Transcript of How I use ChatGPT to run my fashion business
How I AI
0:00Normally, I would hire a technical team,0:02like engineers. I didn't, right? I sort0:04of just came to Codex and [music] asked0:07it to build the website. The idea is to0:09use AI really anywhere in the process0:12where that makes sense. It's literally0:14my technical co-founder.0:15>> There's so much you couldn't do before,0:17practically, from a cost perspective,0:20even from an execution perspective, that0:22is now totally possible. And so, you can0:24like unlock your creativity in a way0:27that wasn't possible before.0:29>> does unlock so much creativity. A lot of0:31my designs start from a hand-drawn0:34sketch, but the other piece of this is I0:37sort of developed the fashion prompt0:40that's behind it. And it describes a lot0:42of sort of what goes into garments. So,0:44it has like the the silhouette,0:47uh the proportion or the volume of of0:50the dress, like kind of like the fabric,0:52how it flows, how it behaves, the0:53construction details, how it moves, even0:56how it sounds. The other piece that is0:58still unsolved, and I'm sort of still1:00banging my head against the wall, is to1:02get it to create patterns based on my1:04design. And there I'm kind of running1:06humans and AI in parallel. So, I I am1:08working with human pattern makers Codex1:10and just saying who's going to get to1:12making my things fastest.1:15>> [music]1:17>> Welcome to How I AI. I'm Claire Val,1:19product leader and AI obsessive here on1:22a mission to help you build better with1:23these new tools. Today I have Yana1:25Welander, and she is building an AI1:27native fashion startup. And she's going1:30to show us how AI can intersect art,1:32creativity, and real-world production,1:35and why computer use plus software means1:38SaaS really isn't dead, at least not1:41yet. Let's get to it. This episode is1:43brought to you by Merge. Building an AI1:46product is one thing.1:48>> [music]1:48>> The hard part is everything around it.1:50Connecting to the tools your team and1:52customers rely on, letting agents take1:55action with the right permissions and1:57keeping [music] everything reliable and1:59cost-efficient once you're in2:01production. Most teams end up piecing2:03that together themselves. So, instead of2:05building the product [music] you2:06actually care about, you get pulled into2:08integrations, permissions, routing, and2:11all the infrastructure underneath. Merge2:13is the infrastructure layer for2:15production AI. It connects to thousands2:18of tools, gives agents secure ways to2:21act inside them, and optimizes model2:23routing and spend without you building2:25or owning any of it. OpenAI, Dropbox,2:29and Ramp already use Merge to move fast2:31and build AI right. Visit2:33merge.dev/howiaai2:37to start building for free. Yana, thanks2:39for joining How AI. I'm an art and2:42fashion girl. You know this. There are2:44like five of us AI art and fashion girls2:47on Twitter. And we all find each other,2:49but I'm one and I love what you're2:51working on. So, immerse us in this new2:54thing that you're working on, and then2:56we'll talk about how you're using AI in2:58this really unique fashion.3:00>> Cool. Yeah, I'm so excited to be here.3:02Thanks for having me. I3:04am building this AI-native fashion3:07brand, and the idea is to use AI really3:11anywhere in the process where that makes3:13sense. So, obviously heavily in the3:17design and production process, but also3:19it's literally my technical co-founder,3:21and so I3:23that's that's what I use it for,3:25everything.3:26>> And can you show us like it's it's3:28probably hard for people to3:29conceptualize what like an AI-powered3:32fashion brand looks like? Can you just3:33like show us a little bit of kind of3:36like some of the stuff that you're3:37working on, and then we can get into how3:38it actually how you actually got there.3:40>> So, this is Yana banana, the the the3:43fashion brand, and you can kind of see3:45like every everything here is built with3:47AI. So, you have some of the different3:50garments that I designed, and I'll show3:53later that's like different ones starts3:55from different like from different3:57inspiration from different points. So3:59often time I start from a sketch. You4:02can see that when you hover over some of4:04these. So I start with with a hand drawn4:07humanly manually, you know, old school4:10hand drawn sketch and then I use AI to4:13turn that into different types of4:15fashion collateral and kind of technical4:17specs and everything along the line to4:19ultimately get to a garment and I kind4:22of use AI for all the things to4:23visualize it to create photos to create4:27videos of some of these garments. There4:29a lot of them kind of allude to4:32different like4:33uh4:34technical4:36uh you know, computer things in a lot of4:38ways because I feel like that should be4:40really part of the theme. It should be4:41obvious that that what it comes comes4:44from and so the fabrics4:46will use a lot of kind of the punch card4:48like computer punch card uh patterns and4:51and things like that so that you can4:52really see that on the other the other4:54way in which it sort of appears so you4:56can see4:57for example for this like uh post4:59keyboard t-shirt which I'm wearing5:01myself right now.5:03Um it also is uh alluding to the fact5:06that now we are not going to be using5:08keyboards and5:10uh in a lot of this stuff but can we now5:13have like give it a second life5:15essentially and to make sure that it's5:17like obvious the world we live in and5:19the world we're entering.5:20>> And I just want to pause because when I5:21first saw this I was like what a fun5:24like little art project. Like I once5:26made this like I vibe coded this app5:28which was um Anthropic as an5:30anthropology website cuz I thought it5:32was very funny.5:33But you're actually bringing these5:36designs to life. Like you are actually5:38wearing this keyboard shirt which is5:40super cool. And so5:42what you know, I hear a lot of people5:44who are in the creative fields be really5:46fearful of AI because I think it is like5:49displacement in a lot of ways. Like5:50let's let's all be honest here. There's5:53like videos and sketching and all sorts5:55of stuff that AI can do. And yet, if you5:59step on the other side of this like post6:01keyboard future, there's so much you6:04couldn't do before practically from a6:07cost perspective, even from an execution6:09perspective, that is now totally6:12possible. And so you can like unlock6:14your creativity in a way that wasn't6:17possible before. So, I want you to show6:19us step-by-step kind of what goes into6:23creating a garment like this, creating6:26and imagining a design like this, and6:27then actually getting it to production,6:29and getting a website like this up.6:31>> I think I think you're absolutely right6:32that it really does unlock so much6:34creativity. A lot of a lot of my designs6:36start from a hand-drawn sketch, but um6:41the other piece of this is I sort of6:43kind of developed this um6:46uh technical essentially stack, but I've6:49for the purpose of visualizing what it6:52looks like, I have this sort of set up6:54as just like the the the fashion prompt6:57that's behind it. And it describes a lot6:59of sort of what goes into a garment. So,7:01it has like the the silhouette,7:05uh the proportion or the volume of of7:07the dress, like kind of like the fabric,7:09how how it how it flows, how it behaves,7:12um the the construction details, how it7:15moves, um7:17even how it sounds. And I can I can show7:19in one of the pieces it kind of has this7:21really interesting kind of sound as7:22you're walking in in uh in the dress.7:26And I take all of this and put it7:29together in in essentially kind of when7:32we if we take it out of my my my flow,7:35uh you can see that it's at at the core7:37of it there's sort of this this prompt7:38that we could take um and plug into chat7:42GPT um, and often times in in when when7:46I'm working with this I will also have7:48an actual sketch of what I want and I7:49sort of am then defining this in7:52addition to the sketch. But just for the7:54sake of us kind of playing in real time,7:56um, I just thought it'd be cool to just,7:58um,7:59just use prompts to start with. So8:02So here we have a sharply tailored waist8:05defined jacket with exaggerated8:07shoulders and blah blah blah, right? So8:09now we're putting this in.8:11Oh, I should have told it to make an8:13image, but maybe it'll figure out to do8:14that. Yeah, there we go. Sometimes8:16sometimes it's smarter than you think,8:18even on instant. And so now it's going8:20to generate an image and often times8:23this will give me kind of like a product8:25photo, but usually what I then want to8:28do with this is to then take it to kind8:30of the next level. And I'll have very8:33specific types of like collateral that I8:35want to get out of it. So once we have8:37this ready, I will ask it to generate a8:40Vogue style runway photo. Oh, there we8:43go. So this is kind of like the product8:44photo of this. Um, so it has these8:46exaggerated shoulders,8:50uh, the waist, a lot of kind of the8:52stuff that I asked it to. And if I had a8:54specific sketch it would really adhere8:55to my sketch much more so than than what8:58was in the prompt. But then if we go to,9:02um, here we say, oh, like9:05now make it into a Vogue style9:09Can you tell that I normally do not type9:11anymore? And so I usually just use voice9:13for all these things. So it's just like9:15so9:16hard.9:16>> It's easier for our podcast guests, um,9:19when you use voice because they don't9:21have to imagine what you're typing. So9:23she's, you know, you're typing make it9:25into a Vogue style photo.9:27>> a photo model walking down the runway in9:32this outfit. And so then I kind of just9:35like iterate on this, and there's9:37um, I go to to photo that's a runway9:41photo, then I go to like9:44influencer photo, and you kind of see9:46this in a lot of from a lot of different9:48angles, and also iterate on it based on9:51kind of what you're seeing. They'll make9:52it, you know, red, make it shorter, make9:55it longer, and all those different9:56things, but I usually just start from9:58from a specific point and then iterate10:01from there.10:02>> I have a question for you, which is10:04why Chat GPT and why Image Gen 2 as the10:09as the model you're using? There's a lot10:11of Image Gen models out there.10:13Why that particular one?10:15>> Yeah, so I've done a lot of10:17experimentation with lots and lots of10:18different image models, and what I found10:21was ultimately10:23Nano Banana and Image 2.0, really Image10:27good at this.10:31Are the best at following visual10:34directions. And often I'm not doing10:36this, usually I'm sort of as I10:37mentioned, I'm mostly giving them a10:39physical like a sketch that I want them10:41to turn into a picture, and what Image10:461.5 did better than Nano Banana and10:49Image Images Chat GPT Images 2.0 does10:53even better is it really well follows10:56the sketch. In fact, it follows the10:58sketch sometimes a little bit too11:00precisely, and so you get something that11:02doesn't look like fabric anymore, and11:04you have to kind of be like, "No, no,11:05no, no, no. You got to like it's got to11:07flow." Like this is kind of why I have11:08this this very detailed kind of11:10prompting11:12where I tell it how the fabric should11:13flow because otherwise it can get into11:15the mindset of, "Oh, she really wants it11:17to look like paper." You know.11:19>> [laughter]11:19>> But11:21but the flip side though with all the11:22other image models that I've tried is11:24that they will make it look beautiful11:26and realistic, but it will look very,11:28very different from my design. And11:31obviously when you're designing fashion,11:33you you need it to look new and11:35different. You don't need it to look11:36like the most similar thing that's11:38walked a runway some other time. And11:40that's usually what you get with a lot11:42of other image models.11:43>> What I love and if you can go back to11:44your prompt flow, what I love about this11:47is this is just Okay, so people are11:49going to be like, "I'm not a fashion11:50designer. I don't have any use for11:51this."11:51What I think is the takeaway here11:54generally for folks is11:57write down your process and write down12:00what a definition of complete or good12:02is. You you and I like you we're product12:04ladies.12:04>> I've product people. Yeah, it's a spec.12:06>> We're product people. We're like12:07official product ladies. And you need a12:11spec. Like your prompt is a spec. And so12:14if you can put in to the effort of12:16saying like, "What makes a really good12:19garment? Like what makes a really good12:21photo? What makes a really good12:22illustration?" And come up with like12:24five things that you can define for the12:28prompt, you can get a really good output12:30of AI. And then what I love, my friend12:32says, "What's good for AI is good for12:34humans." Like let's just say you had you12:35were going to have a human sketch this12:37out or build this. This is the exact12:39information they would need as well to12:41do a good job. And so whether or not12:43you're doing like fashion illustrations12:45or12:46coding, right? Like this at the spec the12:49PRD matters, people. Like the spec12:51matters. And so getting forcing yourself12:54to sit down and be a little bit more12:55precise is going to get you that exact12:57outcome that you that you want.13:00>> And um you know, to be honest, I think13:02that jacket's pretty pretty rad. I would13:05I'd wear it.13:05>> This episode is brought to you by Jira13:08by Atlassian. The teamwork graph in Jira13:11delivers 44% more accurate agent results13:14with 48% less token usage. That's a huge13:18difference when working with AI coding13:20agents like Claude, Cursor, Codex, or13:22Copilot. The hardest part of shipping13:24with AI isn't the code, it's the13:26context. What's the right ticket? What13:29did the spec say? What got decided in13:31Slack? The Teamwork graph pulls all of13:33that from across your entire stack, from13:36Jira and Confluence to GitHub, and feeds13:38it directly to your agents before they13:40write a single line. You assign the13:43work, the agent [music] gets everything13:44it needs, and a PR surfaces when it's13:47ready. No digging, no context switching,13:50no broken flow.13:52Same team, smarter agents. Try and free13:55at jira.dev.13:58That's j i r a .d e v.14:03I love it. So, you've created this this14:05image. I I I also love this idea taking14:07like a core14:09image and iterating it through like14:11runway photo, influencer photo, like you14:15know, stand-alone catalog like photo14:18product product shot. What do you do14:20next? Like, what's the next part of your14:21flow?14:22>> So, a lot of what I do is14:24um kind of iterate on it. And so, I I14:29have this like, you know, change the14:30color, change the14:32you know, change the fabric. Um and so,14:34I I have I have kind of a lot of these14:36different ways to do that. But then, the14:38next piece really comes to kind of like,14:40what is the production process? And the14:42production process is going to be very14:44different depending on what I'm doing.14:47And so,14:48um14:49to give you an idea from one of these.14:52So, for example, I had this14:54uh14:55dress that actually did not come from a14:57sketch. It came from I had this one day15:00where I was just being15:02bombarded with Ruth Asawa's um15:05artwork in different places. Like, I15:07went to I went with my son to this like15:10like kids event, and there was an15:12exhibit across the street that I like15:14accident- like, we were just early, so15:16we went and watched it. And we're like,15:17"Oh, there's a flower model. There's15:18like a permanent exhibit there." Uh he15:21had a piano recital and and the church15:23had one of her pieces hanging like15:25literally almost above my head and I was15:27like, I keep I don't know. And so this15:30was this is literally from my photos. I15:31didn't have a sketch, right? Normally I15:33start with a sketch, but in this case I15:35was like15:36you know, make a photo of a model15:38walking down the runway in a dress15:40inspired by Ruth Ruth Asawa's loop wire15:43sculptures. And it was important to that15:46it had a beige undergarment because15:49otherwise ChatGPT will not like it's15:51like no, you're trying to make me make15:53nude pictures and I don't do that.15:56>> [laughter]15:56>> Very artistic nude picture.15:59>> No, I don't do that.16:01Um actually I think I think even here I16:03tried to make it to remove the16:04sculptures in the background and then it16:06was like, we're sorry, this violates our16:09nudity guidelines, right? This is like16:10happens all the time. It's very16:12[laughter] frustrating.16:13>> Quick side You know what? If I showed up16:16to a date with my husband in this gown,16:18I do not think he would be thrilled.16:19I don't think he'd be like, ooh.16:22It's not quite not quite the erotic16:23gown.16:24>> It's like it's not16:25Exactly, it's not it's not16:28it's not nudity. It's not, you know,16:30it's not porn. Um but16:33but here we are. And so so this one,16:35right? Like this is and I think that's16:38um16:39a lot of garments you would do like some16:41some garments I will drape, right? Like16:43on on on the on the model behind me or16:45some some garments16:47go into they become you create a kind of16:50a pattern and I have different some some16:53AI processes and some non-AI processes16:55that I've done for patterns. But this16:57garment is previously unmanufacturable.17:01And so what I and I and I ended up kind17:04of taking this and and also creating a17:07video of it. This is this is kind of17:09like you can you can see a little bit17:11more of of how what it looks like. But17:14the next step for this this particular17:17garment will be to take it and 3D17:21printed. And the particularly these kind17:24of big ball pieces will need to be17:27um 3D printed. And so I uh17:31then went on to Codex and had it help me17:34create CAD uh files17:39for these balls. I first prepped it17:43and had it like17:44I was like, "Okay, so I made made that17:46into a sketch." Then I made these pieces17:48into uh into kind of more sketch17:51illustrations. Then I17:55it didn't do well. It still didn't do17:57well. We had a bunch of back and forth.18:00Eventually it started doing something18:01that was a little bit more similar. Um I18:03had to do This is a really really long18:05process. And then ultimately I had it18:08using uh computer use um go and build18:11this in 3D software so that I could18:13create like an additional file and send18:16send off uh to 3D print. So that's the18:18process for this particular garment to18:20sort of to to 3D print it.18:22>> I have to again, like I have to pause18:24because18:26if if we're just talking about the18:27creative like generation process,18:31again, this is now a garment that would18:33have been previously like almost18:34inconceivable to make. Not just because18:38it was it's difficult to construct, but18:42even if you were like, "Of course we18:43could 3D we could 3D print that before18:45AI. We could 3D after AI." The like18:48tedium of creating the CAD models was so18:51onerous that like even getting to the18:53point where you could execute on it is18:56is really hard. And so I just think this18:58moment of like unlocking the previously19:01impossible, whether the impossible was19:03technically impossible or if it was just19:05like practically infeasible, is is super19:08important. The other thing um that I19:11love bless computer use and like Codex19:15plus computer use.19:17Love you so much because I can just be19:19like I don't need to know how to use19:21this software. Like you go use that19:23software. You do whatever you want. Um19:26I've gone through this like trough of19:28despair of like SAS is dead.19:31Maybe like we're never going to touch a19:32website again. Do you know what? Like19:35agents are really good at pressing19:36buttons. So like a software is back for19:38but agents are going to use it. It's19:39very very kind of like interesting shift19:42I'm seeing.19:43>> I think that that's a great point and19:45one one fascinating piece is that a lot19:47of these things like for actual end19:50product like a CAD file um Codex as19:54amazing as it is um is not great at19:57generating a final CAD file if you ask19:59it to do it on its own. But if you use20:02um yeah, if you use computer use20:05uh and have it go and do it in a piece20:07of software that's designed for that20:09purpose. I have both for like actually20:11for 3D printing and there's a there's a20:14fashion um20:16uh20:17software called Clo um that I've used as20:19well and it generates this like like it20:21generates patterns and it and it then20:23fits them on a 3D model and and I tell20:26Codex to go and use the software that I20:28haven't learned how to use myself. And20:30it does such a great job at something20:32that it itself couldn't do. So it's sort20:34of like SAS in combination with20:36>> Yep.20:37>> Codex works so well on a lot of these20:40things that20:41uh that don't and I'm sure it's just a20:43matter of time but that's that's where20:45we are today. And I want to do it today.20:46I don't want to wait. I don't want to20:47wait like a year. Yeah.20:49>> I I love it so much. Okay, so20:51you like generated an idea. We've20:53sketched it. We've created images. We've20:55even started to prototype the20:57construction of it.20:58How are you getting something built? And21:01maybe not the gown. Like maybe this like21:03keyboard shirt that you're wearing. Like21:05how are we actually21:07getting this manufactured?21:08>> Yeah. So, for something like that, um21:12what I've done, and I do this a lot, is21:15I will go and21:17now uh now this same tool is becomes my21:21research assistant, and I ask it to21:24identify like um you know, find 1021:27US-based custom apparel manufacturing21:30companies that are similar to the one21:31that I liked, but I want to see if21:33there's other ones. And then I just like21:35there's a bunch of things I want, and I21:37kick off a deep research uh extra high,21:42so that it's extra high21:43while doing it in that in that way. Um21:46and um and then and then I get back a21:49bunch of things, but usually so so21:52for for a lot of my work, I end up like21:54I'm doing something else. So, I'm like I21:56am kicking off this research, and then I21:59go and like drape fabric on on model, or22:02I do like I'm I'm on my sewing machine22:04doing prototypes. Like I'm doing22:06something else. And so, a lot of them22:08kind of the next steps end up being via22:11voice. And so, I like in this case I22:14came back to and I was like, "Well, do22:15you see the22:16um because I'm talking22:18uh apparel and using voice, right? Like22:20the apparel manufacturers that I asked22:22you to find, um now can you like put22:25together the email that I can see that22:27you can send off to them, but like you22:28may want to make sure that you get my my22:31approval to do that." And then I have22:33And then I use browser22:35uh used to let it go into superhuman and22:38like set all of these up for me, but I22:40really want to click the send button22:42before22:43>> [laughter]22:44>> Just to check its work.22:46>> Again, like something that is just so22:48tedious, and you know, if if I was like22:51you're you're you're much more diligent22:52at this than I am, but if I was like,22:53"Oh my gosh, I have this vision.22:56We're going to make We're going to make22:57these keyboard shirts. They're going to22:58be amazing." And then I'd be like, "How23:00can I make it?" And then I'd be like,23:01"Uh I have to like look up vendors. I23:04would just practically give up because23:06it feels so boring. But, if you can if23:09you can get that work off to someone23:10again like the next step and the next23:11step and the next step, and then what23:13you're going to end up with is this like23:15incredible business. And so, I just am23:18like so inspired by people like you that23:19take AI and create like23:23a a niche and like can can then take23:27that and build out a business where23:28maybe one wouldn't wouldn't have been23:30available or possible before. Um, I just23:33think it's is is super cool. Okay, so23:35you're doing this vendor outreach.23:37You're actually going to get this thing23:39this thing produced. Side note for all23:42the cool girls out there.23:44Um, when they're ready, we all want one.23:46Put me on that Put me on the pre-order23:48pre-order23:49>> There actually is. I'm going to I'm23:50going to I'm going to be that person,23:52but you can pre-order it.23:53>> [laughter]23:54>> Amazing. Perfect. Okay, and then last23:58kind of kind of flow you have is like24:00again going back to like pre-orders and24:02how to actually run this business.24:04How are you using it to like be your24:06technical co-founder in all this?24:08>> That's I I I got to say that that's24:09probably like my favorite piece because24:11this I mean, apart from the fact that it24:14is like I have so much AI in this, but24:16it's basically like an AI native like24:20Everlane or something, right? Like24:21normally I would hire like24:24a technical team like engineers. I would24:26have possibly have it I probably I24:28probably wouldn't Let's be real. I24:30wouldn't have a technical co-founder,24:31but like I could have had a technical24:33co-founder. I'm a solo solo founder24:35girl. Like that is just my my my nature.24:38>> It's the life.24:40>> It's the life. But, but I would hire24:42engineers. And um24:44And I I I didn't, right? Like I I I had24:47to build this website um and and I did24:51it and I sort of just came to Codex and24:54asked it to build the website. There's a24:56lot of pieces here that that aren't just24:58like a static website in that it like25:01you can you can obviously pre-order25:03things that I could like add Stripe to25:05it, which is like in prior lives I've25:08been like a tedious big exercise, but I25:10can go back and kind of show you the25:12Stripe flow, but adding that was just25:14like25:15you know, a second. Um I had to create25:18like databases to be able to like track25:21votes as people are voting on these25:23different garments. Um so I can decide25:25kind of figure out like which are the25:27most popular ones to bring to life. And25:29so like I I had to do that. Um25:32In in the earlier phase of this, I had25:34to create a separate dashboard for me to25:36track the votes in real time. Um and so25:39I did that. Now I really just like I25:41just use voice and I'm like, "Oh hey, um25:44how many votes do I have for this thing25:45now, right?" So like why would I even25:47have a dashboard?25:49>> [laughter]25:49>> Um25:51So I don't do that. But like just to25:53kind of show you like Stripe flow as as25:56as an example of of like, "Yeah, I I25:59came here and I was like, 'Oh yeah, use26:00browser to add Stripe to my site.'" It26:05was like, "Uh I don't know. You're like,26:07'I'm not signed in.'" I'm like, "No, you26:08got to sign in in my in my own in app26:10browser." I'm like, "No, I don't know.26:12You got go to mine. I'm already signed26:13up there." And then I was like, "Yeah,26:15you got to do it." I'm like, "No, you26:16can do it."26:17>> [laughter]26:18>> You know, I was just like not going to26:19go forward. It's like, "No, no, you can26:21do it." Um and and yeah, this is it. And26:24and then I had done.26:26Um26:27Everything's uh added in and now26:29pre-order's live. This is like a very26:32very fast process. I'm literally26:33whenever I like need to26:36change something about like the the26:40process, I can I can really just come26:42here and it's just like list out all the26:44things I want to I want to change and26:46and it goes off and it finds the result26:48project and26:50um it's it has like finds uh the26:53repository in GitHub and does all the26:54things. I would say like this weekend26:57my son built his own websites with using27:02like chat GPT sites and I'm sort of just27:05like sitting there going like do I27:06should I really like do I want to27:08migrate off of her cell? Do I really27:10want like you know, like if this is this27:12a whole new like now like now a27:149-year-old can have his technical27:16co-founder like do this and like build27:18build a company for him, right? Like27:20just it's it's just so cool to see this27:23all uh evolves, you know.27:25>> Your your 9-year-old and my 9-year-old27:27are going to be best buds. We can create27:29like a like a Codex club for fourth27:32graders. [laughter]27:33>> We should totally do that.27:35>> Um, this is a very sorry we are doing27:37San Francisco Silicon Valley mommy stuff27:40right now. I just want to I want to like27:42go back to one thing because ripped from27:44the headlines Codex prompting. I too am27:48like hey Codex do this thing and it's27:49like nah, no thanks. I can't. And I'm27:52like no, I'm pretty sure you can. And27:54then like after two like I'm pretty sure27:57you can. It's like you're right. I can.27:59I am I'm going to post about this later28:01but I have this like smart light bulb.28:03You can see it's on my on my thing. And28:04I was like hack it. And it was like no,28:07I really shouldn't. I was like you can28:08do it. And I was like done.28:09>> [laughter]28:12>> So just a little bit insistent.28:15You know, and and you can get get stuff28:17done.28:18This has been so so fun. Again, just28:21like recapping for folks. A business28:23that I think like would probably be28:24inconceivable before AI fully stacked28:27from the creative process creating28:28images creating videos imagining product28:32ideas all the way through like hardcore28:35manufacturing and sourcing vendors to28:39just like being able to voice note your28:41virtual CTO and update anything you need28:44to do and have your engineering team on28:46demand in chat GPT. Pretty incredible. I28:50want to get through some lightning round28:51questions and then we're going to get28:53get you out of here. You know, my first28:55question for you is like, what is the28:56hardest part of this? Like, what is28:57still Maybe like, what is still hard?29:00What isn't solved? What Where are you29:02still feeling friction?29:03>> So, I'd say it's kind of two things. One29:05is that like getting getting it to29:08really do getting image models to do29:11really29:12unique things per your vision while29:15still making it realistic. That it that29:17has been really really hard. I feel like29:18I've solved it. Like, I've kind of29:20cracked that nut, but it still is and29:22every now and again I'm sort of still29:24get something that is really29:26two-dimensional or whatever and I have29:28to really refine it. Um the other piece29:31that is still unsolved and I'm sort of29:32still banging my head head against the29:34wall is to get it to create patterns.29:37So, to get29:38uh get it to create like actually29:41accurate patterns based on my design.29:43And there I'm kind of running humans and29:46AI in parallel. So, I I am working with29:47human pattern makers and uh and Codex29:51and just saying, "Who's going to get to29:52like making my thing fastest like most29:56accurate that like to my vision, you29:58know?"29:58Um and so, that's uh that's been really30:00cool to see. I'm not I'm not there yet30:02with either actually with either uh30:04flow. So.30:04>> Okay. I love it. All right. And then my30:08last question that I ask everybody30:10speaking of like when it's hard and when30:11it fails you and not getting there quite30:13yet, which is how do you prompt AI when30:16it's not working? What's your Are you30:18nice Are you nice mom or a mean mom?30:21>> I can be both, but I I think I'm mostly30:23a nice mom.30:24Uh30:25and and the reason is that I find that I30:28get better outputs. So, like I mean, I30:31am very direct and I'm sort of like,30:33"Yes, you can."30:34Uh you know, you go do that. Like, I'm30:36not just like, "Oh, yeah, I'm going to30:37do this for you." Um but I'm I I be like30:41very polite. I will still say please and30:43thank you. Um and I do think that it's30:46because like30:48the the it's it's trained on body of30:51text. And so if it sees30:54for text output that is like rude and30:58mean, it will like kind of get in the31:00mind frame of that part of the internet,31:02right? Like versus like kind of more31:04professional business context and then31:06it will deliver something that's a31:07little bit more what I want it to be. So31:09that's But I I do it for selfish31:11reasons. It's not because I'm31:14afraid of the future AI overlords.31:16>> [laughter]31:17>> Perfect. Well, this has been super super31:20fun and inspirational and just like a31:22breath of fresh air in terms of a new31:24use case, a creative space that we31:26haven't seen before and something sort31:27of like manifesting the world. So where31:30can we find you and how can we be31:31helpful to you?31:32>> You can find me on X mostly. I am31:35YanaBana on X. I am now YanaBana at X.31:39Um I I changed my handle too much to31:41brand. Um31:43but uh but yeah, the in in terms of just31:45like helpfulness is getting feedback on31:49all the stuff. Like you can come you can31:50come to YanaBana and like vote on the31:53garments that you like um so I can get31:55feedback on like what's what's working31:58for people and what's not. Like I'm all32:00about user feedback and customer input.32:03Um and yeah, uh tell me tell me what you32:05like.32:06>> I love it. Well, I love32:09Poppy love.32:10Let's smash that like button. It is very32:12cool.32:13>> Awesome. Thank you. I will uh I'll let32:15you know when that one is ready to32:17>> available for purchase. Perfect. Well,32:19thank you so much for joining How I AI.32:22>> Thanks for having me.32:24>> Thanks so much for watching. If you32:26enjoyed the show, please like and32:28subscribe here on YouTube or even better32:30leave us a comment with your thoughts.32:32You can also find this podcast on Apple32:34Podcasts, Spotify, or your favorite32:36podcast app. Please consider leaving us32:39a rating and review which will help32:41others find the show. You can see all32:43our episodes and learn more about the32:45show at howiaipod.com.32:49See you next time.
6,171 words · 890 lines
