Ep 621: Anything is possible now – The AI Creative Stack

0:00Braydon:Before we jump into today's all-killer, no-filler episode, a quick word about who makes this show possible. The D2C podcast is brought to you by Pilot House, the performance agency behind some of the fastest-growing D2C brands in the world. Creative, media, and customer journey all under one roof. Performance and brand without the trade-off. Every Friday, we hand the mic to a Pilot House operator to break down what's actually working in their space right now. Want a team that treats your growth like their own? That's Pilot House. Head to pilothouse.co and now on with the show. Talk to me about Fable.

0:32Fable, it's better than Opus 4.8. It's extremely powerful. I watched one guy, he one-shotted a Minecraft game.

0:40Braydon:How has your Higgsfield work gone? Dude, that is an amazing platform. What Higgsfield does is it aggregates all those models into one place. If you do a deep dive on Andromeda, one thing that pops up is like the way that they rank creatives. Part of it is like how relevant it is to the audience and like how good it is at targeting. And the other part of it is how much it affects user experience on the platform. You're actually rewarded for increasing the user experience for the app.

1:15Braydon:It's all killer, no filler. Perfect time for a Friday afternoon session on AI with Brayden from Pilot House. Let's jump right into it, Brayden. Talk to me about Fable. What exactly is Fable? Is it just a model that I'm using or is it something that I got to click into? So I'm sure a lot of people have heard about Mythos, but basically Claude released this Mythos model that was super powerful that they were limiting who it would go to because it was like, you know, it was like hacking into systems and it was basically extremely powerful. And so Fable is a version of Mythos that just has a lot more safety guardrails.

1:52And so it's better than Opus 4.8. It's extremely powerful. some amazing like use cases and ways people have just been doing like single prompt, like one shotting, like amazing looking websites. And like I watched one guy, he, uh, he one shotted a Minecraft game. Like he's like, make me Minecraft is literally one sentence. He's like, make me Minecraft. And he's like, improve this prompt. That was like all he put. And it like took an hour and it came back and it was like a fully fledged playing Minecraft game.

2:22Braydon:That's wild. I remember it was, I think Eric Schmidt or one of the Google founders was on stage at maybe one or two years ago talking about, he's like, people do not understand what's coming. Like if, you know, TikTok were to be, this was when there were a lot of discourse about TikTok being banned. And it was like, if TikTok were to be banned, you could just create a new app with like an infinite amount of content on it that would just be, it would be ready to go. And like, it's hard, maybe it's hard to conceive of. I like that idea of being able to one-shot things because I'm not there yet with my prompts.

2:51Braydon:is quite a labor of sometimes diminishing returns as you prompt more and more. Yeah, it's true, especially with Claude. Claude, you know, once the context window like of your chat gets really big, it eats up a lot of tokens. And that's like one thing that I've found is like have to constantly like either create a new chat and I'll ask the old chat, I'll say like, summarize this, give me a markdown file summarizing this chat so I can move it to a new chat just to kind of like save tokens. Because if you're working in like one chat that, you know, say for a full day, only one chat, It just has to read through everything to answer the next prompt, the next prompt.

3:25And so it just eats up your tokens really quickly.

3:27Braydon:So have you been able to one-shot anything yet? You know what I did? A few things I've done, you know, based on our last conversation was really deep into the landing page stuff. And I gave it one prompt. I said, rewrite my landing page eight different times and create sub-agents. And each sub-agent is a council member and basically saying like one of them is an expert copywriter. One of them is like the CEO of the company. One of them is a, like a random like customer. One of them is a CRO expert and basically rewrite the eight and then have each member of the council basically rate which one is best.

4:02And then from there have a final decision maker who makes the final decision on the best output. And so that's like one thing I did and it was pretty cool. Like it wrote like eight different pages and it basically landed on like kind of like an advertorial style page and it had all the copy and stuff written and everything. And then I was able to take that as a wireframe to, to Claude design and then use my design system that I had built for this client and, and get it to create the page. So pretty amazing. Like, I think you could do that with Opus, but I think it's just like with Fable, it's just supposedly, and I mean, I maybe haven't experienced it as on another level, but like I've seen some amazing new cases.

4:40So I think it is on another level. I think it's one of those things, like maybe you just need to spend some time researching and playing with it in order to feel like you get everything out of it. Like if you're just asking, it's like, well, write me some copy. Like probably it's not a use case. But if you're like, hey, I'm doing a startup and I need you to like basically like develop this thing from scratch. Like maybe these like more advanced type of prompts is what it's good at. But it's insanely good at like web design. I mean, I recommend people to go on X and just like search Fable and just start seeing some of like the use cases that people are doing.

5:13Braydon:I'm just going on Claude right now. And it tells you Fable is the most capable model, draws down usage two times faster than Opus. And it's only included until June 22nd, which is interesting. So it seems they're going to add an upper tier for Fable's use beyond June 22nd, I assume. So yeah, what that means is right now, if you have like a max plan or even a pro plan, you can use Fable. So you only have till June 22nd to kind of freely use Fable within your subscription account. And then after June 22nd, it'll only be based on API usage. So people are having like a max plan and then they're just like using a bunch of Fable credits.

5:50Whereas like now it's going to be like kind of like a per token basis with the API. So that's, I think, the difference after June 22nd is it won't be included in your plan, but you'll be able to pay for it on a per token basis.

6:04Braydon:Oh, interesting. So if I just have the max plan and I want to use Fable, am I able to use the credits in my max plan or will I have to just spend on top of that to use Fable? From the best of my knowledge, what it means is like if you have a max plan, I don't even know if you'll have access to Fable. It'll be like if you want to use Fable, you have to pay for it on a per token basis. Per token basis. Okay. So it's like you give Claude your credit card and you like click Fable and it's not running through your like allotted subscription token amount. It's like just straight charging your credit card for usage.

6:40And that's to my best of my knowledge. I think that's how it's going to work.

6:43Braydon:something you mentioned last time and i've heard it come up on a number of the operator podcasts that i've done in the past couple weeks but this uh but higgs field is a platform which i haven't jumped on to to try yet i should give it a test because we're trying to come up with a bunch of creatives for for d2c how has your higgs field work gone dude that is an amazing platform so you can think of it as like a model aggregator so you know we were paying for 11 labs individually we were paying for vo3 or like we're using flow uh in vo3 you know um our video team was paying for cling what hicksville does is it aggregates all those models into one place so it's got all the best you know nano banana pro uh nano banana 2 it's got uh the new chat gpt image model it has all the c dance 2.0 video models it has you know the google video models it has everything there including 11 labs.

7:37And so it's really like a, it's an aggregator. And what I find is amazing. It's like so seamless to switch between like, you know, I create an image and then I want to use that image to prompt a video. It's like so easy. It's like one click, it jumps you over to the video section. Then you can like write your prompts. You can like split it up. If you want multi-scenes, you can then like submit, you can go 4k, you can go 1080p. You can change all the different size. If you want nine by 16, one by one, four by five, 16 by nine, you can do that. And so it's really, really amazing in that way. Like you can, you know, go in, prompt an image, take that image as your starting scene, put it into the video, then prompt a second image as your ending scene, you know, plug that into the video section and get like a really nice output.

8:21So I've done that for some clients now. Like I had one client who had kind of like a little like character that they had designed back in like 2018 and they were like can you please use this and I was like okay well it's like pretty lo-fi like you know it was obviously made not very well and and so then I was able to take that and then I was able to give it like a Pixar sheen so then I like turned into like a little Pixar kind of style character but then took that image and then you know give it to the video model and was able to create like multi-scenes of this like little character it's like it's just like I think explodes the creativity like whatever you can think of, you can kind of start doing it.

8:54So another example is on a call, they were like, we're talking about Hicksfield and he's like, oh, it's going to be any day now where you can like recreate our founder. And I kind of was just like, oh, I guess I haven't tried that yet. So I like got a picture of them. They have this other thing that's called a marketing studio. And I basically uploaded the product information and all the images for the product. I uploaded the creator, the avatar. So you can create new avatars. So if you want to like target like, you know, 45 year old women or like post-menopausal women, like you can create like infinite amount of these like avatars, these AI avatars, and then you can connect it to the product and then you can just prompt like, Hey, this is what I want.

9:35So I was able to do that with the founder. I was able to, you know, put in all that information and I prompted like a full, like 30 second, really good explainer of having him in his office, explaining it to the customers with like really great shots where it's like engaging where he's like close to the camera it's cutting to b-roll it's like um you know having slow zoom ins it's like so engaging it's like him putting on the mask and just like b-roll of him so it's really really cool and then from there i was able to then train it on 11 labs on his voice and so then i was able to then apply his voice to all of the videos so then i was able to get a final 30 second ad that is him it's super believable and it has his voice so i sent it to him and his mind was blown.

10:16But I was able to do that in like, basically the call was at two and ended at three. And by four 30, I had that video created. And so it was just like, whatever I, my, my imagination comes up with whatever it's doing, it's just so doable in Hicksfield. It's, it's crazy.

10:30Braydon:Are you running those ads in New York? No. So I'm sure you saw this and, and I'm not sure if anyone else has, but New York has basically created this law that says, if you're running synthetic AI ads. So synthetic characters or people, synthetic people in ads that you have to disclose it. And so there's like a thousand dollar fine if you are running like AI avatars and kind of like presenting them as real in New York. And then I think it goes up to 5 ,000 if you're like kind of like a continued offender. So, you know, we just got this news. It just came into effect on the 9th. And so we're doing audits on our accounts, just making sure we understand who is running AI avatars, these synthetic kind of people.

11:16And then just figuring out best practices in terms of how do we disclose that? You know, can it just be in the ad copy or does it need to have like a big blaring, you know, a thing on the ad? So yeah, that's definitely changed the game and probably shooken things up for a lot of people.

11:31Braydon:Be careful out there on your synthetic humans. I saw some interesting counter-programming this week from a bunch of DTC Twitter celebs It's just sort of talking about human creativity outperforming AI synthesis, but it's always going to be a spectrum. And especially, you know, you're always trying to tell all these different stories on your customers, you know, engagement with your product and such. There's going to be lots of room to test things that are both synthetic and authentic. Yeah, I think, and some of the underground places I go and learn about DTC stuff, they've kind of been mentioning like, you know, you don't necessarily have to try to pull the wool over someone's eyes and have like a fake AI person that's like super real.

12:14like you can lean into like an ad that's like claymation or an ad that's like in that pixar style and i've seen a lot of those ads kind of popping up on my feed lately where they're kind of leaning more into the fact that it is kind of fake and ai and it's not a real person but it's telling a real story and offering real value and i think that's kind of the the direction it's like not to be scared of ai synthetic ads as long as they're like offering value and not like scam Obviously, you just don't want it to come across as a scam. So if I see an ad that's like a synthetic AI avatar that's telling me about something that's actually relevant to me and that gets me interested in a product, I think it doesn't necessarily matter if it's human or AI, but it's probably good to try both things.

12:59And I think with just the amount of volume that everyone keeps requesting, clients and team members and meta, everyone's just saying, more creative, more creative, more creative. I think you kind of have to use all these tools at your disposal. 100%.

13:15Braydon:Got to get someone from Higgs field on the pod to, uh, to chat sometime because I've heard, yeah, a lot of people mentioning it. So the other thing we're doing, we're, we're working on, on some stuff right now about AI discoverability answer engine or SEO as Google just kind of silenced everyone a couple of weeks ago, just sort of saying AO, you know, GEO or whatever. It's, it's still SEO. It's still, it's still very, you're still writing for humans. But then there was another report that came out just this week that said that when you when you really look at it, listicles are still the number one source for how these answer engines are generating their answers.

13:46Braydon:Something like 58 % of a lot of their answers were sourced still from these cheesy listicles, which is funny because you wouldn't, you wouldn't think that's real, like because listicles can so easily be faked. You would think you would hope that they wouldn't play as much of a role in, in ranking here, but, but they are apparently. Crazy. I mean, I think that's another thing that people are just able to produce such high volume now is landing pages. But it's still, you really need those authentic mentions, those authentic social mentions. Here's something else. I just did a podcast with Ari Murray from Salt and Stone, CMO of Salt and Stone, which is one of my favorite brands.

14:21Braydon:She was talking about something that we're starting to see on our ad front. This is not necessarily AI related, but it's this idea of socially native ad creative. So ad creative that looks less like an ad and more like a social post. So like removing buttons from your ads, removing like very obvious calls to action, really making them seem like they're formatted for the medium more than an ad on the medium. Is that something you're seeing at all on the creative side of things? Yeah. If you do a deep dive on Andromeda, one thing that pops up is like the way that they rank creatives is part of it is like how relevant it is to your, to the, to the audience and like how good it is at targeting.

15:02And the other part of it is how much it affects user experience on the platform. So if you have an ad that like, kind of ruins people's experience as they're scrolling or as they're using Facebook, Andromeda and their AI algorithm, it doesn't like that. So it actually demotes you. And so there's actually like, you're actually rewarded for increasing the user experience for the app. So if you can have an ad that increases user experience, that that's going to get a positive vote as opposed to something that degrades it. And so one thing I have been hearing a lot, and I think that we are seeing on certain accounts at Pilot House is that social boosting, that posting a lot of stuff organically, and then finding what's working organically, and then putting a lot of spend behind that is a really good way of finding more scale, finding new ads, because it is exactly doing that.

15:55It's an organic piece of content. People get it. People don't need a CTA button to know that they can click on something and people kind of get, they've been around, they've been using Facebook for years now. We've been running tons of ads. People have seen ads like they get it. They don't need a CTA button in a red color to like grab someone's attention. So I think that when you, you know, it's advantageous for meta to have companies creating ads that feel organic and feel natural in the ecosystem and actually increase user experience. So I think that's like really probably what the person was talking about And what I'm seeing as well is when we get something that's organic and the organic person on the team says like, hey, like this thing's doing really well.

16:32We just love to hear that because it just means we can just put a bunch of spend behind it. And usually it does quite well. So nice.

16:38Braydon:Well, thanks for the quick check in today. We will stay in touch on all things AI. Thanks for coming on today. All right. Thanks, Eric. Maybe next time in office. We just had that great in office podcast with Duncan last week. So we'll have some more in office throwdowns coming up soon. Maybe even a DTC after hours. I'm always down. Okay, talk to you later.

17:24Thank you.