team.

Rory Flynn (44:39)

Yeah.

Host 2 (44:39)

I take that.

Host 1 (44:42)

Anybody in the world would take that.

Rory Flynn (44:44)

Three weeks of trial and error and working backwards. That's one thing I can, you know, definitely impart onto people here if they're interested in building workflows like this. The idea doesn't start unless you go from the end. So you have to be able to get good output first. You have to be able to get. Oh, this is how the output should look.

Host 2 (45:02)

That's agreed. Yeah.

Rory Flynn (45:03)

This is the prompt I used. How do I get that prompt to scale 50 times? Right. What do I need to write a system prompt to get that prompt 50 times? What do I need to input? So it's like total reverse of how you probably think about it.

Host 2 (45:16)

That's a great tip. So start with the outcome and then reverse engineer into scale.

Rory Flynn (45:20)

It's the only way to do it as a pro.

Host 1 (45:22)

Tip. And so you're using weave and then you're basically exporting these images and then running ad tests and doing everything in meta or any of the core platforms.

Rory Flynn (45:32)

Right.

Host 1 (45:32)

Just so that everybody understands kind of the end and flow as we close out here is what I'm trying to make sure.

Rory Flynn (45:36)

Yeah. So personally, I'm not running any ads. I'm teaching businesses how to create this stuff so they can do whatever the hell they want with it. Right. That's my. It's like, take it. Here's even like, you know, here's like the stock car that I just gave you. Go build upon it too. Like change it up, make it different builds, you know, put the performance tuners on it if you want to. So it's basically like, here's the. Here's like get off the start line and move. But yes, that is the idea is to just sort of eliminate a lot of that work. And I hate to say eliminate. Cause I still think the photography is the most important thing to anybody. Like, I wouldn't say, go get rid of your photographers and your creative team. I would say this is a supplement.

Host 2 (46:12)

Yeah.

Rory Flynn (46:12)

This is a way to test direction and, and then get things done professionally. If you, you know, if you have the means to, you know, to keep building out or to be like, this is our entire look and feel now. Like you can reshape a brand. Yeah. By testing and, you know, like Audience response. So there's just so many different ways to go about and use it. Just I try not to tell people to be like, thinking about as I need to get rid of people because I have this. It's like, yeah, yeah, you know.

Host 1 (46:36)

No, it's not about that.

Host 2 (46:37)

Yeah, well, it's kind of like code. Right. The vibe coding means it's easier to create more code than ever, but it just means there's an unlimited want for more code. And so it just allows people to create more software. I'm sure there's an unlimited want in the world for great creative assets, but historically, the only people who could get them is people who could afford to do that through creative design agencies or photographers. And now it kind of democratizes the ability for everyone to get that. So the people who might use it may never have used a photographer or a creative design studio anyway because they couldn't afford it.

Rory Flynn (47:10)

Sure.

Host 2 (47:10)

And I think that's what's happening is like, there's pent up demand for this stuff and now everyone can kind of get access to it and kind of we can end where we started, which is create. Right. I think this is an incredible time to be a craftsperson.

Rory Flynn (47:22)

Absolutely. It's just like, do it. You'll figure something out that you like or some useful way to use it.

Host 1 (47:29)

Go build. There's a bunch of tools. Rory, you've taught us that an image can be a core building block for lots of things. Text, video, different types of imagery, ads, and that's like a core takeaway here. But you have to have proficiency in reverse engineering things and have a tool or a set of tools that you can then play around with and build and test assets on. And you've walked us through and given us some baseline workflows and examples, and I think you're going to share those so everybody can go and get started right away, which is awesome.

Rory Flynn (47:57)

100%.

Host 1 (47:58)

Rory, this has been awesome. Thank you so much for being on the show today. We'll talk to you all real soon.

Rory Flynn (48:02)

Awesome. Thanks, guys.