Ep. 79: AI Workflows Brands Actually Pay Five Figures For

0:00:00Cold opener

Play from 0:00:00 8:88:88 0:00:00 Drew Brucker (00:00) What's up, everybody? Fast Hours Podcast, episode 79. Seven Rory Flynn (00:06) Seventy nine. Drew Brucker (00:07) 79. I don't know if that's a lucky number, but it might as well be. We're almost at 80. And it feels like we've like, because we've been on this little run, you know, between Rory Flynn (00:18) It's a run. Drew Brucker (00:20) episodes 70 through 80 has basically gone at light speed. So I'm pumped about that. Dude, we've been knocking these out. I know we got a good episode today.

Play from 0:00:30 8:88:88 0:00:30 But before we get into that, man, what's happening in your world? Rory Flynn (00:33) Well, if we're gonna be candid about the situation, if you guys can't see the bags under my eyes, wife gave birth to yes, Drew Brucker (00:40) Whew, they're rough. They're rough. Rory Flynn (00:42) they are. It's I'm pale, I'm I'm I'm puffy, I'm tired. I'm gonna s I'm gonna sound like a total idiot on this episode, but here we are. Wife gave birth to our son, Roman. He's about five weeks early.

Play from 0:00:56 8:88:88 0:00:56 He's cool, he's chilling, he's you know, burping, farting, sleeping. Eating, that's about it, making some noise. And I haven't slept in two weeks. So, you know, perfect time to do episodes, right? I mean, like you guys will get the Drew Brucker (01:10) Daddy times two. Rory Flynn (01:11) times two. It is way more chaotic than I expected. It's like, you know, when one starts, the other one gets going too, and it's just like, all right, this used to be peaceful here. Now it is now it is a tornado. But hey.

Play from 0:01:25 8:88:88 0:01:25 Drew Brucker (01:25) Was it ever was it ever peaceful though, let's be honest. Now now it will seem peaceful because now you've got two and they came at different times. You've got sort of the baseline there, Rory Flynn (01:36) I was getting I was getting soft. You know, I was getting like s eight hours, seven hours of sleep a night, like just just cruising, feeling good. And now it's like, you know, I'm eating leftover pizza at three in the morning because I'm just trying to stay awake and my body's just failing me. Dude, it's a lot harder at thirty seven than it was at thirty four. Like, what?

Play from 0:01:55 8:88:88 0:01:55 Drew Brucker (01:54) Dude, that's I was just gonna say that. You just told me that. Yeah, you you texted me. You're like, dude, the difference between thirty four and thirty-seven is drastic. Rory Flynn (02:03) it's it's like I feel like a corpse of I wake up in the morning like no joke the other day I took the I took Luna out I take her out every morning at 5 45 like so you know there's no there's no negotiating this one back here she she pounces on me at five forty five in the morning like get up dude time to go so like even when I do get two three hours of sleep she's right there to wake me up. So regardless, I take her out in the morning, we walk to the park, I throw the ball for like 45 minutes, we go get coffee, I walk home. Right.

Play from 0:02:32 8:88:88 0:02:32 When I got home, no joke, I could not remember the walk in the morning. I was like, am I like blackout drunk? No, I'm just sleep deprived. This is chaos. I don't even know what I'm saying, you know, a little bit sharp with everyone too. Like people like trying to be nice to you. And I'm just like, yeah, whatever. You know, such a I'm such a you know, Drew Brucker (02:50) Don't mess with me, man. Don't mess with me. Rory Flynn (02:52) don't mess with me. Don't don't pull don't pull some Higgsfield stuff on me right now because I'm I'm ready to Drew Brucker (03:00) Dude, they got lit up again, bro. I I just I just can't. I mean they got lit up again. Did you see that?

Play from 0:03:08 8:88:88 0:03:08 Rory Flynn (03:02) Catch astray. Yeah, they're like they're you know, maybe we should you know what? Let's just we Drew Brucker (03:10) That okay. So we we we Rory Flynn (03:11) give them enough airtime. Right, you know, we don't need to talk, but they suck. Sorry. Drew Brucker (03:14) I agree. I agree. The the the sh the short is Higgsfield gonna Higgsfield. So Rory Flynn (03:19) Yes. But other than that, you know, I've I've been doom scrolling in the middle of the night, trying to watch like, you know, YouTube videos, whatever to stay awake. And you'll you'll appreciate this one. I know I don't know if we've mentioned this one on the podcast. Maybe we have. I saw some I saw a video of a dude being like, All right, you guys have been sending me movies that can make me cry, part seven. And then he puts it up and he's like, homeward bound. I was like, No. I was like, You are Drew Brucker (03:43) yeah.

Play from 0:03:45 8:88:88 0:03:45 Rory Flynn (03:44) going to melt down on camera. Don't do it. You know, and then he You see the guy like halfway through bawling. And I was like, he's not even to the hard part yet. Like you know, he's not even to the end. It's like when Drew Brucker (03:55) Yeah, yeah. Rory Flynn (03:57) when shadows and they're like that muddy pit or whatever and they can't get out kind of stuff. And then you see him coming over the hill at the end, and he's just like it's like the ugliest cry you've ever seen. I don't know if I could watch that again. Like I don't know if I could actually sit through it. I might I might internally combust if I watched it, but Drew Brucker (04:14) I you know what? There was a there you we're we're doing we're like on this movie run with the kids right now. You know, like I I love Rory Flynn (04:20) Yeah.

Play from 0:04:21 8:88:88 0:04:21 Drew Brucker (04:21) like family movie time. You know, it's just like it's you know, when they're under five, there's only so many movies that that you can really choose from. I mean, I I've told you this before, I get a little excited about someone where I'm like, I just gotta show you this, even though you're probably not gonna understand 90% of it. And so what I end up doing is like pausing the movie every five minutes. It's like, do you guys understand what's going on? Just like. Rory Flynn (04:42) Okay. Drew Brucker (04:43) And like, you know, and then I explain it and they're like, okay. And then I do that like throughout the whole movie. But we watched a movie the other day. and we're doing like a lot of these, you know, like little cartoon movies, right? The Magician's Elephant. I'd never heard of this.

Play from 0:05:00 8:88:88 0:05:00 Rory Flynn (04:55) What is this one? I don't know this one. Okay. Drew Brucker (05:00) 2023, 2023 movie. But this one at the end is a little tearjerker. I mean And the whole family was just like wiping. You know, we're just Rory Flynn (05:14) Yeah. Drew Brucker (05:14) like, and then like trying to tell the kids like, it's okay to feel, you know, like to feel sad during a movie, like just a movie, right? It's good, but it's good to feel the feelings, blah, blah, blah, blah, blah. But this was actually a dope movie. This is a good movie. Highly recommend Rory Flynn (05:27) Okay.

Play from 0:05:28 8:88:88 0:05:28 Drew Brucker (05:27) if you've got kids. Highly recommend. Rory Flynn (05:29) I'll keep this one. Drew Brucker (05:30) The the only thing that my my gripe though with this in a lot of these movies that are made for kids. Please tell me why it's always necessary to mention killing. You know, like K-I-L-L. Rory Flynn (05:44) yeah. Drew Brucker (05:45) Like throughout. You know, it's like, dude, how do I I I that kids don't need this? You know what I mean? Like there there Rory Flynn (05:50) Terrible. Drew Brucker (05:51) could be something simpler to explain here. But anyway, I'm I'm off. Yeah, I'm off my high horse. I'm just saying Rory Flynn (05:54) Let's not go back to the nineties then.

Play from 0:05:59 8:88:88 0:05:59 Drew Brucker (05:59) that like that was a that was a really good one in in a tearjerker too. And it was like, I think that was like the first movie where we were all like kind of watching, where it's like, my God. my God. You know? Rory Flynn (06:09) I mean, if we go back to the trauma of the nineties, it's like baked in. It's like every movie had to start with something absolutely terrible. Like, you know, Bambi Land Before Time. Like it you can't you Drew Brucker (06:19) Light do Bambi. Rory Flynn (06:21) you can't make it out of the you couldn't make it out of the first ten minutes without like I you know s some some of the the closest human like the closest relationship you could have just being severed. And then it's like, you know, the rest of the movie is is all like well now it's happy and fun, but yeah.

Play from 0:06:37 8:88:88 0:06:37 Drew Brucker (06:35) And it's like an a you know, an animal too, because it's like, dude, innocent. Innocent like just just don't. You know what I mean? Like Lion King, Rory Flynn (06:42) We're bad. Drew Brucker (06:44) Air Bud when he throws the ball, leaves him on the island. Like, come on, man. Come on. So Rory Flynn (06:48) There's so many bad ones. But it's the nineties, I don't know, I guess we were just you know, we were we were traumatized differently back then. It was w just just setting us up for lifelong trauma that we could that we could just process differently because we've been exposed to it at age five.

Play from 0:07:03 8:88:88 0:07:03 Drew Brucker (06:55) We were. We were. And and now our cathartic experience is just sharing it with everybody else. We're just projecting. Rory Flynn (07:06) Yeah. Dad just sneezed, it's okay. He's not really you know, that's not w the water works running down his face. Don't worry, he's not say happy tears. Happy tears. Drew Brucker (07:15) Happy tears. Yes. Happy tears, man. well, I think a little segue here. I think I got some little Rory Flynn (07:21) Yeah. Drew Brucker (07:22) happy tears with this one because this is we're gonna do this episode diving into nodes and getting in further with Weavy workflow specifically, Figma Weave now. this is something that people have been requesting for a while. And we had sort of like several guests in a row, we couldn't get it.

0:07:22Why a full Figma Weave episode

Play from 0:07:40 8:88:88 0:07:40 You know, right away. But we did have that episode a while back, maybe a month or month, month and a half back now, where we showed a little bit of something there. We were kind of going through it and and people spoke up and said, We you guys need to do a whole episode on this. So we're like, say less. So I think Rory Flynn (07:54) Done. Drew Brucker (07:55) today we're gonna get deep into that. Like we're gonna go super deep into that. shout out to Figma Weave as well. Cause I think it's you know, it's just an awesome tool, but I think the people behind it.

Play from 0:08:07 8:88:88 0:08:07 too we've got relationshi special relationships with both of them because I think they are a perfect example of people behind a brand that are extremely relatable and care about the people promoting their stuff. And that's not true for a lot of these other companies. It feels much more transactional. But shout out to their team. I mean they're they're amazing folks over there. Rory Flynn (08:28) Yeah, and that's why we figured we'd partner with them on this video and get a little bit more in the weeds. And speaking of speaking of being just like a human in the AI world, I woke up to to this one day. It was a handwritten note from the Weave team. They sent it to me and a bunch of onesies when they heard I had my son or my wife had my son. I did nothing in the in the of the sort. when Yes, I d you know, you Drew Brucker (08:51) Thanks for qualifying that, bro. You were about to

Play from 0:08:55 8:88:88 0:08:55 Rory Flynn (08:54) You never feel more useless than a man in a delivery room. You're just there. and it's like don't move, don't speak, don't breathe, don't sneeze. We'll we'll handle all of this. And you're just like, What do I do with Drew Brucker (09:04) Yeah. Dude, that's so true, dude. Rory Flynn (09:07) what do I do with my hands? You know, like one of those things. But they sent me a handwritten note. They sent me onesies. I mean, like, that's a look. I consider them family at this point. They are they are really genuine people. They've been building this the right way. It's always been community focused, not glazing.

Play from 0:09:24 8:88:88 0:09:24 This is reality. I'm sure anyone else who's interacted with them or their founding team or anyone that works there, you know, specifically Neta, Ron, Itai, you know, Lior, Jonathan, everyone, they're great. They're great, they're great people. They just they really want to make a good product. Yeah, Roy too. Drew Brucker (09:37) Roy, shout out Roy too. Yeah. Yeah. Rory Flynn (09:40) Just want to make a great product and they want people to like it. And I think that's why, you know, it feels less transactional most of the times when you're dealing with them because they just, you know, they want to build a good product. They got acquired in a year. Obviously did something right.

0:09:53How node workflows went mainstream in a year

Play from 0:09:53 8:88:88 0:09:53 Right, so yeah. Drew Brucker (09:53) Dude, that was insane, right? Cause like last year at this time, last year at this time, Rory, you were one of the few people that I think were talking about node-based systems across social platforms. Like there, there were I I I think I could count on one hand. Like I I you were definitely one of them. You converted me onto it as well. Like once I kind of like stepped back into.

Play from 0:10:21 8:88:88 0:10:21 doing my own thing again and and really trying to scale. Like I remember you spent like a whole session with me, just kind of like getting me up to speed on that stuff. But a year ago at this time, nobody was talking about this stuff. A year has gone by, and this is such a core part of a lot of individuals' process, teams process. You and I have both worked with clients that want sort of these build-outs, these workflows. And we're gonna like we're gonna get into all of it today. We're gonna get into the tool itself.

Play from 0:10:51 8:88:88 0:10:51 Why you know teams or individuals need to be using a tool like this, right? The capabilities that exist, like the the actual benefits are there. We're gonna get into we might even do a live build here today. We're gonna get into some of our workflows today and share those with you guys. So I think like we're gonna try to cover all the bases here on this episode. So I'm I'm pumped for this Rory Flynn (11:09) Yeah. Drew Brucker (11:09) one. Rory Flynn (11:10) Yeah, I think it's like, look, this is this is what gets used in production. There's a lot of other tools that we use. I mean, I use Claude every day. I use some of the other tools, right? But like this is the one that when I'm building something for production, systems for teams, things that people can reuse. This is it has the most sort of flexibility and customization outside of maybe ComfyUI, but I think that's where it fit a need it fit a hole, right? Like ComfyUI is extreme.

Play from 0:11:39 8:88:88 0:11:39 Right. Like it's it's extreme. That's where I was like, Yes, we can build these workflows, but it's like can anyone operate them once they're built? Does anyone understand what's going on? Can you tell a traditional, you know, designer or marketing team what is going on in this ComfyUI workflow? Or do you need an engineer to understand what's going on in this workflow? Right. So what they did was they they stripped it back. They made it very easy and visual and it's it's digestible. there's like a there's a there's a time and place for every sort of tool and sort of infrastructure that you're building. But this one, when you need something to get scaled across more than yourself, you need something that's easy. And so not not gonna say easy, something that is like digestible. and that's where they fit and that's why it works so well. a lot of different people

Play from 0:12:26 8:88:88 0:12:26 resonate with node based workflows, you know, the people in post production, people in 3D, VFX, you know, designers, things of that nature. So it's it's familiar. But it's also, you know, in a visual space that makes it look somewhat appealing as well. So, you know, I think the the relationship with Figma too is gonna be really interesting. We're just starting to see sort of the whole thing come together now and how it's gonna work and how it's gonna look. They presented a lot of stuff at Config on like the the future of the releases and how this stuff's gonna be integrated. So it's gonna be, you know, more integrated than I probably thought in the beginning. But I think there's a lot of stuff that we can do with it now.

Play from 0:13:04 8:88:88 0:13:04 And I keep finding new uses for it. And it's it's also like my favorite new little thing now with the release of the MCP to just like give Claude little tools that I build and weave. You know, it doesn't have to be like these giant workflows. It can just be like you know, it can be like a all right, here's a reversioning workflow. Take this asset and scale it into, you know, nine different aspect ratios. The workflow is already built. I just send Claude the picture, it just goes and does it. It's like now it's like a little extension for Claude, right? I don't have to leave Claude.

Play from 0:13:33 8:88:88 0:13:33 just go do it and we even come back, deliver it. Right. So there's a lot of different ways to use it. And I think that's the one thing that should probably hit on a bunch of that is that it's not, you don't have to use it one way or the other. You can use it however the hell you want to. You can use it for systems. You can use it for design. You can use it for doing your own thing. You can use it for testing. You can use it for, you know, just going absolutely crazy and doing node explosion in there. But it's it's sort of, you know, if there's, there's a creative way to use it and then there's a more like engineering systemed way to use it. So I think we'd try to hit both. If we accomplish that, we're doing pretty damn good on this one, I think. Yeah.

0:14:10Why use a node-based AI tool at all?

Play from 0:14:12 8:88:88 0:14:12 Drew Brucker (14:10) Yeah. Yeah. And I mean, let's let's let's touch on this for a second. you brought up, you know, systems, scaling, right? Like, why should somebody be using a node-based tool in the first place? You know, I think like let's start there too Rory Flynn (14:23) You know, I think like what what's Drew Brucker (14:25) and unpack that because at the end of the day, like one of the foundational elements of this is you're getting essentially an aggregator of all of these tools, So instead of all these individual subscriptions that you might have, everything's in one place. That saves you a ton of money. Like I'm just thinking about that from the the personal creator side, let alone, you know, working in-house somewhere where they're you're trying to justify maybe everything under a budget, Like, why do we need five tools, six tools to do this? Well, you got one that can probably do a lot. So

Play from 0:14:59 8:88:88 0:14:59 it's the it's the model aggregation, the subscription costs. You're also I think getting to try a lot of these tools, right, without a a a strong commitment to any particular model or tool too, right? So like the LLM Rory Flynn (15:13) Uh-huh. Drew Brucker (15:14) side, right? You get to test different LLMs. The image side, you get to test Nano Banana versus ChatGPT versus Reve versus something else. Image side or video side, same thing, right? And so you're kind of getting exposure to all these tools at scale, which is very much aligned to this idea that you should not be really Confined to any specific tool, right? It's you need to be a little bit tool agnostic and understand maybe the differences between the tools, what tools do better than others, the

Play from 0:15:45 8:88:88 0:15:45 use cases for those. But this opens up the lane for that. So I think like that alone is a big, big thing to note is just the idea that you can put all these things together. And then, you know, the second part of that is what you mentioned, Rory, like if if you're trying to build a system. In-house, or maybe you are the person that's going to integrate that somewhere in-house, How do you activate people? How do you activate a team? So there's going to be the learning curve of one tool versus five or six tools. And in order to move fast, right, you need to almost create templated use cases. And doing things like this gives you a templated use case where people can build on the same foundation. And so people aren't working in Disconnected or disjointed fashions are working in the same space and they're

Play from 0:16:33 8:88:88 0:16:33 working from the same templates and they're working from the same concepts. Right. And so this pulls all of those pieces in place. And there's a big difference between creating a one-off asset or a couple of hands, you know, five assets versus creating hundreds or even thousands of assets. Rory Flynn (16:49) And I think that's from a internal and a scale process too. The good thing is like not everyone has to learn this stuff. Right? Like you can have someone that's really good at building these things and they can deploy these tools that everyone can just use in a very simple way. Right. Like so the the example with that being, like with the weave workflows that you Create. They don't have to be run individually. You don't have to go and click every button. You know, if they're systemed and engineered correctly, you might just need to put in an image and like either type something or select something and

Play from 0:17:22 8:88:88 0:17:22 push go. And then it runs through the whole process. But again, you might not have to build that. Someone could build that on your team and be like, here, take this and do it. Like this is my I'm gonna keep coming back to the asset reversioning thing because I think this was just always a pain in the ass as a designer. It was like, I have this asset in one by one. Aspect ratio. Now I need it in nine by sixteen, three by four, two by one, four by three, whatever, right? Like and you used to have to go and manually do all that. Now we can just scale it out, make it one click, and it'll go do it to all of them. Now that can also then be a tool that you've created, like harden, and then be like, here, just send that link to someone and they just put their image in and hit go. And that's it. So like they don't have to worry about building it every time. Like not everyone has to be the systems engineer. You can have one person that does it.

Play from 0:18:09 8:88:88 0:18:09 And then everyone else can get like feast off of it. So it's and like that's the that's the other thing too that we're definitely gonna get into is like when do you need like when is it good? And like what are good places for this? Because I think what I see a lot of most of the time, people try to like way over engineer, like get these giant node strings, and it's like, well, one thing breaks in the middle and then the rest of it doesn't work. It's like sometimes it's easier.

Play from 0:18:36 8:88:88 0:18:36 Like with we, like you you know, these tools function this tool function now is to have five tools that Drew Brucker (18:40) Micro. Yeah. Rory Flynn (18:41) yeah. And then just be like, run it to this one, run into this one, run into this one, run to this one. You know, if you're using this whole MCP thing, we're gonna have to get into it a little bit more because there's so much to unpack with like how this can all start to be strung together in like multi step processes. But even just starting with an empty canvas you got here, like this is a this is a good one. Like Drew Brucker (19:02) Yeah, and I think like just just like backing out of this, right? Like we were just talking about all the different capabilities within here. I mean, this is the toolbox. I mean, you can literally do just about anything you want in here. I mean, you've got sort of like these traditional like photo elements, right? Where you can resize and crop and composite, right? And then you've got, you know,

0:19:02Inside the Figma Weave AI toolbox

Play from 0:19:23 8:88:88 0:19:23 sort of like these masks, you've got text tools if you're getting into ads or text in general. You've got iterations. You've got all these other things that are sort of like the connecting dots. You've got the image models, which go on for days. You've got, you know, the edit capability within those. So nano, ChatGPT, right? You've got, I mean, there's so many, there's so many. I'm just like Rory Flynn (19:43) So vaš. Drew Brucker (19:43) very I'm glancing over this, but there's a lot in here, right? In painting, out painting, background removal, aware fill, 3D, Rory Flynn (19:48) Three D audio.

Play from 0:19:53 8:88:88 0:19:53 Drew Brucker (19:51) audio, you know, upscaling images. So you got like Magnific Rory Flynn (19:54) Upscaling, yeah. Drew Brucker (19:56) here, the video L. I mean, so like this is literally like your one stop shop type of deal. And so to your point, like again, like zooming out from this, if you're ever integrating AI and adopting AI inside of a business is already tough as it is, right? Because if you're not hiring outside help, you need somebody internal that is really going to be your advocator and like the implementer and the mover of these things. And so how do you eliminate friction? Well, one of the best ways is if you've got a tool, a tool.

Play from 0:20:27 8:88:88 0:20:27 That then can take advantage of all these other capabilities because you're lessening, you're flattening the learning curve and you're reducing friction, right? There's not all these bottlenecks that you have to get up to like these are the things I'm always thinking about whenever I work with clients too, is just like, how do I activate new employees or people that just maybe they're not savvy with AI, right? It's like, well, I gotta like flatten this curve as much as I can. Things like this do that job very well.

0:20:53How to sell AI workflows to businesses

Play from 0:20:54 8:88:88 0:20:54 Rory Flynn (20:53) This just reminded me from a couple episodes ago when people asked to show like business cases or like the business stuff of like how we're doing this, right? one of the things that I gained traction with when early on, When this was just new, no one even you know, what is this spaghetti string sort of canvas that we got going on here when I send this to people? It was like here. This is like we had a conversation, We talked about what your problems were, what you guys were trying to fix, how you might be interested in AI, what parts there were. I would just build a workflow in Weavy after that call and then like send it to them, be like, Hey, solved it. You know, like n free. Right. Like a lot of times it was something small, but it was just to show that it could be really simple. Like here, just like put your image in,

Play from 0:21:39 8:88:88 0:21:39 push go, watch what happens. and then, you know, most of the time that as a sales piece. That's like okay, what else can we do? You know, like that's a that's sort of like the you have to Yeah. Drew Brucker (21:49) Show instead of tell too. Yeah. You're like you're you're providing the proof, you've done this enough times where it's not a big lift for you, but it's a it's a major thing for them. Rory Flynn (21:58) And selling this stuff in to businesses as well for anyone who's who's you know interested in this and doing it maybe on their own. I've always had more success simpler. Like, don't don't show the crazy you know, 50 group node package that gets you to the end result. Like that is so intimidating and it feels so out of touch for most people that don't understand what this stuff is. When you

Play from 0:22:24 8:88:88 0:22:24 can solve really simple problems, like I said, I'll say it a million times on this. The asset resizing thing, like here it is. You put your image in, you push go, you get 10 new images that are resized to the right specs, and everything fits. And it's like, that's cool. Or if you want to localize text, right? Like if it's like, here's my copy. I want to localize it to five different languages, right? You show someone that can be done in one second, and it's just like, I get this now. This is a you know, basically we're building little tools, not.

Play from 0:22:55 8:88:88 0:22:55 Here's this giant graph of stuff that you can do in all these crazy, you know, branches. It's like keep it simple, show the value, solve simple problems, then the big problems will come after that. You can, you know, have the time to work on that. so so the little sidebar. I know people have asked us about this, but a lot of times in my sales process, I was giving away free workflows early on. Like just free Drew Brucker (23:14) Yeah. Yeah. Yeah. Rory Flynn (23:15) ones. Just like try it out. Let me know if it works. And then they come back. Like, that was that was great. How do we how do we get more of this? Slide right in. So Drew Brucker (23:24) brands pay some money for this too, guys. I mean, you know, this Rory Flynn (23:26) Yeah.

0:23:24Why brands pay five figures for AI workflows

Play from 0:23:27 8:88:88 0:23:27 Drew Brucker (23:26) is this is five figure stuff. So I mean, you know, the the this matters. This speaks the language, I think you brought a really good point up, which I don't think a lot of people, even that are playing with this, are thinking about, which is micro. Yeah, like those big things look impressive, but when you present that, after they're like, whoa. Their second thought is like, shit, there's no like what how the hell am I gonna do this?

Play from 0:23:53 8:88:88 0:23:53 Right? Like, there's no way. So, just the practicality of handing off in in almost like categorizing use cases is a better way to organize sort of the this. Change management as a as as a as it stands right now, right? And because there are going to be obvious use cases that you know of offhand, and then there are gonna be some that you discover later down the road. But again, if you were to trip up anywhere along the road of like this big wired piece, it kind of sucks having to like zoom in and look it through all that stuff, right? And so, you know, to make it very simple, is the is the shortcut and is the hack here.

Play from 0:24:33 8:88:88 0:24:33 Rory Flynn (24:33) this is actually what you just said was good because I want to show you how, like, and then you know, dovetailing on what I just said, starting small, and then how this scales. Like, I'll show some examples from real quick, I'll just show like what that looked like at at SharkNinja, when we first went in there over a year ago, All right. You tell me if you can see this. Can we see this? Drew Brucker (24:54) We got it. Rory Flynn (24:55) Cool. Right. So Shark, I'm pulling up slides because this is actually in my Figma presentation, but it makes sense, right?

0:24:33How SharkNinja localizes across 35 markets

Play from 0:25:02 8:88:88 0:25:02 Like when you hear the context of why it's needed, then you're like, okay, I start to understand. Like we've created the situation ourselves. Right. Like whenever I say the whenever like, why do you have to use AI? I'm like, because we've created the environment where it's necessary. It's you know, someone like Shark, they're huge, right? Like they're absolutely massive. They have hundreds of products. They have 25 categories, 35 global markets, right? So think of just think about that from a localization standpoint. They also launch 25 new products a year and they do 300 unique videos every year of pieces of content.

Play from 0:25:37 8:88:88 0:25:37 That's insane. Just like extrapolate all of those out, like times each one of them by each other. And that's how many assets you have to produce. Right. So again, getting that traditionally done, there's a way to do it, but there's also like a way that you can you can use AI in this, right? So that's that's one of those things, like localization. That was what they came to be originally for. And I'm I'm showing slides, guys, because it's just easy to to to like understand the story. But like this stuff what works in the US, right? If I show a picture of a product in a US suburban home, it looks radically different than what it looks like in Paris or Germany. Like it's a different design design aesthetic, it's different preference, right? Like people don't have giant suburban homes in in Paris. I'm just I'm using this generally. I understand everything's

Play from 0:26:26 8:88:88 0:26:26 different. Don't come after me. But like that doesn't that doesn't make sense for a flat in you know the middle of Paris that's like 200 square feet. Right. Like it's a it's a different environment. Right. So what we started with was like, all right, here's the problem. How do we sell this? Like how do we get this into the team? Because it was everyone's a little bit apprehensive. How do we use this stuff? How do we get in there? Right. So Retouching was the first thing. We had these these older photo shoots that were not getting utilized. They were stuff that had been scrapped. There was things that were wrong. They spent money on it. It was just basically sitting in the vault, not getting utilized. So, all so let's show an example of that. Like, how do we take, you know, these products here that we saw that were, you know, some of these shots that they didn't love?

Play from 0:27:08 8:88:88 0:27:08 There was like, you they were like, we hate this fence cutting through this guy's head, the lighting's bad. It feels like it's very like hollow in here and not real. Drew Brucker (27:13) Yeah. Rory Flynn (27:15) And it was like, all let's just change you the first thing I did was like, all right, let's change this hero angle, right? Let's just drop it down thirty degrees, come and push up closer, make the lens the, you know, the lens wider and we'll remove the fence of the back. Bang. And I was like, that's we can do that. It's like, yes. I was like, we can change the lighting. And again, we're talking about this is before Nano Banana came out. We were like using LMArena on some of these to be like or LMArena to be like, well maybe we'll get that thing called Nano Banana before we It's good with Google because it works really well. Right. So this sort of like fed the rest of it. Then it became the localization. Like, all right, how do we take all the different e comm

Play from 0:27:55 8:88:88 0:27:55 assets? You go to a product page on e-com, right? Typically there's like eight to ten different product shots. They're all tell a different story. It comes from marketing. Each one of them has a brief. They're all part of different, you know, descriptive sort of processes. Like, we need to show this one because this one shows us you know, that it can be packed in a bag. It's portable, right? Like they'll they'll give you that sort of like as a brief. then also like we want to show that it can cook from frozen to real. Like it doesn't have to be defrosted, right? So you have like all these different ways to show these images. But you know, fried chicken in the US might not fly from you know from the US might not fly in Germany. They might not be at, you know, it might be more like Schnitzel there, right? Or it might be more something else. So again, like that we're we're being able to take all these different pieces that we might have the same brief for and then localize them to whatever it is. That's where it started to expand to, right?

Play from 0:28:44 8:88:88 0:28:44 Then we started being like, can we use old compositions that we like, turn them into sketches, you know, rebuild new products into it, you know, environment fixes, like we got we we were getting things back from, you know, legal and the product team. this thing doesn't work good on bumpy rugs. Can we flatten the rugs? Right. Like simple things like that. Or like, can we just composite and change the color of the product and product swap? Like these are these little places that you wouldn't think all the time. you know, we have things like this. It was like, we need to create these thick cut French fries.

Play from 0:29:14 8:88:88 0:29:14 Great. Let's put them in the dish. But also, like, how do we make sure that the the fries can sit in this type of dish all the time? So we can basically remove the dish and then we take these fries and utilize them as other assets. So we keep the references so we can build asset libraries. Right. And then there's the, you know, then there's more full campaign style aesthetic where we're doing more things that you guys have seen before in the past. But like this is just a good way to show how it starts with one thing and then there's all these little areas, right? They can go into video.

Play from 0:29:41 8:88:88 0:29:41 Where it's, you know, localization here, instead of it being the US, it can be somewhere maybe in the UK. Simple things. We're swapping in one frame, and that's really it. Right. You know, things here where we're taking older assets, swapping products. the hybrid stuff is really cool. Make things seasonal. it's a lot of different uses for it. And it's not like we're building giant workflows there. It's really little pieces that help in the day-to-day for all the things that need to get done, right?

0:30:07Why recurring tasks belong in workflows

Play from 0:30:08 8:88:88 0:30:08 Drew Brucker (30:07) And they're recurring, right? Right. I mean, like these are things Rory Flynn (30:09) Yes. Drew Brucker (30:10) that they're going to need again. So, like if you're thinking of a one-off ad hoc asset, maybe that's not a good use case. But anything that you're doing twice, three times more than that, Rory Flynn (30:21) Mm-hmm. Drew Brucker (30:22) and you've got things that are interchangeable, this is where this wins, right? Like this is where the systems are needed. This is where you can really win because you're solving a problem, you're front loading it, and then From there, it just runs. I I I love front loaded work. Like I'm I've always been a big fan of front loaded work. I love just doing something, spending Rory Flynn (30:44) Yeah.

Play from 0:30:45 8:88:88 0:30:45 Drew Brucker (30:44) extra time on it up front, and then knowing that the rest, I'm just gliding downhill like you're on a bike. You know, you're just like, all right, baby, here we go. You know, like everything else is easier. All you might have to make an adjustment, but like that's so helpful for a lot of people. People don't want to do tasks more than once and become real menial and meticulous with these things. Like that's why it's important too. Rory Flynn (31:08) a hundred percent dude.

Play from 0:31:10 8:88:88 0:31:10 'Cause that's what I think, you know, how I mean, how many times do you do this now? Where you're like, I've done this one time and then it comes up again. I'm like, I'm just building a skill for this. I'm not doing this again, ever. So I mean Drew Brucker (31:18) Well, I mean and like too, let's be honest too. Like how much can you actually hold in your memory now with like how much we're doing with AI? If you haven't done a use case and let's just say two or three months and it goes by and then all of a sudden you need it again, you don't have to rethink it. You've already got it. Right. Like Rory Flynn (31:34) So maybe we should build something real quick, just for anyone who's not familiar with the tool.

0:31:34Live build: your first AI workflow

Play from 0:31:39 8:88:88 0:31:39 Drew Brucker (31:38) Yeah, let's let's do that. I think that would be helpful. let's build something live and you guys can kind of follow us along. Rory can talk through too, like, you know, what some of these things do and why he's making the selections that he's making, right? Like just kind of get into the thought process behind it as well while we're doing it, because I think by now we've hammered home the importance and the benefits of doing this, right? But but now sort of like into the to the meat of it.

Play from 0:32:04 8:88:88 0:32:04 how those things connect. And and obviously we're probably talking to a couple different people here on this this episode, right? There's some people that have never been Rory Flynn (32:12) huh. Drew Brucker (32:12) in here. There are some people that have probably been in here like in a commuter fashion, right? Like from time to time. And there are some people that are locals that have that are in here all the time. So we're gonna try to hit all three of these. But I think connecting that dot like just very specifically and precisely will help. Rory Flynn (32:28) Yeah. And I think it's, you know, a lot of the best ways to work in here is like, let me let me just I have this idea. Can I get from point A to point B by using the tools? And then once you do it, it's like working backwards from there. It's like how do I take this out of the process? Like I used a manual prompt here. How do I turn that into an LLM that's doing it? And it does it every time.

Play from 0:32:49 8:88:88 0:32:49 Right. So that how do I how do you take like all the the variables that are like you that you did, all the decisions you made, and then how do we system make that a system? Right. Does it need to branch this way, or was that just me testing something? So again, you have the ability to go in here and use all these all these damn tools, which I I love having them. I love just having things in here that I didn't, you know, that I didn't know I needed. you know, a lot of times like color correction levels, things like that. There's this alpha masking and things that you just somehow you come across them and you're like, all right, I need them. Right. So, you know, even if you're Drew Brucker (33:18) Yeah. Yeah, they're like stuff that would be in Photoshop or Lightroom that you could that are right here.

Play from 0:33:26 8:88:88 0:33:26 Rory Flynn (33:25) Yeah, so like let's just start let's just do like a couple icons or something like that, just so people get an idea. Right. So just drop a prompt in here and then like we'll go through a couple different things that we can do. Let's just do you know, isometric, 3D Apple. in chunky pixel block style isolated against a white background let's just start here let's just try let's try let's do flux maybe Flux let's do Flux 2.

Play from 0:34:05 8:88:88 0:34:05 Drew Brucker (34:04) this is also like one of those things where, hey, you're not sure what image model to use? Doesn't cost me a lot to just run those against each other. Let me let me speed run four Rory Flynn (34:15) Yeah. Drew Brucker (34:15) or five different image models and see what best interprets that particular ask, right? Like there's a difference between photorealism and these other styles. Maybe you got a model that you really like for photorealism. Maybe you got a different one for isometric icons, like like we're talking about here. Like Rory Flynn (34:32) And you know, I'm just using ones Drew Brucker (34:32) But you can test really quick.

Play from 0:34:34 8:88:88 0:34:34 Rory Flynn (34:33) that are fast, right? Like I just want to use stuff that's fast right now, like for for this. But you know, maybe I'm gonna go over here, change this to, you know, just square. Let's see what what comes out, Cause then we can take this and we can go in a million different directions with it. Like maybe you're developing stuff for a website or for a presentation. Like, all right, cool. So here's a know, here's the apple, right? Like we did that, like can we can we edit it? Let's do, you know, Nano Banana 2 here. Just drop this in Nano Banana 2. Let's drop. And this is showing like what we're doing here is basically like stream of consciousness, like work. If you you know, concepting work, things like that. Like I would do this first.

Play from 0:35:18 8:88:88 0:35:18 If you're new to this stuff, you're not really familiar, and then go back and systematize it, right? an example of that. Would be like, all right, now I see how this came out. Now I want to create like a a little system prompt here that I would say, you know, basically into this every time. Like no matter what, I might just do I might make this a variable, right? So that it could be a banana, coconut, whatever, and it would come out exactly like this every time. So it's just then that's a workflow, right? Then you just type in Apple and the rest of it goes. Not I have to write this every time, I have to add this node every time. Right. So that's sort of the thinking.

Play from 0:35:54 8:88:88 0:35:54 We're doing this stuff for scale. So if we say like let's let's make this reflective Drew Brucker (36:01) just like tactically too. I you you can literally just drag and highlight something and copy and paste it to create a new one, right? So if you've got like to your point, if you want multiple, you know, boxes, let's just say you're running different models of instead of flux, you want to Rory Flynn (36:14) Mm-hmm. Drew Brucker (36:14) run four other ones, you can just copy that, copy, paste, copy, paste, okay, change the model, right? It's there. So like just little shortcuts like that. And then, you know, just for our beginners out there, like every Rory Flynn (36:24) Let me just do Drew Brucker (36:25) everything is connecting. So The way that this is laid out is is done really well because you've got sort of these color coordinated nodes that exist, right? So your pink one here is connecting to a prompt level

Play from 0:36:39 8:88:88 0:36:39 and then your green one is connecting to sort of like the image itself. And it and it labels those things. you know, that may seem a little bit overwhelming or you know, you're not sure about it at first if you're a beginner here, but it's very simple. Each model that you choose, whether it's an an image model or video mod or whatever, those those different node elements may change and it and it captions them in terms of what to connect. So it's that is not a hard thing just to call that out.

Play from 0:37:09 8:88:88 0:37:09 Rory Flynn (37:08) Yeah. So you're seeing here we're running this same prompt three times. We get three different options here, right? Like I might actually like these more. I should probably stop this and use this image, right? Okay, like I can, you know, right now this image is connected here. Let's drop this one in right here. Right. So let's use that one instead. You have options, you can do this. you don't have to just stick with with what you've engineered. You can start to you can move it around, right? So Typically Nano Banana is really fast. because we're doing a live, of course. You will this this this will take forever. All right, cool. Right? Reflective Chrome, good enough for now. Right now we have this. So let's just say I wanted to then make this somewhat like an asset. Let's go, let's make this like a let's add another prompt here. Stick this in. Let's do a perfect.

Play from 0:38:01 8:88:88 0:38:01 Three hundred sixty degree Or three hundred sixty degree rotation. And then we'll go to like, let's just use grok. Grok's typically really fast. it's another thing to know, right? If I'm testing, a lot of times I I I'll use grok because you can get you can get token poor in your or credit poor in here real quick if you're not if you're just like, let me run 17 different in instances of Seedance at one time. It's like there goes your month, maybe two. But Drew Brucker (38:29) this is a really good also the by doing things this way, this is a really good testing ground for new tools. Right.

Play from 0:38:37 8:88:88 0:38:37 Rory Flynn (38:36) Yes. Drew Brucker (38:36) Like Weave is really fast with importing new tools as they come live, right? So like, I don't know, Seedance 2.5, as soon as that came out, right? Usually within hours, maybe a a day or two, right? Like they've got it in there. And so you're getting to test. the new tools in real time as well. I know that's always like probably a challenge too for other people is like, where do I go to use this new tool? Like, you know, so again, like this is this kind of keeps things in one place for you and gives you the ability to test those.

0:39:09What the Gen Effect node can do

Play from 0:39:09 8:88:88 0:39:09 Rory Flynn (39:09) So what you know, all right, so let's just see what Grok did for us here. All right, so we got this 3D rotating, you know, pixel apple, right? Great. Now let's just say we wanted to, all right, I want to like stylize this or something like that. This is where like I love this new tool, Gen Effect node. Like Gen Effect can come in. What this is, like you can just you can tell it to do whatever you want. It's gonna create like a effect, like a shader, almost like a shader.

Play from 0:39:38 8:88:88 0:39:38 in you know Figma standpoint. So like I don't know, what do we want to do? Like a maybe like a half tone or something like that. Or maybe let's see. Drew Brucker (39:46) Yeah. Yeah yeah. Yeah yeah. Do a half tone. That'd be dope. Rory Flynn (39:50) Actually hold on. I did one the other day that was this was this should be cool. I'll show us how to do it, but I just want to pull this one. you remember the Game Boy the Game Boy camera from back in the day?

Play from 0:40:05 8:88:88 0:40:05 Drew Brucker (40:04) Game Boy Camera. Rory Flynn (40:06) Nah, alright. This one. This Drew Brucker (40:07) No. Rory Flynn (40:09) one was good. Let me see if I can find it real quick. Game Boy Camera. There we go. Drew Brucker (40:11) No, Game Boy Camera. Rory Flynn (40:14) I'm just gonna copy this one just so we have it. But we we can go and we can go and look at it over here. Let's see how this works. Drew Brucker (40:19) Yeah, that's dope though. Yeah. Rory Flynn (40:21) let's see. Where is my all right? So just take this. Drew Brucker (40:23) I I like I think this is a really cool new feature, right? Because we we always talk Rory Flynn (40:27) Yeah. Drew Brucker (40:27) about like how do you make things feel customized, And you do have some ability within, you know, the models to do that to some degree, right? Like there technically are thousands, millions, I don't know, dare I say billions of styles that exist, right?

Play from 0:40:45 8:88:88 0:40:45 Rory Flynn (40:44) Mm-hmm. Drew Brucker (40:44) But like finishing this off, making it custom, tailoring it to maybe different surfaces or different campaigns. Maybe there are things there that you can really add. And this is this, I think, adds a nice little customization element once you get to the finish line of something, right? Where you can really make it unique or yours. And so you're just doing this with a prompt, Rory Flynn (41:05) And this is why I I love this stuff. Drew Brucker (41:06) right? Like so this is with the Gen Effect stuff, you're basically just describing what that effect is, right? You want grunge style scratches over there. You want like a I Rory Flynn (41:14) Yeah.

Play from 0:41:15 8:88:88 0:41:15 Drew Brucker (41:14) don't know, like a grain, you know, noise, all that stuff. Rory Flynn (41:19) So I'm just pasting in the prompt that I used to create this. This is a long one. It doesn't have to be like this, right? Like I had Claude do this. You can also add an image reference for what you're looking for. But the the whole general idea here is you just hit that and then it'll go and start creating. But the let's see, I go back. I'll just grab the actual note. I feel like we're doing like a like a cooking show here.

Play from 0:41:44 8:88:88 0:41:44 You like we're just like, here, put it in the oven, and then like they reach in the oven two seconds later, it's done. Drew Brucker (41:50) Yo. Hey, one other thing to call out too, right? Like, you've got the LLMs in here. we say this from time to time. Use AI to use AI, right? Have it rewrite prompts for you Like I know when I've used this too, like I love having a Almost like a prompt template. Like so we're talking about like you've got something here with the Apple. But again, like if I want to make variations and just change that object, but I want it to be that same style, right? Like establishing that up front through the LLM and you know, the system prompt becomes very important here. I guess what I'm trying to say is if you're unsure and you're not a prompt expert or engineer in that sense,

Play from 0:42:31 8:88:88 0:42:31 like let AI guide you with the prompt. Have it develop. that look for you, right? You could provide maybe a visual example of that and bring that into something like this directly, or you could take that and take it into an LLM and have it describe what that, you know, effect is, right? So then you know how to prompt for it. Like there are just several different ways to do it. Rory Flynn (42:52) So now that we have it in here, right, like we can see now that it's creating this sort of like Game Boy pixel effect. Let me take the noise off. So we can see it. Right. And then it's like, all right, cool. Now if I want to, it's giving me because I asked for all this stuff. It's giving me like the block size so I can change sort of the pixelation, like the exposure. We can bring it down if we want to. Change the shadows, right?

Play from 0:43:15 8:88:88 0:43:15 Drew Brucker (43:13) Dialed, really dialed. Rory Flynn (43:15) Right. It's like fun to then play with. And then, you know, let's see, fix sensor integrity. Let's turn the maybe. Bring this back a little bit. Let's just say I like it like that. The other thing I love doing now with this stuff is they have the toolbox. Drew Brucker (43:32) you've also got a reference. So assuming you like this and you're gonna do this for a campaign and you're gonna do this for several things, now you've got something that is tangible and templatized. So you don't have to worry about somebody else, you know, creating something that doesn't quite look right. You've got the references now. They're locked in.

Play from 0:43:56 8:88:88 0:43:56 Rory Flynn (43:55) And also love I love this so I can see sort of what it looks like on both Drew Brucker (43:58) that's dope. That's dope. Rory Flynn (44:00) sides. Right. You know, this is this can also be take this asset right here, turn it into, you know, if you really want it to be a 3D asset, you can just go into, you know, some of the 3D tools here that we have. All right, let's just make this, you know, try whatever it would be. Meshy V6 I mean, you know, I'll run this in the background. And we're doing this, we're doing this crazy guys.

0:44:00How to turn an image into a 3D asset

Play from 0:44:21 8:88:88 0:44:21 Like this is just really shorthand, but just showing sort of like the way that you can go. To these things a pretty quick. we'll just see if this all wrong without me having to do anything. Now, if we want to do the same thing, right? Like you might just go and be like, all right, I want to take this image and let's just get rid of it. Let's just say I don't like I don't like this one anymore, and I don't like this one anymore. Great. Now we're starting to like trim it back. So we have our, you know, this is how we create our image over here. This is how we create our, you know, our texture, this is how we create our video, and this is how we have the you know the effect.

Play from 0:44:56 8:88:88 0:44:56 Right. But if you want to go change this and you could go scale this to like, let's go, this is gonna be an orange, this is gonna be a banana in this style, this is gonna be a coconut, whatever, and then apply the same exact things here. I don't know if does Grok video have a seed? I don't know if it has a seed. But again, I would use the same thing. And it could be the same way, and then you have a whole asset set of these spinning, you know, sort of Game Boy camera things, right? and you can change it, and once you do, you know, something to one, it does it to the whole thing. We can add more to it, right?

Play from 0:45:24 8:88:88 0:45:24 and then you can see sort of what the difference is over here. But this is just a perfect example of the the the I'm gonna get in here and free create and just do something random. and then you know you could end up with a 3D asset and spinning Game Boy camera image by five minutes in here. So this is not the extreme version. We're not taking this that far. Again, just wanna make that very clear. This is just how you get started.

Play from 0:45:51 8:88:88 0:45:51 By doing this stuff. if I wanted to go back and systematize this so that this would happen for any object or any sort of thing that I put in here, I would write this prompt more extreme, right? And then it would be, you know, I'd add add more into the prompt here, like, you know, the prompt rule. So I'm just putting this here as an example. And then like I would go here and let's just go to prompt, right? And then this could be, let's just say this is whatever you want to insert here.

Play from 0:46:19 8:88:88 0:46:19 car. And then maybe I would put that right right there. So let's just make this the variable. Drop it in. Put this right here. And now this is depending on whatever it is, whatever you change it. So isometric airplane. Now that's injected, right? So that can be changed. Right? That whatever you need that to be or then like the prompt rules if you need them to change every single time. Or this is maybe like your style The style of the prompt that you're trying to be in here, like, you know, I'm adding my prompt style here. Right. And this could go right there. So whatever you're doing, like this is going to do the same thing now. You all you have to do is change this. Once you get the system prompt down and you know these variables are, then, you know, and we'll get into the system prompting stuff,

Play from 0:47:08 8:88:88 0:47:08 which is where LLMs gonna come in handy. Then it's, you know, you can basically have a workflow, And this is this is something, you know, now we have a we also have a 3D asset. Over here, if we wanted to go play with this. This is something that we did in five minutes. Is it great? No, but can we go use this in a compositor? Yes. Like we can go put this in a in the in the rest of the scene. If we had another scene where we wanted to place this, right? If we wanted to go put this on a countertop or an environment, we can go do that. there's a lot of different things that you can do here. We can also even, you know, from this video.

Play from 0:47:43 8:88:88 0:47:43 Right, we can take this and color correction. Let's say we didn't like the colors here. We want to make it a little bit more vivid before we run it through this Gen Effect, even though it's going to take all the color out, just showing a different way to do this. Right, like we can change the exposure, the contrast, change the saturation a little bit, maybe make this more blue, give it more pink tint here, give it some highlights, some shadows. Right, so we can.

Play from 0:48:10 8:88:88 0:48:10 Definitely play with this in a lot of different ways. Just showing one simple way to get in here and start like making stuff. That's where it all starts. Then once you realize I can make stuff, then you just go. Right. So think yeah. Drew Brucker (48:20) That's like the that's the friction, right? Like people that are haven't been in here yet, it feels overwhelming. Where do I start? This is how you get started. Rory Flynn (48:26) Mm-hmm. Yes. So this is this is as simple as it gets. then you know, we can kind of keep pushing from there.

Play from 0:48:36 8:88:88 0:48:36 Like this is how I think everyone starts. Right? You start with kind of like, let me just drop some stuff in here, see where I go. How does this work? How does this fly? Because then it's like, wait, Drew Brucker (48:44) If you Rory Flynn (48:44) no, now I need to build systems. Drew Brucker (48:46) If you're I if you're doing it for the first time, like the best way is to give yourself a little assignment, something that just comes to mind that you want to create. And you you actually did that for me when I was just starting with this. You just said, I think like come up with can't you remember what you said? You're just like, Different angles of a character and do this and that and the other. And it's like, okay. You know, and so I had sat with you, we had, we had gone through it, but like the hardest part is just getting started. So we tried to cover that here, right? Like just getting going. and I don't know if I'm, you know, jumping ahead here, but like this was the first one of the this might have been the first thing I created in here,

0:48:46Scale one fashion image into reusable assets

Play from 0:49:24 8:88:88 0:49:24 actually. which is still like an awesome example. I'll show this from time to time for people that. really aren't familiar to with what's possible in here. But this is really cool because like it's just this idea that you can create net new assets in perpetuity, And this applies in so many different ways, right? We talked about the micro use cases. This is a little bit less of a micro use case in that fact it this may seem a little bit overwhelming to some, but the idea itself is very simple, right? what if I was a clothing brand And I had, you know, an outfit or attire or a piece of clothing that I wanted to showcase.

Play from 0:50:03 8:88:88 0:50:03 Okay. And then let's just say I had a model, whether that be real or not. I think like the fact that you have a real model is probably even more valuable, right? Because you could capture a model now and think about having them sign a release where you can create new images of them from different perspectives or in different locations. Okay. I've got my model. Dude, I mean, like what? Rory Flynn (50:20) I would love to get royalties that way. Do whatever you want with me, I don't care. Drew Brucker (50:24) Dude, that's what I'm saying. Like, and so like that way, also you're working off real material here. So I like for this example for this assignment, I just went on Pinterest. but basically I came up with three ideas, three core pieces of this, right? The outfit, the person, the location. Okay. And then combine those into one image. It's not connected here because I think I did this like outside of that.

Play from 0:50:49 8:88:88 0:50:49 but you can actually do that here as well, right? With with just Nano Banana or whatever the the the model is that you're using. And then once I have this, right, I've established my core piece that I'm gonna use throughout this. And I've got sort of this prompt. And this is where it came down to sort of like what you were telling me. It's like, hey, like think about different perspectives or angles. So it's like I wanted to kind of like go really wide with this. I wanted to test a lot of different angles too to see how they would be interpreted. So describe 20 distinct.

Play from 0:51:17 8:88:88 0:51:17 Camera angles, for each scene, highlighting the element from the image, capturing the look and feel, keep every description under 30 words, separate each with an asterisk. This is your, this is your little thing right here. the way that you like to set this up. So essentially, then I have this prompt and I have this image and I'm connecting it to an LLM node now. And so this piece is key with separate each with an asterisk because it's then creating sort of like these, these breaks in.

Play from 0:51:45 8:88:88 0:51:45 Each of the 20, right? So I've got a close up, I've got a mid shot, I've got an aerial high angle, right? And now I've got these established. Cause once I click this, it's it's gonna give me those those camera angles. But I can also say, hey, look, like maybe some of these aren't quite as distinct enough. Maybe there are two that like just don't feel right, right? And I can either regenerate that or I can just pull from other material and just say, no, I want a low, I want a low angle shot. I want to do like You know, something from like an aerial shot. Maybe that didn't list it here. Okay. And then we get into this part here. And maybe maybe this is getting a little technical. Maybe we're getting a little ahead here. But essentially getting into something called array, which then separates these out by that asterisk mark. which that was such

Play from 0:52:31 8:88:88 0:52:31 an awesome idea by Rory to set that up that way. And then I've got this piece right here. Which, which is these these 20 boxes that you'll see here. But list, these are lists, right? So now I can just go and select, which of those camera angles it's taking. Then I'm driving that over here, right? Where I've got the 20 different angles. So I'm creating 20 different image, you know, boxes. And I'm making sure like when I click that first one, right, which model I want. Also, I I did state the The aspect ratio originally, but you can also change that here.

Play from 0:53:09 8:88:88 0:53:09 And now all I'm doing is once I've got these 20 boxes, is highlighting this. And then I can just hit run. Right. It's gonna tell you how many credits it's Rory Flynn (53:17) Bang. Drew Brucker (53:17) gonna take up. Boom. Right. Done. And and so, like to the the scalability of this is well, what if I don't like one of these images? I just select and rerun it, you know? or right. Rory Flynn (53:28) Go back and change that prompt. You can see it. It's right there. It's listed. Yeah. Drew Brucker (53:32) It's all right there. And maybe what you do before you do the 20 is maybe test two or three. See if it interprets it the right way first, right? Before you spend all the credits. And then it's like, okay, yep, head in the right direction. Let me run the rest. I love how simple it is here. All I got to do is drag, run, boom. Right. And so it's just like this is insanely powerful for a number of reasons, right? But like if you're a fashion brand or something like that,

Play from 0:53:56 8:88:88 0:53:56 you could take one existing image that you have, or maybe there was a missing image or camera angle or the lighting wasn't great or something like that that now I can just create from and I don't have to send the photographer back out there. I don't have to spend additional money re-renting the studio space or doing this. Or maybe I've got like this shirt, Rory, and I forgot to do a colorway with it, but I've got the hex codes or the, you know, the color codes for those and I can just swap in the color codes. Right. Or I've got a different hat that I want to show on that same outfit that's in my inventory. Or I've got different colorways or variations of this Nike shoe that we offer on our site. Right. Like, so you can quickly see how you can scale this across. And it's extremely powerful because a

Play from 0:54:42 8:88:88 0:54:42 lot of these companies that are spending a lot of money on photo shoots, Now is the photo shoot still necessary? I I think in a lot of ways it it still is to some degree. But now you don't need as much starting material. And you can fill in these gaps and you can scale in the net new, you know, Opportunities in perpetuity very quickly. so it kind of comes down to like just Rory Flynn (55:01) That's the idea. Drew Brucker (55:01) having, you know, a single product shot or several shots of that same product around. So you get all of that tangible material about the product so that it's transferable and can be scaled without, you know, any of these models hallucinating it. And that's also how you keep the accuracy up.

Play from 0:55:17 8:88:88 0:55:17 Rory Flynn (55:18) So I think you highlighted something there. Going back to the might as well go through it. the array stuff, right? let me just take over on this one and I'll go because this one, I'll just I'll we'll build it. We'll build it live. You can see how to build a Drew Brucker (55:30) Let's do it. Yeah, let's do it. Rory Flynn (55:31) batch, a batch processing engine, We'll do it again, super shorthand. A lot of this stuff has to get dialed in. I spend way more than the five seconds you're seeing on this, you know, doing this kind of stuff. Right. So let's maybe just duplicate this. We'll bring this down.

0:55:18How to build a batch-processing engine

Play from 0:55:45 8:88:88 0:55:45 Like that batch processing engine is great because it'll have it'll have like multiple things, you know, you'll get multiple outputs with one click. Right. So let's say we wanted to use our our Apple asset here and we wanted to make it into three different things, right? So let's just write a prompt. we're going to trying to figure out how to do this on the fly, make it not crazy and make you guys sit here. But Apple and turn it into an orange.

Play from 0:56:15 8:88:88 0:56:15 A banana and a what's a good fruit? Give me a fruit. is that you just you're on mute? Drew Brucker (56:19) Kiwi. Rory Flynn (56:20) Kiwi, good one. Drew Brucker (56:21) Kiwi. Rory Flynn (56:23) Kiwi. let's see here. write three prompts to preserve the exact style placement. Fire. Fruits this is where little system prompting comes in again this would normally be I'd be running this through Claude and just like tightening this but we're we're writing this live like format or like output guidelines Let's just say no additional commentary. Just produce the prompts. and each prompt with an asterisk. You'll see why. And then each prompt should be fifty to seventy-five words.

Play from 0:57:06 8:88:88 0:57:06 All right, so this is the prompt I'm taking. This is how I take one prompt and then turn it into multiple. Right? This is the branching element. So if we do an LLM here, you can do run any LLM. any LLM. Okay. So here, just connect the image, connect the prompt. This one should now basically you can select whichever model you want. you know, typically I just do the Gemini because it's really fast. here, I mean there's different uses for all this. Claude is sometimes better. I like ChatGPT for a lot of things. For the purposes of this, I'm using Gemini because it's really fast. So We'll just see if it does what I'm what I'm asking here. hopefully it will, Drew Brucker (57:46) what you said though is important though, which is like if you think about it in terms of like a, you know,

Play from 0:57:53 8:88:88 0:57:53 like a templated prompt, right? Where you know you want to include certain things each time based on the ask, right? You can Rory Flynn (57:59) Mm-hmm. Drew Brucker (57:59) quickly adapt that. And what Rory's saying is just because we're doing this live, like normally he would spend some extra time just tightening down what exactly that framework of the template looks like to ensure he's covered everything that he wants to it's. Explicitly call out, generally speaking, in a lot of these, the more sort of like structured and specific you can be calling things out, do's and don'ts, the better. but again, like we're just kind of running with this on the fly. But that would be one of those pieces of it, spend the extra time up front on, right? Outline those things. And also after after you do this a few times, you're gonna kind of know what those core pieces are.

Play from 0:58:39 8:88:88 0:58:39 Rory Flynn (58:38) So this goes back to what you were saying before, like with the with the array and the run or the with the with the array and the prompt splitting. Now I have three prompts in this one box, But there's only one output. So if I put it straight into another Flux 2 node, right? It's gonna ru it's basically gonna run Drew Brucker (58:57) All three. Rory Flynn (58:58) this as one prompt and it's it's gonna be like crazy, right? So we need to sort of decouple it from that sense. So, you know, here that's why.

Play from 0:59:07 8:88:88 0:59:07 You can run the array. There's a few ways to do this. Array is one. Meaning, like the reason we had this here, basically end each prompt with an asterisk as asterisk is so that when you connect this in here and it says split by asterisk, that'll turn it into three separate inputs. And then from there, you can do this. You can either run this straight into, List, list selector. And that's one way to split these out. And you can do I like to do it this way. You can do it other ways. because I like to see everything.

Play from 0:59:42 8:88:88 0:59:42 Drew Brucker (59:42) I you know what, I I I like it this way too, because it breaks everything apart. So it it feels more understandable of what's connecting to what. And I can I'm just speaking from like when you first Rory Flynn (59:51) Yes. Drew Brucker (59:52) get in here, that's the way it felt for me. It was just like, well, I d because I see three boxes there, I know what I'm getting, like I've got three prompts there. And so just seeing Rory Flynn (1:00:00) Yes. Drew Brucker (1:00:01) that number match made it more digestible of what's going on. Rory Flynn (1:00:05) So now we can run this will be prompt one, right? We'll just duplicate this two more times. Right. We have our two and our three. Right, we'll make this one to this one. Come on. And then we have our tech we have our text prompts cover, but we also need our image in there. You can just go drag this straight over there,

Play from 1:00:27 8:88:88 1:00:27 or you can use something that's like a little bit more maintenance specific. You can use a router. Which will just like run this into here. So you don't have to run three specific strings over. You can just run it straight into that. And then just bang, bang, bang. And then you can run these all at once. Right. Drew Brucker (1:00:45) Is your last is your last list connected to the image? Okay. Okay. Rory Flynn (1:00:50) it's not good catch though.

Play from 1:00:54 8:88:88 1:00:54 Let's see, we'll cancel this one once it Because there's two, there's another way to do this, right? You can do basically with the text iterator, the other thing here. Where I'll show you the two branches of this. Regardless, let's just run this back, do this again. All right, so we're creating our our pixel. Our pixel fruit. But this is one branch, right? Like you can do it this way where the array, this will split it out. Here's three, one, two, three, right? The prompts that it's creating runs the three separate nodes. This is if you want to keep things a little bit tighter. Text iterator. It's the same thing as the array, except it does array and list in the same thing. So when you connect text or array into here, split by this, you know, you have your three prompts. It's all broken out. And then if I do

Play from 1:01:42 8:88:88 1:01:42 Flux Pro 2 here, right? Let's make it closer so we can see it. Hit the run model. And this will now run all three at once in this node. Or all four at once. Yeah, so Drew Brucker (1:01:53) so it's just gonna tab with the arrow, right? Showing there's three images. Okay. Yeah. Rory Flynn (1:01:57) this way. god, I forgot to connect the things. See, so sometimes you know, moving too fast, you have no sleep, you're doing dumb stuff. I'm moving fast and not not thinking because we're doing this live, of course.

Play from 1:02:10 8:88:88 1:02:10 But regardless, it's it's still gonna show up the same way. It's running all three in here and you can toggle through whether you want to look at them all together. You wanna look at individual by batch, And then if you wanted to, let's just say break these out and I'll use these as individual nodes now or all individual inputs. You can go here and you can create you know, either you can unpack, right? Which will break all these out, and then you have them all as.

Play from 1:02:38 8:88:88 1:02:38 individual input nodes or individual upload nodes or you can turn this into an image iterator which is like okay now if we put in five of these images and we said make this all chrome right like if I've ran this thing now through our prompt up here make this all chrome it'd run all five at the same time do that system prompt. So again ways to do batch processing but These two things right here, this is like the end, this is like an engine. I plug this thing into everything that I do. Where it's either, you know, prompt and then into any LLM into array, into list, into batch, or you know, prompt and then like a system prompt, right?

Play from 1:03:21 8:88:88 1:03:21 Which might be way more extreme, which is what I'm doing more so of now, which is just like Drew Brucker (1:03:24) Hey, Rory, just for people that are newer to it, explain the difference between Rory Flynn (1:03:28) Mm-hmm. Drew Brucker (1:03:29) prompt and system prompt really quickly, just so they can help delineate between the two. Rory Flynn (1:03:33) Correct. So okay. Prompt here, I'm running this directly into the run any LLM, right? So this is going into the prompt section the same way that you would do in ChatGPT Claude, here's my prompt. Claude goes and answers. System prompt is what works behind the scenes to tell Claude how to respond or how what to do every single time or how to make ChatGPT do that every time. So if I want to take these, this would probably be more so.

1:03:24Prompt vs system prompt: what's the difference?

Play from 1:04:01 8:88:88 1:04:01 suitable for the system prompt. Drew Brucker (1:04:05) Rules structure guidance. Rory Flynn (1:04:06) This rules structure. How you want it to output, how it should output every time, w how it should, you know, how many characters should it be? How you know how do we want this to if you want it to be more creative, right? In Drew's example, it's like, okay, I want to do 20 shots, but like of the 20 shots, maybe five should be close-up, five should be long shot, five should be from a crop perspective, five should be this. Like typically use this type of camera, use this type of tilt or whatever.

Play from 1:04:35 8:88:88 1:04:35 This type of lighting. Like there's a w system prompt's most malleable thing in here, but it runs everything. So basically when you connect that into here, it'll work every time. This can change and it'll just be like if I said output nothing in the system prompt, and no matter what I put here, let's say, you know, regardless, I'm I wonder if it'll listen if I do this. So it's like the rules, right? There's just a s maybe a good way to understand it. Output nothing. Only output, you know.

Play from 1:05:04 8:88:88 1:05:04 Dot dot dot. We'll see if that works. Probably won't because I'm trying to do this on the fly. But like essentially it's like governing to that extreme. Like I can put in whatever I want, right? It's just be like, Nope, I only do this. Right. So that's the system Drew Brucker (1:05:14) Yeah, there you go. Yep. Rory Flynn (1:05:16) instructions, right? So you can get as deep or as crazy and we'll go into some of that stuff, as you want to. But Drew Brucker (1:05:22) It's like the o it's like the overlord of what you're trying to do. Also, it that that part feels more of like the static piece versus like the prompt itself could be dynamic. Maybe you're changing things out, but these are the the governing rules, guidelines, et cetera.

1:05:38Turn a node graph into a reusable tool

Play from 1:05:39 8:88:88 1:05:39 Rory Flynn (1:05:38) Exactly. So this is this is the other thing worth showing. now that you have both of these, I'm just gonna take these out here now for for cleanliness purposes. We have this little workflow here, now we can plug in basically any idea. It'll take this Apple pixel as reference and we can change to anything. Right? This is this is pretty simple. To make this like an actual workflow, you have to use these output nodes. So the output right here.

Play from 1:06:06 8:88:88 1:06:06 You just connect that. That signals that like it's the end of the workflow. This is where you stop. Right. So I'll make two more of those. So we know we want, you know, if we're doing three at a time, right? This is logically, I know some people like this makes no sense, but just bear with me here, showing a lot of things all at one time. Drew Brucker (1:06:23) Yeah, hang hang tight 'cause this is it got some value here. Rory Flynn (1:06:27) So now that these are all connected, now you actually have a tool. And when you go into tool, you'll see here's my, you know, here's my regular prompt, here's my system prompt. And then I can just basically write, you know, instead of an Apple banana, let me go back to I took out a crap. let me write these output rules again.

Play from 1:06:49 8:88:88 1:06:49 Dapper rules, you know, only Yeah. Prompts. Output that's a that's a rough one for me, Drew. Only output the prompt, no additional. Sorry, we're doing this again. Commentary. and every asterisk each prompt should be each prompt should be fifty to seventy five words. Great. we'll run this again. But as you'll see here we can see how this works, but within the tool, you know, you'll be able to see what's going on Drew Brucker (1:07:23) I I think there's like Rory Flynn (1:07:24) There we go.

Play from 1:07:25 8:88:88 1:07:25 Drew Brucker (1:07:25) a big the the system prompt, just to elaborate one one more thing on the system prompt. I mean, like if you're thinking about this from a team standpoint, multiple people being a part of this, right, you can see why an image looks the way that it does, right? Like you can go back to sort of the foundation of this. Whereas if it's just a a like, for example, a prompt history of something or a prompt itself is is like You're like digging for fossils trying to find those things.

Play from 1:07:56 8:88:88 1:07:56 And I bring that up because there are a lot of enterprises for for legality purposes too that want to know what the prompts are. so, like, you know, imagine digging back through hundreds, even thousands of prompts, which I I've had to do. right. Like this is an easy way to see exactly why you did or didn't get the result that you want. And it's and it's immediately. discoverable. just so it it just saves a lot of time. So I think that's a just a nice little thing to bring up, especially if you're doing this work for for a client.

Play from 1:08:30 8:88:88 1:08:30 Rory Flynn (1:08:30) that's a good point, by the way. There is there is kind it's also a great way to learn. You can go back and see everything that happened where stuff went wrong, and it's easier to fix this way. but now because we have these outputs here, this turns into a tool. Right. So that meaning, if I just rewrote the prompt here and uploaded the the image ref, everything would everything would fire and it would just run here. Let me go back to the canvas, make sure everything is connected here. One quick correction. We'll do an upload here and an import. And then let's add this Apple back in just so it starts the workflow. So this can be like whatever you want it to be.

Play from 1:09:07 8:88:88 1:09:07 The file here back to the router and we're good. Now this tool should be we should have that input there. Great. So now you can change this to whatever you want. But as you'll see here, this is what you can hand off to someone. This is now a tool. Like I don't have to look at that workflow. I can just be like, I want to do, you know, I can make this so that it you don't see any of it. You just type in the fruit. And it's just like we'll label this fruit, and someone types in you know, coconut, enter, and then it goes, right? System prompt here, you see that it's labeled most of the time. I like to I like to get rid of that. The one way to do that in the tool view is to lock it. So if you lock it, you can't touch it, but also in the tool, you won't see it. So Drew Brucker (1:09:44) Yeah.

Play from 1:09:46 8:88:88 1:09:46 Rory Flynn (1:09:45) then, you know, you can also drag this up here if you want it to look exactly like that. And then, you know, we hit run. And then this will all run in here. All three will happen at once. And this is much easier to give somebody on your team. Be like, here, just drop in the image and tell it what you want. And then, and then Drew Brucker (1:10:02) Dude. Rory Flynn (1:10:02) it'll run. And then you can do that. Right, the verse like trying to engineer this. But I built it once, now someone else can use it. Right. Drew Brucker (1:10:09) Consistency, baby. Consistency at scale is the product. That's the thing that Rory Flynn (1:10:14) Yeah.

Play from 1:10:15 8:88:88 1:10:15 Drew Brucker (1:10:15) brands pay for. And so this, you know, the the one shot generation stuff, you'll you'll be flying through hours. And so the the you're trying to get high quality and consistency at scale. This is the best way to do it. Rory Flynn (1:10:31) And it comes down to a lot of what's in here. I would do the same thing. I would probably have this same prompt. Like again, if you wanted to do and that's the great thing about these, you know, these effect nodes, right? If we were to bring this back to maybe bring these back, Exposure grain. There we go. Like if you wanted to use this now, this also works on This also works on images, doesn't have to be on video. So now we have, you know, our orange, which kinda looks like our apple. We have our banana. We can just

Play from 1:11:01 8:88:88 1:11:01 run that into here. And then we have our it might take some little finessing here based on Drew Brucker (1:11:06) Yeah, you could but that but that's the that's the best part about like the you can literally Rory Flynn (1:11:09) Yeah. Drew Brucker (1:11:10) because it's like this shit you it's just this threshold you can drag. I mean this is the fun part too. So if it doesn't quite look right, adjust it. Rory Flynn (1:11:17) And that's how you know we can get from point A to point B on this stuff. Now this won't be part of the workflow because it doesn't have the output attached to it. Right. If I had run this through if I change this, you know, if I change this to here, and then this output gets changed to this, right? It would want it run it through one extra step. So the orange would basically become the only thing that's outputting in this style, right? Have to change it to this would have to be the output.

Play from 1:11:46 8:88:88 1:11:46 And then this would have to be the output. Right. And now the workflow would go automatically. If this makes sense, it would run through everything. Like if we'll just run this one more time so everyone can see this. it'll go straight to the Game Boy. And we'll go to the, you know, we'll go to the the flux nodes. Drew Brucker (1:12:04) It's whatever that last step is before output, right, Rory? I that's the way to think about it. Rory Flynn (1:12:07) Correct. So we just finished that one up. That was for like anyone who's never used this before. Let's go, let's get a little bit more extreme now. this Drew Brucker (1:12:16) I like extreme.

1:12:17Engineering an automated listicle video

Play from 1:12:18 8:88:88 1:12:18 Rory Flynn (1:12:17) yeah, this is more like engineering. So this is a This is a tool for like those listical videos. Like you know you never you ever seen those where people just like either they write on the whiteboard where it's like, here's the best five tools to use for this and they write it and you somehow get hooked to it because it's very popular format and people like it, right? Yes. Drew Brucker (1:12:35) And it's perfect for dwell time, dude, 'cause you're like, I wanna I gotta stay till the end now, right? So Rory Flynn (1:12:41) Because I want to see what they're gonna say. You know, they'll use some catchy hook. Like one of the worst five restaurants in downtown Atlanta. And you're like, I want to see if that wanna Yeah.

Play from 1:12:50 8:88:88 1:12:50 Drew Brucker (1:12:48) Wait till you see number one. Yeah. Rory Flynn (1:12:52) Right? Stick around till number one. Right. So this one's just fun. Like I did this with you know, I did this with an anime character. Cause I'm like, this would probably be different, like an anime character in a real in a real setting. Oddly enough, I made this, you know, a couple weeks ago and then I saw it. Jboogz doing this style. And I'm like, this is so sick that he's doing it like it his way, which is like you know the real extreme anime in the you know realistic situations.

Play from 1:13:18 8:88:88 1:13:18 If you haven't seen that episode, go back a few. JBoogz good one. But this, Drew Brucker (1:13:21) That was a good one. Rory Flynn (1:13:22) this is way more engineering, way more complex than what we just did. So the best thing, the best thing that I can say for anyone who's been using this stuff and you want to start to build more. engineering type systems. You have to work backwards. You have to know what the end result is. And then you have to know how to get there and how you can control it. And that comes with the using the tools. You're not going to be able to step into this day one and build this. Right.

Play from 1:13:47 8:88:88 1:13:47 So, what we're doing here, right, is it a lot of these also have they have cuts. So it's like quick cuts. You either jump in, you jump out, you know, and you'll see it. I'll I'm going to mute this just for now But the goal is all right, we have Five, we take it off, right? Then it'll go to the next one to four. We'll cut. She'll take it off. Same thing. We'll cut three, two, one. Now, there's a lot of things to control in here, right? You gotta have basically everything stay the same. The camera's gotta stay the same. The movement here. Then also we gotta get all this stuff to work. Now, I'm knowing that meaning probably I'm gonna need to do like first frame, last frame. Right? I'm gonna need to have First frame is where it starts, last frame where it ends. So like I'm gonna have to start with everything covered, and then the last frame will be when you just see

Play from 1:14:35 8:88:88 1:14:35 five and compounding. Same thing. And then we loop that back, then we loop that back, and then we loop that back. So, you know, for this, probably we need something like Seedance or Kling could probably do this, or Grok now could do this, but a couple weeks Drew Brucker (1:14:51) What do you what do you think is the right now is the is are are models like H3 and Seedance able to kind of do all of this in one shot, meaning go through one through five versus, first frame, last frame, here's before the reveal of number five, then after the reveal of the number five, here's before the reveal of number four, after, right? And stitching those together. What like are we at that point yet?

1:14:51One-shot video vs modular AI production

Play from 1:15:20 8:88:88 1:15:20 Rory Flynn (1:15:19) We could be, but also like I think about this from production standpoint, right? Like let's just say this was being you were doing this for the marketing team and they said, you know what? We want to change less errors to, you know, error resistant. Right? It was like Drew Brucker (1:15:33) Good point. Yeah. Yeah. Rory Flynn (1:15:35) some stupid change. And you're like, well, now I have to hope this whole thing works and run, you know, 30 seconds over again and over again and over again and over again. Could it do this? Probably. right now, this was before.

Play from 1:15:45 8:88:88 1:15:45 Seedance 2.5 or 30 second generation. So we're stinging it together, string it together. But again, think about this as like more engineering standpoint. So that's how we have to start. We have to get there. It's going to composite all of it. Where that starts is all right, well, we need to get the board, we need to get the character, and then we need to get like the videos to work. So we go all the way back to the beginning. I like to use character sheets. Pretty great way to to give, you know, reference to a character. Again, like multiple angles of a character, multiple facial expressions, Drew Brucker (1:16:14) Yeah.

Play from 1:16:17 8:88:88 1:16:17 Rory Flynn (1:16:16) right? Gives it all the data that it needs when it can read all of that. What I'm inputting here, because what we want to do is first I'm going to create the actual board. I think it's easier to do it this way is to create what's actually on there and then add whatever effect you want to put in there, whether it's you know torn cardboard or something or it's not written yet. you know, I would probably stay away from writing something because it it's probably just going to be weird with the AI models. But The listicle, right? We have basically what I'm listing here is like what is in the listicle, what's gonna go into here. So I'm splitting that out so that listicle can always be changed. This can be anything now. You just write in one through five, good. that'll take that and put it into the prompt here, which is I gotta unlock so we

Play from 1:17:02 8:88:88 1:17:02 can look at it. Which is always it will come in here, and these are two factors that I have coming in. So we have you know the listicle actual list itself coming in as one variable. As variable one and then as two, the location. Like I can make this indoor, outdoor, whatever it is, right? So that it knows like this is basically the input prompt. Here's what I'm trying to do. Here's the location that it's gonna be. you know, you can also add images for this. Now, the system prompt, this thing gets really extreme. you know, you look at the JSON, see if I can text. Can I expand this and actually can we actually look at this? Regardless.

Play from 1:17:40 8:88:88 1:17:40 JSON prompt. I typically run through a very similar structure here on most system prompts. So we can just look at it. So we'll have our role. What it's going to do. And why am I using a system prompt here, right? Because this is going to run into an LLM and I want the LLM to output things that are very specific. So I'm going to give it specific instructions, This will be used as the input into the LLM. This will be used as the system prompt saying, like basically do this every single time, right? Your role is you're going be a Nano Banana prompt engineer. What are you going to get? You're going to get this character sheet and the input from me.

Play from 1:18:14 8:88:88 1:18:14 Like the text, the text listicle and the text location. Those are the three things you're gonna get always, What you're gonna do is you're gonna analyze the character sheet, you're gonna parse the listicle, right? You're gonna understand what the rendering constraints are. And then we're not gonna add anything else, just that. We're gonna have the same identity lock. We're gonna have the environment. you know, should be should fully be defined in the setting and the whiteboard or whatever it is should stand in a very specific way, there's a lot of information here The format, how it looks, the lens type, the assembly order, how the the whiteboard text should look itself, like the how the handwriting should look. Human and flawed, but every word is readable and correctly spelled, right? Like it's not because if you don't write stuff in like this, like it'll be way worse than what it looks like here from

Play from 1:19:02 8:88:88 1:19:02 an AI standpoint. Like this actually doesn't look terrible and then I'm saying output. Only this. So basically I'm telling it exactly how the LLM should write the prompt to go into Nano Banana. Right. So that's that's what I'm writing here. So essentially, like we said, every time you just input the listicle, the outdoor this system prompt takes over the rest. We have the LLM here. Write the exact prompt. We gave it the exact format to output it in, so it'll do it every time, which is great. This now will be able to take the character, whiteboard setting.

Play from 1:19:36 8:88:88 1:19:36 Place the listicle on there every time I did it with let's see if I can go back. I don't know. maybe I don't have the one. Do I have the ones for me? No. I had I ran this with myself doing it, and it was hilarious. I guess I should have kept it in there. regardless, now I have this. Drew Brucker (1:19:53) I wanna see that one. Rory Flynn (1:19:54) I gotta find it. I think this is a duplicated version of it. Drew Brucker (1:19:57) That's awesome. Rory Flynn (1:19:58) but here, this system prompt now, what I'm doing is adding the cardboard over.

Play from 1:20:03 8:88:88 1:20:03 Text right. So every time you get this, now you get this. So I can run this start to finish, and this is where it'll start. Nothing here is I once I put in the listicle, the outdoor location, and the person, this would run basically up until this point. I put the checkpoints in here because I want to stop and make sure this is right before I go and spend a bunch of Seedance credits. But what we have now is the batch processing system, which will take this.

Play from 1:20:32 8:88:88 1:20:32 And this system prompt is absolutely insane. Really long. But basically it says you're gonna create the next One, two, three, four, five images in the sequence. Exactly what happens. So you start at five, It's gonna say, you know, you start at five, you go to four, you uncover three, you uncover two, you uncover one, using this as the first image in, so it can keep the text exactly the same. So the handwriting is the same across the whole thing.

Play from 1:20:58 8:88:88 1:20:58 Right. So it's basically like first frame, last frame, but with image generation. and then you know, again, we should go through our array. It's it's creating those prompts. As you can see here, we have everything, and I'm telling it explicitly every time that the person should be holding the piece of cardboard and in a different position. Right. So it's like we're taking it off with a different hand every time, and she's in a different position, like she's talking. She's not just like in the same spot. That'll help us with the movement and the cuts so it doesn't look totally flat.

Play from 1:21:26 8:88:88 1:21:26 Now, now that we have this, basically we're we're putting in our script. I'm putting in the script like whatever I'm saying here in this input. System prompt saying, take the script, write the prompts for Seedance that are going to be first frame, last frame, right? Again, same batch processing system that we just built before. And then I'm adding a just a voice reference in here for Seedance. And then it'll go and do these. Right. So it'll go and take every single one of these.

Play from 1:21:52 8:88:88 1:21:52 And then we use a compositor, which is basically just like a timeline editor. And if you do edit, right, you'll see everything in here is just baked in. Right. And we just run through it all, same way that you would. And this can now be turned into a tool if you wanted it to be. As you can see with here, I probably took one of these prompts off. Which one did I which one did I unlock? This one. Great. Let's lock this one back up.

Play from 1:22:22 8:88:88 1:22:22 So it should just be the character sheet, the listicle, the script, and the location, and then this voice input. So now this is what I would send to a team member who wants to create the listicle video. They don't have to do any of that crazy engineering that I just did, right? They can just do tests. So that's that's a more extreme example, right? Very, very very, very like.

Play from 1:22:47 8:88:88 1:22:47 technical and going from here to here and this has to be this way and this has to be this way, but it runs like a tool now. Now it's one click. And it's really as minimal. My goal is always to trim the input to as minimal as possible. Like I don't want anyone to have to change a lot of stuff. I I don't want this to be like, well, once I put the stalistical in, I want it to be like this and the person has to be and have to choose the camera and it has to be this way. That's just all if you're gonna do something that's this extreme, that's gonna be way more way more complicated and way more possibilities for error.

Play from 1:23:20 8:88:88 1:23:20 Drew Brucker (1:23:18) Well, well, to your that's that goes back to your idea, though, that have a singular use case in in in a lot of these situations for these. That would because otherwise you almost have to make it a bit more ambiguous in order to apply to several different use cases at the same time. It also makes it more complex, Because you're able to get very, Rory Flynn (1:23:41) Great point. Drew Brucker (1:23:42) very specific here. And this is one particular use case.

Play from 1:23:47 8:88:88 1:23:47 Right. Maybe now maybe there are parts of this right that could plug into a different use case. But Rory Flynn (1:23:52) Mm-hmm. Drew Brucker (1:23:52) the point is, is anytime you're going to do something like this, and put myself in the audience shoes, anytime you know you're going to do something, you now know exactly where to go for that. And you know that there's not going to be a lot of confusion in trying to we rework that. You've got a very specific thing and you're able to be very specific because of it.

Play from 1:24:12 8:88:88 1:24:12 Right. You start to introduce all these other variables in trying to broaden it too much, then it it's just it it becomes a little bit more difficult to manage. So I love that you Rory Flynn (1:24:20) Yes. Drew Brucker (1:24:21) you sort of like really frame that up. I think that's one of the best probably framings of this entire episode is thinking about it like that. Rory Flynn (1:24:29) 'Cause it's tough, man. And look, I think probably the the other thing that we should look at here is just like usage like usage. Like this doesn't have to be the full production engine in this, Like it doesn't have to be you don't have to do everything with it. It can be more so like BarkBox. Where is this? This one, right? Good example here. Big problem. If you're not familiar with BarkBox, there's subscription toy box for dogs.

1:24:29How BarkBox fixed one production bottleneck

Play from 1:24:54 8:88:88 1:24:54 They get sent out every month. They typically have new partnerships every month. They have multiple variations of boxes. Meaning they have to get all these toys and things into one location every month to take the shoot the photo shoot to then also scale that photo shoot across all the different assets, right? So all the different marketing channels, all the different assets they need, email, paid, social, internal stuff, you know, presentations, whatever they have to do. It has to get all get done. So we sort of looked at like where.

Play from 1:25:22 8:88:88 1:25:22 Instead of like, how do we build a machine to create all of our organic social? It's like, let's find the one piece that's causing the friction in all of this. So like going back and like digging through, it's like, these composite shots, because it takes so long to get them done, and then sometimes some of the pieces don't ship or they get shipped and they're damaged. What do we do then? Now we have to wait longer. We might not get the photo shoot done, which then all the marketing assets get held, and that's how we keep our subscriptions going. Right. So this tool.

Play from 1:25:51 8:88:88 1:25:51 how do we just get that one reference image that we can use as like the vector to everything else? And here it's just putting inputs in, right? So we have a box, we have our PNGs of the toys that are going in, PNGs of the of the treats. We can swap any of these as we need to. We run it to the compositor node, which basically we do some traditional compositing here, and we just place it in so it can it can look like that. And then our system prompt, we run it.

Play from 1:26:21 8:88:88 1:26:21 Just this retexture prompt. What it's doing is it's saying you're gonna get this PNG with no background. And what you're gonna do is you're gonna add depth, like shadows, light, and shading so that it looks real in a white studio. And then we have this asset, right? This becomes, you know, something that's really important in everything. I don't know how far I'd have to go back in this presentation to find the close-ups. Drew Brucker (1:26:44) That's sick though. I mean it's just that difference, right, between the composite and then just that those little tweaks to get it to that that state.

Play from 1:26:55 8:88:88 1:26:55 Rory Flynn (1:26:54) Yeah, 'cause it's like right here, flat, right? Like you can see there's no depth shadows, everything just looks like a like you did it in Canva. But we add Drew Brucker (1:27:01) Yeah. Rory Flynn (1:27:02) the lighting and it looks real, which gives it the dimension, which Drew Brucker (1:27:05) Yeah. Rory Flynn (1:27:06) then we can take and we can bring it into something like this, which is again another batch processing engine. These are sometimes the different avenues that we're going for whatever marketing assets we're trying to do. Some of them are like sort of these vivid studio shots where we're just using this. This one vector piece becomes like an asset for everything we're doing. Like maybe these are for the social content, right?

Play from 1:27:29 8:88:88 1:27:29 Drew Brucker (1:27:28) Zoom into those too, Rory because I want to just sort of like think about people that may be out there that push back, right? On how real AI integration into the process can get. Guys, this is as real as it gets. Like that's insane. Like if if you did not know that AI was a part of this process, you would not know that AI was a part of this process.

Play from 1:27:55 8:88:88 1:27:55 Rory Flynn (1:27:56) And that's the thing, I'm not Drew Brucker (1:27:56) That's what Rory Flynn (1:27:57) trying to create like Hollywood films. Like these are things that have to get done. They have to get done somehow. Right? Yeah. Drew Brucker (1:28:03) But this is but and this is practical, right? It's useful. at the end of the day, right, if we're saying, AI, like AI is a part of this is like a great example of taking individual pieces, the product itself, the box, the logo, itself, like things that are tangibly have to be correct and are real, and then infusing AI into the part of this process to create this thing that is indistinguishable. And at the same time, like People shouldn't care that this is AI. It's that good, right? Because it serves the need that it needs to serve. just like extremely well done. I've seen, like, I've seen a few of these from a while back when you showed me. I mean, they're incredible.

Play from 1:28:44 8:88:88 1:28:44 Rory Flynn (1:28:43) And some of these are obviously mess ups and ridiculous, but some of them, like this one, you know, I would go and find like, there's a scale problem here. Why is this pug so large against these toys? They're not that size. Right. There's fixes for that. I wanted to show there's not all of these are perfect, but again, like this one's a little bit extreme. But I'm running this in in batch. So sometimes, you know, that's what you get. Right. Like that's the that's the reality of these things when you run this stuff in batch. you get things like this one, this one could play.

Play from 1:29:13 8:88:88 1:29:13 Drew Brucker (1:29:12) Ha ha ha. Rory Flynn (1:29:12) Like we can we can work with this one. That one's actually really funny. Right. Drew Brucker (1:29:15) That's awesome. Rory Flynn (1:29:16) And like BarkBox has a sense of humor. Like this is also good. Like this could also play. Again, we're thinking about this for email assets, for things that are, you know, that we have to do in a lot of email marketing campaigns, things like that. Right. There's also like some of the more ridiculous stuff where we're trying to do like like flash photography with it, you know, with just like extreme facial expressions where it's stupid, like whatever the marketing message fits.

Play from 1:29:40 8:88:88 1:29:40 Drew Brucker (1:29:37) yeah. That makes me think of the dogs that go that makes me think of the dogs that go underwater fetching like the ball or the tree that like Rory Flynn (1:29:44) Yeah. Drew Brucker (1:29:45) those photographers would do. Yeah. Rory Flynn (1:29:47) Exactly. So, you know, we're trying to get stuff like this, because this is just different. It's goofy, right? So that one that one little asset, right? That one finding that one little piece, this one little compositor and relighting can lead to this is just one of many things. This is like how do we create the assets, right? 'Cause then the other side of it, going to that scale problem, again there's just another instance where they do a lot of sales pitches because a lot of their their stuff is licensed, right? So a lot of their licensed material.

Play from 1:30:18 8:88:88 1:30:18 They have to pitch to these bigger brands and they have to mock up and get things sort of visualized to get the sale in so that they can get these new box promotions. Maybe it would be a South Park box one month. And then, you know, that they have to release that, right? So we're doing these sales promotions, they have or just you know, sales presentations, things like that, are just trying to visualize overall. A lot of their designers, but probably everyone designs their toys in 2D vector. That's what they get. Like that's how they visualize the toys.

Play from 1:30:45 8:88:88 1:30:45 That's how it's done from an internal system. So this is like this is a way to show how you take what's already happening in the organization and then just like you don't have to just scrap everything for AI. It's like, well, we have 2D vector. Okay, how do we work with 2D vector? Then same sort of theory applies with the system prompt here. It's like we input our 2D vector, we ask it to texture and materialize, add the lighting, make it 3D. Like now it's a toy, We also want to take that and you know we'll use our system prompts saying, here's the toy. Give us the nine angles that we need for every single toy. Right. So this is gonna be every single time front, you know, left side, right side, back, three quarter, three quarter, back, three quarter, back, three quarter, low angle hero. It'll do this every time, And that's the that's the goal.

Play from 1:31:33 8:88:88 1:31:33 But the other thing we're adding in here is the materials, like into this into the process. So we're adding the types of materials, the seam logic. So the seams are all the same on every single one. You'll see there's like the little, the little seams here, the little seams on the side of the head. The materials are exactly the same that we want to keep it that way. so then we have this, but we'll composite it into one sheet. We have all of these that we can now use if we need to use references for any of them. Composite it into one sheet to use as a reference for future stuff. But The real thing here is now that we take this and we want to go create some lifestyle images with it, you know, depending on where we're at. Let's see. There's a ton. We always got to this same point of holy crap, there's a scale issue. Like let me find some earlier ones here where

Play from 1:32:20 8:88:88 1:32:20 the perfect example. Like this is this shouldn't be that size compared to a a St. Bernard, right? Like this is probably more realistic in terms of what the size is, spec wise, to you know, like a greyhound, or this is what it should look like next to a I don't know what those are puggle, chihuahua, whatever you want call it. so we built we started building these types of diagram references into it to give it the visuals. You can type in measurements you want all day, it'll never come out correct. So we started using these 3D visuals where it's like, okay, here's the size of the toy against the different sizes of the dogs, So this problem here, it can be solved with where is it? There it is. Better

Play from 1:33:07 8:88:88 1:33:07 example of what we use. Really, like so we know that the plush toy is eight by eight. So we can give it scale reference against the Chihuahua, against the Beagle, against the Golden Retriever, right? That makes things a lot easier for us function-wise so that things don't look wrong. But again, it doesn't have to be, don't have to go and create these giant, like, you don't have to go create a full film. This one little thing we use more than anything, right? This compositor with the lighting text, this system prompt, like gets redistributed throughout a bunch of different workflows now.

Play from 1:33:41 8:88:88 1:33:41 Same thing. So it doesn't have to be all crazy. You don't have to go absolutely extreme. You can find these little small places to inject it. And then, you know, you can start to scale that because this is now this one system prompt is reusable a million different ways. I can also just you can save this, you know, you can save this node and then redistribute it however you need to. But, you know, if I was to go if I was to go back, I'm pretty sure I have like, you know, the and that workflow that workflow is repurposeable.

Play from 1:34:10 8:88:88 1:34:10 It'll work for any character that I put in now for that plush. I might just have to change some of the dimensions in the system prompt. But see if I can find the stand one. There's the Stan one. same process, right? Same end result. Just exactly the same thing every time. I could run Kyle through this, and then I could run whoever else, Mr. Hankey the Christmas poop. You can run through it, come out the same exact way, right? So we have our 2D, 2D going in.

Play from 1:34:38 8:88:88 1:34:38 Plush, angles. All this stuff is about production. Where you need it. Again, I like to break it out this way because then we can always change one if necessary, if we find something that's wrong. If the flap on the back of the neck here is different, if the you know the the mesh or the the pattern on the hat or the texture on the hat is different, we can always go back and change it. It's easier to see it this way. So I think that's like from a use case standpoint.

Play from 1:35:07 8:88:88 1:35:07 Easy to see how you you can just take this in a direction that doesn't have to be so big and it can be small. And that's how you can bleed it out to everyone. And then it becomes tangible and doesn't feel like anyone's getting their jobs taken. Drew Brucker (1:35:20) there's two there's two angles of this too, right? You're in-house somewhere. you have a mandate to start integrating AI and adopting AI into processes. If you haven't done that yet, or you haven't done this specifically, this is one of those big wow moments that you can provide internally.

1:35:20Where AI workflows create business ROI

Play from 1:35:40 8:88:88 1:35:40 About how your creative team is adapting and really cutting time, cost, headcount, improving consistency, eliminating some creative compromises, all as a part of that direct ROI. If you're on your own, and this is maybe something that you could provide as a service to a client, these are like Rory has done this.

Play from 1:36:06 8:88:88 1:36:06 I have done this, but clients want this. Clients want this, they will pay money for this. This is one of those things that you could leverage in a lot of different ways. You could embed yourself in the team, and maybe you're the one do you know building the workflows. Maybe you're the one that's upskilling people and handing off the workflows, right? Like, but a lot of brands are thinking about how they do this. And the the challenge.

Play from 1:36:31 8:88:88 1:36:31 remains consistent. We've brought this up in in other episodes, but these people in-house do not have time to just play and tinker on the clock. Most of the time they have deliverables that are due. They have timelines. They have other responsibilities. They have meetings. There is not this large gap of time where they get to play with these tools. So there's an opportunity to be sort of like that expert that comes in and does these things.

Play from 1:36:57 8:88:88 1:36:57 And I think like what we'll do too is like we'll probably share a few of these workflows in the description for you guys. so that way you guys can have some some jumping off points. And I would also just say with that, we will have an affiliate link that you can use that will give you, I think, 15% off, something like that, if you want to sign up through that. So I would highly encourage that as a part of your process. Now, Rory, I was just saying In general, with this, Systems can be delegated and they can be handed off. Like prompts and some of these other little things, like it's much harder to do that. There's the scale problem we already talked about. a system is a great way to scale. And it stops being so luck or you know, slot pulling,

Play from 1:37:47 8:88:88 1:37:47 right? So you're getting the consistency, which we talked about being a bigger factor. the other note that I just wrote down right here is just like t your taste, gets encoded and codified here, So it's at scale. So playing to your brand guidelines, your particular visual identity, right? So it's not something that needs to be re-performed by different people. You've got something that can be duplicated very easily. the other part that we didn't talk about, but is a problem is your workflows.

Play from 1:38:18 8:88:88 1:38:18 Survive model changes. These Rory Flynn (1:38:20) Mm-hmm. Drew Brucker (1:38:20) models are changing all the time. There's new models all the time. Well, rather than adapting to each of these models, right? Like you've got a place where your workflow doesn't have to change. But maybe instead of Nano Banana, you're now using Nano Banana Pro or using ChatGPT or or whatever it is, but the whole system isn't breaking as a result of that. And then each one of these workflows that you may build is is equity, right? Like you're you're building things that you can leverage for the future. We talked about front loaded work component of this. And I think lastly, I would just say like node, like this idea of nodes, node literacy is transferable here. Like this is a skill that a lot of people have not gone super deep in.

1:38:20Why workflows survive every new AI model

Play from 1:39:07 8:88:88 1:39:07 But this has got insane value. And again, like we were talking about this last year. Yes, more people have gotten involved, but in large, this is still something that most most people, most brands aren't doing. They might have the tool, but are they actually doing anything with it? That's a whole nother thing, right? Because the adoption and integration part is a whole nother beast in itself. So Rory Flynn (1:39:29) Just problem solving, right? I mean, that's what it is. It's all the tools there for. But like the other thing, you know, is we typically drop some nuggets here at the end, if anyone's still sticking around. I forgot to I forgot to show this. We're gonna run out of time doing this anyway. One of the things that's really cool now with Weavy is if you go and if you go and highlight that whole workflow that we just did, copy it. Just like, you know, right click, copy, drop it into something like ChatGPT or Claude.

1:39:29Paste your workflow into Claude to debug it

Play from 1:39:58 8:88:88 1:39:58 It will be able to read the whole thing and it'll be able to talk you talk to your workflow. So you can just be, what is happening here? Why is this not working? And it's just like, yeah, because this c this Drew Brucker (1:40:06) I didn't even think about that, bro. Rory Flynn (1:40:08) connection is broken. This doesn't this isn't working here. This doesn't make sense with your system prompt, blah blah blah. So because it exports and you know, if you copy the whole workflow, right? Let's just do do one more thing here real quick. Just show this. Drew Brucker (1:40:20) Do it. I didn't even think about that. That's a really good call.

Play from 1:40:24 8:88:88 1:40:24 Rory Flynn (1:40:21) Sure. Look, let's see. Like, let's let's find a small one. This one, maybe no. Where's that? Where's that box one? Yeah, I pulled up a lot here. pulled up a lot for you guys. Is this one? No. Nope. Drew Brucker (1:40:38) I'm surprised your computer didn't just give up at this point. Rory Flynn (1:40:41) Yeah. Didn't just explode. Let's just say, like, here. Let's just copy. This is a Seedance. Seedance prompting one. Let's just go to This one, if you just paste, let's see if it'll do it. I just copied it, pasted it, and you'll see it comes out all in code. So you can see that's that's what that's what Claude is reading. That's what ChatGPT is reading. You just paste that right into the 96 pages Drew Brucker (1:41:05) that's crazy, bro.

Play from 1:41:07 8:88:88 1:41:07 Rory Flynn (1:41:07) worth of. of chaos into there and then like can you read this? Yes, I can read this. Cool. What o what am I doing wrong here? What can I how can I change the system prompt? You know, how can I can you re-output a more optimized version of this now that it knows the the node structure? Right. And you could do some can't unfortunately do this with the with the MCP yet. but once that gets going, that's gonna be sick. So you think about that's the last thing that I'll leave that with, right?

Play from 1:41:37 8:88:88 1:41:37 Like this doesn't have to be like the whole thing doesn't have to be weave. Like we can go. I don't even know if I said this. I'm I'm basically hallucinating on this podcast. I'm so tired from not sleeping. Drew Brucker (1:41:46) This is a lu This is a lucid dream, bro. Rory Flynn (1:41:49) basically, I don't even know what happened. Blackout. But like think about if you're building multi-step processes in Claude Skills. Like I know one of the things we've done is build like a CAD process where it goes CAD to Blender. Blender's building mat assets. Assets is now.

Play from 1:42:05 8:88:88 1:42:05 Once those are produced and you know we're taking shots within Blender, like the shots of the product that we need, we're sending it to the one of the weave workflows that renders material texture on top of it. So it's just like it goes there, then it comes back, and then it goes somewhere else. It doesn't have to be the whole workflow, it's part of the rest of the workflow, So you can use those little tools, like we said, like, I need the I need the asset, like the the asset created the on the white background. Bang. You know, it goes and does that, but that might be part one.

Play from 1:42:33 8:88:88 1:42:33 The next run by, you know, the MCP is going to be much bigger as it gets going. It's like, now I have my my Claude skills that create the advertising, you know, the the social media assets. And we do it in a very specific way, but now it has the reference images because it's pinged to that vector created. Now it's going to go ping and create the assets. It's going to go ping, go and localize the text and go ping, right? So there's My brain's fried when you start to think about the possibilities. Yeah.

1:42:57Use AI workflows as callable functions

Play from 1:42:58 8:88:88 1:42:58 Drew Brucker (1:42:57) No, I mean d I mean the the MC P stuff's really interesting because I mean like like it it's really got an effective use when you think about maybe like the callable function of it. So like maybe you're Rory Flynn (1:43:08) Uh-huh. Drew Brucker (1:43:09) you're in ChatGPT or Claude or whatever, and you're just like, Hey, take these ten product shots and run them through my catalog workflow or whatever and give me the results, you know, and then you just get it. You know, so I I think like Rory Flynn (1:43:19) Correct. Mm-hmm.

Play from 1:43:24 8:88:88 1:43:24 Drew Brucker (1:43:24) that's the other piece of this. There are so many AI tools, dude. And I mean, like you and I are both popping around and the context switching is crazy. Right. Rory Flynn (1:43:34) Uh-huh. Drew Brucker (1:43:34) But like anytime you get a chance to sort of like rein in the context switching, take advantage of it. Like a tool like this, again, has all the tools already in there. So you don't have to do that. But then like this N MCP component, if you're already in an LLM, right? Just like those little things, because otherwise you're just gonna be.

Play from 1:43:54 8:88:88 1:43:54 Having 50 tabs open and popping between 30 of them and the other 20 are have been sitting there since January. And you you Rory Flynn (1:44:01) Yeah. Drew Brucker (1:44:01) swear you're gonna look at them. But I just like like the the context switching is like such a big thing. especially if you're doing your own thing, like if you got your own business or whatever, it'd be just because it's like you're trying to do like three or four things at once and then just hopping around in different places. But I think like that's that's a really cool feature. I think that just launched like last week, like Rory Flynn (1:44:22) Yeah.

Play from 1:44:23 8:88:88 1:44:23 Drew Brucker (1:44:23) Something like that. So it's like it's a brand new feature for them, which I I think is well needed. Rory Flynn (1:44:28) And you can only you can only access the tools that you've built, meaning If you're gonna go do this, I'm gonna try to find like this one, right? Tool when you when you have you can only access the tool when you have the output nodes, right? The output nodes means it's a workflow and it's done. Then we go to tool and you do create tool. This will publish it. Once it's published, then it'll show up in your in your Claude MCP and you can use it. You can't just can't just like Drew Brucker (1:44:52) Good call. Good call. Yeah.

1:44:28Publish your workflows into Claude MCP

Play from 1:44:55 8:88:88 1:44:55 Rory Flynn (1:44:54) be like go to this workflow and then like work in there. It has to be published currently in its current iteration. This will probably change. That's how it works right now. But just think about giving Claude, you know, Claude can now generate images for you. Right. Like you can be like, use this tool where it's just like you can publish a tool essentially where it's a prompt box and an image generator with like four, four different images. Right. You could and say, like, you know, go use this workflow and prompt for this. And bang, and now it's got an image tool or be for video.

Play from 1:45:22 8:88:88 1:45:22 Drew Brucker (1:45:19) Yeah. You got your own image tool right in there, yeah. Rory Flynn (1:45:23) Yeah. So like this one, Seedance, same thing. I built it so that this is can work with the Claude MCP. It'll go ping this. It'll fill in the necessary categories based on what I tell it. It'll populate with context that I have in video in a populate with context that I have in folders that I need. I'll tell it to I basically built the system prompt already off one of the Claude skills I had for writing. Seedance prompts and then it'll just, you know, aggregate the references and go run it. But now I just say that and it comes back in chat. I don't have to like go build this every time that I want to do it. It's just like go, you know, fix this. You know, I don't want that. Change the dialogue here, whatever. It's like a central con central hub and then like vector out to the workflow. You know, middle out. Middle out.

1:46:09Resources and wrap-up

Play from 1:46:10 8:88:88 1:46:10 Drew Brucker (1:46:09) Middle out. Man, we do we need a like we we need a Silicon Valley in the AI era. Like, dude, we I mean, that would be great. I was also just gonna point to this really quick. So Figma Weave has a really, really solid educational thing here on YouTube. Their channel is really good. like obviously we jammed as much as we could into this session, but you know, one on one, where to start, like. They go through very specific use cases as well, I would also just say, like, again, we'll we'll try to link a few, workflows for you guys in the description. You'll have, the discount link there that's there for you. And yeah, man, I do we we did a lot for I feel like we just sprinted for for two hours.

Play from 1:46:58 8:88:88 1:46:58 Rory Flynn (1:46:57) Blacked out. Might go to sleep. Drew Brucker (1:46:59) It's a lucid dream for you. Yeah, it's it's over. Rory Flynn (1:47:02) I couldn't sleep. Drew Brucker (1:47:04) as we as we wrap this up, who we who we calling out for subscribers this week? I gotta know. Rory Flynn (1:47:11) your your local node specialist. I don't that's that's not that's not a good one. Weavers. Tell Drew Brucker (1:47:14) Ooh, okay, okay. How about just like we weavers weavers in general, like basket weavers? Like Rory Flynn (1:47:21) basket weavers.

Play from 1:47:24 8:88:88 1:47:24 Drew Brucker (1:47:24) Just any variation of weave, like weevils? Yeah. Rory Flynn (1:47:25) That's a good one. so we can get into like crocheting. Your your local crocheting club. that's a good one. We haven't we definitely haven't tapped that audience before. If yeah, if anyone knows anyone. Drew Brucker (1:47:37) No, we haven't. No, that's a that's a different age demographic we haven't touched yet. I like that. I like the white space there. Yeah. Rory Flynn (1:47:43) Same mindset. Same mindset. Who they they'll be good at this stuff. maybe your local your local pasta roller for the you know, for the noodles, your hand picked noodle, you know, your one of those yeah yeah, noodle makers, your pasta makers.

Play from 1:47:59 8:88:88 1:47:59 Drew Brucker (1:47:54) Ooh. Yeah, you're noodle makers. Yeah. You're what what do they call the what do they call those people? Rory Flynn (1:48:00) I don't know, the guy the guys are just like, you know, you sit there Drew Brucker (1:48:02) Yes. Rory Flynn (1:48:02) with the the Chinese restaurants and just like they're hand pulling the noodles and they Drew Brucker (1:48:06) Yes. Rory Flynn (1:48:07) look awesome. But I appreciate everyone for for sticking with some of that technical talk coming from someone who hasn't slept in let's call it like eighty two hours something like that. But we I guess we got through it.

Play from 1:48:27 8:88:88 1:48:27 Drew Brucker (1:48:26) Pray for the pray for the pray for this man. Rory Flynn (1:48:29) Yeah. y you might get some hotter takes for me out of two. Drew Brucker (1:48:30) Dad of two. Dad of two now, man. Rory Flynn (1:48:34) I caught up. We're tied. Drew Brucker (1:48:36) We are tied. Yeah. If you if you have any Rory Flynn (1:48:37) We're tired. Drew Brucker (1:48:38) more, I'm gonna be taking lessons from you. Rory Flynn (1:48:43) But that was good. We we haven't done one in a while where we go deep into stuff. I know we've been going we've been going on the guest direction. Yeah.

Play from 1:48:53 8:88:88 1:48:53 Drew Brucker (1:48:51) It felt good. Yeah. Rory Flynn (1:48:52) If you guys if you guys have specific questions to drop it in there on the on you know what we are doing. Obviously everyone's using this stuff in a different way and we want to be, you know, probably do follow up episodes on it at some point. it's also good for us content wise if you follow us on any other account. Like if you're have specific questions. A lot of times that's where Drew and our, you know, my content comes from. So when people ask questions, just go do it. So if you're interested, that's how we'll, you know, that's how we can continue this conversation. But I would say drop something in the bucket. You gotta do one thing.

Play from 1:49:24 8:88:88 1:49:24 Drew Brucker (1:49:23) Drop something in the bucket. Like the the keep the comments and the questions coming. We like Rory Flynn (1:49:27) Yeah. Drew Brucker (1:49:28) we are taking whatever you put in there and and it's like we like we we earmarked this several weeks ago. We're just like, we need to do an episode on this. We had a run of guests, so we couldn't get it in, but now we've we've got this. We still, I think we're gonna do that episode too, where we go deeper into sort of like the Claude or the ChatGPT ecosystem of, you know, running a business, running like.

Play from 1:49:50 8:88:88 1:49:50 Optimizing your your ecosystem with you know skills and systems and workflows and like blah blah blah blah blah. So I like we've got we've got that. But like any other ideas that you have you want us to shortlist, put them in the chat, make a comment. We'll make sure to note them down. We'll we're gonna keep sprinkling in some guests here as well. So we we're hitting really like the the broader array of like what's possible and what people are doing because there's just a lot of people doing really cool shit in this space. But I got nothing else. I mean we we called out the weavers, so I'm I'm I'm set.

Play from 1:50:22 8:88:88 1:50:22 Rory Flynn (1:50:21) Weavers. It's good. Like I said, everyone keep keep doing what you're doing. You guys have been killing it in the comments, been killing it for suggestions. The guests, we've been getting some awesome feedback on that. I saw there was a comment last week. It was like, Does does burnout make you guys better interviewers? I don't know. Maybe, maybe it's nice to hear someone else doing stuff instead of my voice. Drew Brucker (1:50:40) Yeah. Yeah, man. I feel like, you know, we we we had guests for like several weeks in a row. So I think we were like starting to polish up there.

Play from 1:50:50 8:88:88 1:50:50 I'm excited to because I think we've like we've we've also got some other great guests coming up. So we're gonna we're gonna have this thing rolling for you guys. But Rory, I think maybe on that note, let's let's wrap it up, man. Rory Flynn (1:51:01) Wrap it up. Drew Brucker (1:51:02) We'll see you guys next week. Thanks for joining us. Rory Flynn (1:51:06) See ya.