Transcript · May 17, 2026 · 15.2K words
Ep. 68: The Prompt Is Dead, Long Live the Reference + The Monet Trap
Captions uploaded with the episode. Expect the odd misheard word. Tap any timestamp to hear it on YouTube, or go back to the episode for the summary, receipts and chapters.
0:00:00Cold open
Play from 0:00:01 What's up everybody? Fast Hours Podcast episode 68. Rory Flynn, Drew Brucker back in the building. Two week hiatus. This one is your fault. I'm blaming on you. Completely your fault. Blame Rory. ah Not always, but. uh No man, I I uh this is I'm glad we're back though. I was really I'm really chomping at the bit for this one.
Play from 0:00:26 I think. you know, lots just changed over the last two weeks, obviously, we're gonna we've got a lot to talk about. I think we've we're chatting just before we hopped on here, about a number of different things that we can bring up and talk through that's gonna be helpful for everybody to hear. But let's just back up a little bit, man. How's everything been since we last talked? What's new? Good, [name]'s second birthday was this week. So I now have a two year old.
Play from 0:00:53 So that's gonna be, I don't know where the time went. Anyone who's been listening to us basically was here like right before the baby. So now she's two, which is, I don't know, just nuts, it's weird. that craziness where it's like, again, everything is moving so. Fast and I say it every podcast and I just open X and I turn you know I don't know if anyone else does this use open X make like five or six scrolls and just be like I can't do this right now I can't is a bad idea bad idea.
0:01:12AI news fatigue is real
Play from 0:01:25 It's like it's like you have to like post and go on X Yeah, I don't want to see it anymore. Okay. Now it's like you build something and it's like by the time you build something There's there's something that comes out. It's like that just made it ten times easier. It's like what am I even doing? Yeah, it's It's a huge problem right now. Very big problem. And it's kind of crazy with the, you know, I think everyone probably here, if you're not on the Claude code or the, you know, Codex train, like, it's like become a central hub for everything.
0:01:44Claude Code runs the day now
Play from 0:01:59 Like it manages my entire life and my day and my computer and my work and everything. And then when it goes down, it's like, it's like, so you ever go to, whenever you have the last like office job that you were in, There are days when the internet's down and everyone just looks at each other like, what do do now? I remember this used happen in New York every once in a while. The building would go out and you'd be like, everyone sit there like, internet's down.
Play from 0:02:24 And it's like, well, I can't open my email. I can't check my documents. What do I do? I don't remember too many of those days for me, unfortunately, but that sounds that sounds nice. It was just funny because it would just be like everyone kind of like, like nervous pacing around, you know, like trying to look like they were doing something, but everyone knows the Internet's out. Just like, uh, I think I could work for my phone.
Play from 0:02:50 remember like early in my career, you know, when you're like, just getting started and you're trying to figure out how work actually works. Just like purposely trying to make yourself look busy in certain situations. Yeah. Or just like act like you know what you're talking about when you have no idea and then try to Google it later. Oh yeah, totally. And then like go on Google with what the hell did they mean? Because you don't want to look like an idiot. Oh, yeah. Dad, I mean, I don't miss driving into the office.
0:03:11Remote work and coffee shop crimes
Play from 0:03:17 I don't. But I definitely do miss sort of like the camaraderie aspect and. You know, just kind of feeling, I don't know, like its own environment, you know, like just COVID, obviously, and the remoteness is has taken that away. And, you know, I think you and I spend a lot of time on our own, unfortunately. ah You know, just doing work heads down. get. I get caught up in this basement sometimes.
Play from 0:03:43 It's a nice office, but I don't have a window. So, you know, that's a problem. I got to go coffee shop to coffee shop. Even when I'm at the coffee shop, I don't, you know, usually strike up too many conversations. I'm just like focused on getting stuff done. And uh yeah, I mean, I'm not the most social, social person, but I do need more social interaction that I'm getting right now, for sure.
Play from 0:04:08 For sure. my problem was the opposite in the office. If someone got someone got chattering, I was like, I don't know how to get away. You know, like, don't know how to, I like can't get away. And then people would always come to my desk and like just start and I'm just like, Okay, okay, okay, okay, okay, okay. fix your eyes on your screen, hoping they would get the picture. oh something like that. Like I oh, sorry. I had a call in like two minutes You know like one of those things then the coffee shop was always like a I did that when I started freelancing like ten years ago and it was like the coolest thing I was like I'm working effing coffee shop and then I you know I'd go to the coffee shop and then you start to realize like all the little things about a coffee shop like especially in New York I don't know if it's for everyone but I'm like I gotta go to the bathroom because I'm drinking coffee.
Play from 0:04:52 I can't just leave my computer out I got to pack everything up go to the bathroom, come back, someone's probably in my seat. Like now I gotta go, yeah, like, I can't, it's like my whole life. was like, you know, it's like, I can't lose that. can't also, I also at that point in my life, couldn't afford to buy another one. So it was like, I can't lose this. can't have something happen this thing while I'm there. too is they'll have like some of them, it's like one bathroom, one or two, you know, and it's like a crowded place and it's like, you can't, I'm not gonna be standing in line for 10, 15 minutes.
Play from 0:05:26 So like, you know, I'll get up every once in while and check and you know, the other day I at this coffee shop that I love, but it's just like, they have one bathroom, you know, and it's like, this is a coffee shop, man, you know, like it's crazy. It's crazy. that eventually I'm just like, I'm just gonna wait till I get home to take a pee, you know, but. Want to know one of the most particular like like weird like particular that I feel like I'm Larry David right now like like curve my enthusiasm like I would I would Purposely switch coffee shops if their chairs were too low Because I don't like having like the desk like like a high table and trying to type like I hate that like I like if it wasn't the right like table to chair ratio where I'm like comfortable and you know, if you like if you like Slipping keys or like your hand you're always like readjusting your arms.
Play from 0:06:15 I don't want to do that. So here's It's like my one, there's like two or three things in my life that I'm particular about. It's definitely desk to chair ratio. Like higher, better. Lower, not good. the exact measurements of what you're looking for, like if you were to buy a new desk and the chair height and stuff like that, or what? that can go up and down, you know, like that's the that's the whole point is like some days I'm feeling maybe I want to sink in a little bit, but some days I just want to be up.
Play from 0:06:42 I don't know. I like to be like face in the computer, which is probably the most unhealthy thing you can do like for your eyes and your brain and the radiation and whatever. But I. I like I'll use two screens for this, but I'm not a two screen guy. I'd rather have one screen sitting right in front of me. That just tab, you know alt tab, alt tab kind of stuff and. I just feel like it's closer to me. It's easier to control. I can move faster.
Play from 0:07:10 Do want to my real desk? Right there. I lay down. That's how I work the best. lay and work. If I lay, I'm done. There's no way I'm on my stomach though. I'm like, puts me like right in the, everything you guys have seen has come from a track pad and me laying on a bed and everything. I only take calls from the desk when I have to sit upright. don't like, you know, these are weird things.
Play from 0:07:35 These are weird, weird things. You know, this is, this he's a creative, you know, like that's the best, I had this conversation with someone in the park the other day that was like the best thing that ever happened. to like creatives was like everyone started dressing like a creative. Cause then when you go places and it's like semi-formal attire, if you show up in like a hat and like a sweatshirt, everyone's just like, they're a creative, know, like you're like an ultimate pass.
Play from 0:08:02 oh, the, the creatives here, like, you know, don't try to control them. Yeah. There's something, there's something to be said about that where it's just like, well, I'll certainly take that pass whenever I can get it. Let's go. Let's go. but you know this is last little thing. I feel like I just got to tell you this. I don't know how relevant this will be to the rest of the audience.
0:08:213 Ninjas nostalgia break
Play from 0:08:27 I apologize. I showed the girls the movie Three Ninjas last week. Loved it. They loved it. And that was such a big part of my childhood. Me and my brother loved Three Ninjas. And then, you know, we watched, of course, Three Ninjas kick back. We had to watch second one. I don't don't acknowledge the third one because the third one's just absolutely terrible.
Play from 0:08:53 But that was a fun little last week experience. Got to show the girls three ninjas, then we did three ninjas kick back. And if you guys don't know what we're talking about here, I'm sorry. This is that 90s nostalgic uh era where, I mean, those movies don't exist, but it's just like, aw man, it such a key part of childhood. Me and my brother thought we knew karate after watching that stuff. That's the basketball scene.
Play from 0:09:19 One of my favorites. Yeah. Classic. I haven't watched it in years, but I yeah, there's we so little hack on YouTube didn't know they had a bunch of movies for free like kids movies, especially like the old ones like they have. We fired up the other day Fern Gully and Rock a Doodle. I know if you remember Rock a Doodle.
Play from 0:09:45 Great one. chicken it was like a chicken that was like animated but it was it was it was inspired by Elvis so it's like you know chicken like big rooster is Elvis oh it's a great one it's a great great throwback like 90s kid movie oh another good one yes there again why do we even why don't we even try just go back to the classics I mean it's so good.
0:10:09Freepik becomes Magnific
Play from 0:10:11 Alright guys, we're done. We promise. uh Let's talk. Let's talk what's going on man, cause like we obviously were out for about two weeks there. I mean there's I wouldn't say there's anything major major that's changed, but there has been a lot of different things that have changed. uh Maybe just kind of starting with something that feels like very straightforward. I mean we saw the the whole update with. Free pick free pick is gone.
Play from 0:10:37 FreePic is no longer there, it's now Magnific. Which has been an interesting kind of story, right? We talked about FreePic and their evolution, you know, from just being this icon and illustration shop to then, you know, going AI route. And then they expanded this offering into like this whole sort of like creative operating system with AI. And then they just up in one day changed the name, which I like Magnific name better.
Play from 0:11:03 And it sounds like that's why they did it. But I just didn't, I wasn't expecting to see that because just to take some of you back, I mean, they acquired Magnific. I don't know how that was a long time ago now. I two plus years ago, I had no idea that that would sort of take shape. I would have expected the Magnific website to fully dissolve by then, but it hadn't, it kept being Magnific and I always use Magnific in Magnific, not FreePic.
Play from 0:11:30 So just kind of a like really interesting. branding change. then you had I think I think there was also like they hit what was their ARR number? I think I saw saw their ARR number. Was 200 mil ARR. good. They quietly have one of the best like enterprise agreements from an indemnification standpoint for anyone that's a you know, looking for enterprise stuff.
0:11:47Why Magnific works for enterprise
Play from 0:11:56 It's like very, it's very business friendly, which is I think like a really good. It's a really good move because there's not many people that have the indemnification but they do so I mean, you'll corporate accounts, which is Yeah. have a enterprise client who used them, but they were very strict on which models they could use. They only could use, think it was, you know, NanoBanana Pro, not even two yet, you know, like things like that, Adobe Firefly, you know, is very constrained.
Play from 0:12:31 And then they onboarded here and they ended up getting all the clearance. for all the other models. I'm just like, why couldn't you all done that sooner? Because I was operating out of those constraints that you guys had for a long time. But that's a great point. The indemnification of what they offer is interesting. And I think they've got, I was trying to do some research. think they have like almost 40 image models on there now. And I think 36, 35 or 36 video models accessible now in one stack, which is crazy.
0:13:00Model aggregators vs internal tools
Play from 0:13:01 I mean, that's massive. Yeah. They're interesting because like, first off, I've had a few clients in the past that like tried to build their own like internal solutions where it's like they're pulled, they're trying to be like a model aggregator, but with only approved models. like, you know how much effort you're going through to build this thing and to like keep it up and service it? Just like go to free pick and like just choose the models that you want and you are indemnified. You don't have to worry about any of that stuff. And it's like, you're probably, you're probably doing spending just as much per API hit as you are while using the corporate account, right?
Play from 0:13:35 The corporate or the enterprise account, which is just like, this is boring stuff people, but these are the conversations that we get into where it's like, yeah. Yeah. the other constraint we had was like, with that client, if you wanted to get a new model approved, I mean, it just took weeks, maybe months, right? So it's like, okay, if know, banana two just dropped, how long before we can actually use that? And then by the time we get approval on that, is there a new model?
Play from 0:14:00 Right? So it was just always lagging behind. If you have something like this, you don't have to worry about that. And I think that is an interesting. footnote, I think for anybody to know if you're talking about potential clients or clients that you have that they do have some of these constraints. mean, these larger companies do have some very specific constraints that they have to operate in uh from a security standpoint, from a liability standpoint.
Play from 0:14:27 yeah, I think that's really interesting thing to mention too. Because FreePic to me is just like, it's very user friendly because it's not one thing or the other for business specifically. It's not like you have to use nodes or you have to use their own. You can sort of use whatever you're comfortable with and they have everything in there. Probably the fastest at getting stuff in there now at this point.
Play from 0:14:53 And they just keep making improvements. It's so big, it's so hard to even train anybody to use it unless you've been using it for a while. Well, I noticed that they are starting, they've got people that are now a team that's onboarding these clients now. mean, because when I first had them on there, they were asking me to do trainings for them. It's like, yeah, let's kind of break this into phases. There's a lot here, right?
Play from 0:15:19 ah And then eventually, I guess as part of these agreements now, they're starting to factor that in and doing it, which is great. I I think that part of it for us is, is hard because there are so many different models and platforms. you know, obviously I knew that platform quite well, but, ah you know, we can't be expected to know all these things because they're all changing. The relevancy of them is changing. The different things that they offer is changing.
Play from 0:15:44 So it does make sense for larger, you know, brands to be able to offer that. I know Runway, for example, does that, you know. Yeah, well, I mean, that's it. mean, like, look, the consumer game, the B2C game is can only take you so far. You got to be business friendly at this point. Like, that's who's going to use it the most. That's who's going to spend the most. And that's who's going to keep everyone else afloat. So it's, you know, I think it was interesting because it felt like it felt very funny.
Play from 0:16:10 Like, you know, I think you mentioned this before we were talking, like if you're using like seed, seed dance on there, wipe out like half your monthly credits for one video. and it's still being called like free pick. was like, let's not really have free anymore. So maybe that's why they, know, it's impressive though that you can at this stage can just like totally flip branding for, you know, name, identity, everything. And it's just like, eh, whatever, pick it up and go.
Play from 0:16:35 all that, I think like all that sometimes overvalued, like not, you're the branding guy though. I mean, like the, sometimes I think if you just make the change and everyone's still just like a sycophant, they'll still just be a sycophant. like it doesn't really matter, you know, it's not like that attached to the word or the logo free pick. uh Nike different story, but like, you know, free pick, we're going to change our entire name and appearance, but whatever.
Play from 0:17:03 Cool. We keep on moving. and I mean, most of what they did, I mean, it's like the site is exactly the same. It's just different colors and everything is laid out. You know, so, I mean, maybe that'll change right eventually, but like in terms of their actual launch and deployment of the brand revamp, it's just the name and the colors on the site, you know? ah I mean, they've internally, you know, got to change all those existing assets, but you know, there wasn't really like this big change.
Play from 0:17:30 uh And then to your point, I think maybe the name itself, like Pre-Pick, it doesn't really lend itself to what's actually in that product anymore or the way that they're pricing these things. definitely not free. But interesting nonetheless. you you got that. You know, one other just like little footnote too before we get into some of the larger topics is, I didn't want to bring this up because we did mention, well, no we didn't.
0:18:01Pinterest builds its own AI image model
Play from 0:18:01 You and I were talking about this before, but Pinterest has now developed their own internal AI image model, trained on their own training data, which I thought They have good stuff. interesting because uh to have that available as an internal tool, I don't think we've seen that yet, which is very, very interesting for maybe other brands that rely super heavily on, they've got all this training data available visually.
Play from 0:18:34 um That's kind an interesting thing. I'm curious to see how that takes place, but it's an in-house model trained on exclusively just their content. I don't know, it was sometime in Q1. I um don't think that's a tool that's available outside of Pinterest employees, but it's called Pinterest Canvas. um And yeah, I just think that's like one of those things we hadn't seen yet, which could be one of those things that we could see more often here in the future.
Play from 0:19:09 I'd be surprised. I wouldn't be surprised if you see it from Meta sooner or later. wouldn't be surprised if you see it from, you know, I mean, obviously TikTok did it. ByteDance did it with uh Cdance and like CapCut and everything that they're using all that data to generate the model. have the most, probably the most vertical format data in the world video wise. Google's, you know, take the same model.
Play from 0:19:35 Whoever has the data wins right now as you can. Pinch has got excellent stuff. Pinterest has like quality photography. You know, it doesn't have to sift through the trash, right? Like it's got really good stuff because people only put pretty things up on there really. um I still go all the time. Like I'm on Pinterest like. of the better algorithms too. mean, like anytime, like as soon as I get back on there, whatever's on that initial screen, my landing page screen without even a scroll, there's something there that catches my eye that I'm super interested in saving or, know.
Play from 0:20:07 like there's, I don't, personally I don't do too many pins. You know, don't do the pinning route, but I do have boards, right, that I save. Yeah, that's what it is. And yeah, like, you're right, though. I mean, there's some there's some really quality stuff there. uh Cosmos is similar has a different different type of feel to it. But yeah. But it's just, it's interesting.
Play from 0:20:33 Like I think a lot of these, they realize now that that's just, they're sitting on a gold mine, you know, that's, that's a gold mine. And, know, if they're using it for internal stuff, I don't know what they are. They trying to build more stuff for Pinterest, or are they trying to just like use it themselves and not have to worry about paying for anyone else's stuff, which I don't know. I mean, I don't know how that plays out in the long run. Always like, just like, is it better to develop internal now? for things like this, or is that, yeah.
Play from 0:20:59 I think that's the debate. I don't know the answer to that. When I first heard that, I thought, that's a great idea. But then I also thought about that, I think, in all the perspectives of is that actually helpful or not? if you're training, then you've got to continuously training. And you've got to continuously do these things.
Play from 0:21:25 Otherwise, you know, it's just a sunk cost. I mean, you would have to be all in on that. It's also curious to see like, you know, do you need additional headcount for that team? Right? I mean, I'm sure that's, know, Pinterest got a lot of money, but still, those are things that you have to weigh into consideration. I mean, it's not a cheap thing to do, even if you've got all that training data. So yeah, I don't know.
Play from 0:21:50 I guess TBD on whether that's a good thing or not, you know, because Again, like if you're a brand and you're trying to create on-brand imagery for these different things, at the end of the day, it all comes down to the context of your brand. So if you're able to communicate that with AI and have that sort of identity wrapped, and you don't necessarily need your own model, but... it's still an interesting case.
0:22:11Salesforce goes "headless"
Play from 0:22:16 there's, you know, probably transition into the, with that being said, like I've seen, seen some interesting decisions lately that kind of like caught my eye. Like the, whole, like Salesforce to me made a great, great move. This is just my opinion again, that they went, they're going more of like a headless system where they're realizing that, you know, having their whole sort of business model based on their you know, UX UI and like having people get on because Salesforce, I don't know if you ever use Salesforce.
Play from 0:22:43 I like my first company, like we got Salesforce and then because I was like the young kid and I don't know anything about dev at that point, they're like, oh, you can set it up and like, you can figure it out and like build it for us. I'm like, is so hard. Yeah. first of all, I haven't to answer quite, I haven't used it in a long time now, probably 10 plus years now. I always thought, you know, HubSpot was the better CRM sort of thing for marketers, you know, and it just kind of like it had more of that UX appeal and design appeal.
Play from 0:23:19 And it just like, yeah, it just felt like easier to navigate. Salesforce to me always felt like just they didn't care about that. It was just like jamming as many things as possible into the tool, into the platform. And I remember when I was at a company, yeah, about eight or nine, 10 years ago now, I mean, they were people, their whole job was knowing Salesforce as an expert internally to educate and get people integrated and adopted.
Play from 0:23:50 for the company. This was like a series D, C, or Z company, but they had those people. just to get you using it and knowing how to use it, because it's such a complicated thing, there's so many things about it that, I don't know, I just thought, I've never liked uh what they had to offer. uh You had to have like an internal person that they like placed in your team to just like operate Salesforce, you know, to build it to your specs to like maintenance it.
0:24:08Higgsfield, MCPs, and Meta ads
Play from 0:24:18 But that's what I'm saying, which is like seeing them go headless, which is interesting because it's like, we're recognizing that it doesn't really matter anymore. It's like, everything's going to be agentic. It's like, if it can just interact with the infrastructure, I can just tap into Claude and Claude goes in there and pulls the data that I need and brings it back and runs reports and does whatever I need to. Like it doesn't need to be this whole. UX UI experience, which you're starting to see it with the other companies now like Higgs field.
Play from 0:24:43 I was mentioned beforehand Higgs field for all of their shitty marketing in the past and whatever they're making smart moves now, which is like, you know, you've got to highlight when I think things are interesting and smart. They, know, this, this whole being able to tap into Higgs field from Claude with an MCP, right? Like that being the, not everyone having to worry about Building the UX UI so can see I mean it's gonna be if you think about how many models they have free pick has whatever At a certain point.
Play from 0:25:12 It's gonna be so overrun It'll be like going in Photoshop when everything has like 70 drop downs and like 70 options like I don't even know what to use so it's just like let's Just open up the operating system and the infrastructure in there and just have Claude sift through it all and like this is what I want to do You know having skills based in there and things like that. I think it's just smart. I think like everyone's starting to realize that Dllm and your little you know, your your agents are gonna be like The way you interact with things like that's that's who's gonna be touching stuff not like less manual I just think it's future future pacing which is interesting to me I think it's smart what they've done and now with supercomputer and some of the other things coming from like the So I've seen it like I was always wondering why no one was catering to the performance marketing world, right?
Play from 0:26:02 I think I've mentioned this on the podcast before Where it's like, performance marketers, meta advertisers, they create so much stuff, like so many assets, and most of the stuff doesn't work. So you got to create more, and you got to create over and over and over again in bulk, whether that's static or video, and you just have to keep going and keep going and keep going. Claude has an MCP for your meta ads manager.
Play from 0:26:28 Now Claude has an MCP for Higgs Field to build. like workflows and you know, like generate and batch and they also just open source this you know, virality analyzer for your stuff so you can see like where they're building towards and who they're catering to it's always been like social run some of this stuff, like worth showing? for the Higgs field MCP m now.
Play from 0:26:53 Higgs field stuff like or supercomputer, like do you have any examples? Because I have not run anything with Higgs field. I've been trying to keep up with what they've got going on. And just in general, My take on Higgs field is similar in the sense that it always feels like they do know what people like us want. It's just like their business practices and.
Play from 0:27:21 ethical approach and transparency and you know, like the things that give them credibility and build trust has been the loose end. But like the way that they sort of package things and you know, all that has always kind of been pretty solid, you know, they've been ahead of the game. Yeah. Well, they realized it was just like people don't want to engineer at scale They want to just tap into something that works, right?
Play from 0:27:46 It's like they build the little workflows. They built like the camera controls They built all these things from the start right that it was like I would just make this easier for you um You know here I didn't even see this number one affiliate marketing playbook in 2026 like this is like Prime performance. They know who's in here who's spending money like it's interesting like again, you have your Higgs field connector your meta connector. This is what the stuff this is just smart to me It's you know, you get your marketing skills that they've added wherever the virality thing is that we're seeing supercomputer now, which is like again, you can access Via your browser or remotely like you're talking to it almost like you're talking to you know open claw via this way, but it's It's just
Play from 0:28:34 smart. I think I think you're gonna see a lot more people doing this, you know What's it called? Kling was doing something different, you know, recently in a very similar fashion. This is just to me, business wise, like, I don't know how much this makes sense for the, for the personal user to, to be like spending all these money, money on API credits and hits and generating a bulk scale. like, that's what business does.
Play from 0:28:59 there are plenty of people that listen to this now that I think have clients in this space too. And you guys know this, mean, the way things are changing, we have to change with it. it's just, we like, well, you kind of mentioned it before, but we find a process or workflow we like works really well for a couple of months, then all of sudden it's like, I could still do it that way and probably be okay for a little while. But it's like, if you're not, at least, playing around with these other potential, how do you really know it's the best way to do it?
Play from 0:29:30 We have to experiment, we have to kind of keep up with this stuff. And I think this is one that I think personally, I need to probably try to find some time, however I'm gonna do that, to just kind of play with it and see what it would potentially mean for me and where it fits. ah But I do think the Higgs-Feels thing is interesting. So that, like the bulk conversation, this probably, this is something that I wanted to, have you used Codex for image generation yet?
0:29:51Codex for GPT-Image-2 workflows
Play from 0:30:00 Like for ChatGPT image? Inside of Codex? No. Oh, it is a it is a very interesting experience in there. And it's it's cool because now Chadgy BT image is good again to utilize, but also so I was the other day, let see if I can actually find this here. Do I have it? I have so much stuff in here.
Play from 0:30:28 It's unbelievable. uh Open right now because I'm putting together presentation. just released Codex for, it's on now mobile too, so you can actually do the code stuff from your. Yeah, it is. They have computer control, which is, you know, you can do this stuff like the mobile experience now to me is so much like I opened it up the other day and was doing some stuff like it's so much easier than Claude makes it from the remote access.
Play from 0:30:53 Yeah. mode on ChatGBTs level too, because you can do the voice mode with Claude, but it's not nearly as long. You only get a certain amount of time. It's very short. ChatGBT, I you can go for minutes. ah And so if I'm in Claude, I always use the voice with Claude there. And then I'll... I'll get capped and then I have a tab open for ChatGBT and then I'll just like continue it there and then copy and paste the transcription of what I said, audio back over to Claude.
Play from 0:31:25 But Claude has fixed that. And you may be saying Whisper Flow, I've used Whisper Flow, I like Whisper Flow, I just don't wanna pay for that sort of capability. Again, I think that's like one of those things where I've seen people vibe code, like free replacements for Whisper Flow too. But ChatGBT is kind of like. ahead there or open AIs ahead there with the voice stuff. Yeah, this there's like, that's what I'm saying.
Play from 0:31:51 Like this, there's like interesting stuff with Codex. Codex to me is really fast. I like it a lot. I use it for certain things. I think it over engineers a lot of stuff, but Claude kind of like if you use them in tandem together, they they sort of like balance each other out. Like Claude Claude always brings like the skeptical being like, yeah, that's, know, the same way Chow LGBT is over ambitious about certain things and Claude's like the realist. um But like this was interesting. The other day I was trying to...
Play from 0:32:19 I saw this video on Instagram that had a really cool color grade. was a very flashy... had a very good... edge-halation and stuff. And it was a great color scheme and I was like, just want to replicate this. had it... A codex you can do just about anything. So I had to download the video, pull the frames, analyze the frames, and then create a style around it.
0:32:45Pulling style from video frames
Play from 0:32:45 Like I want to just like create basically like a standalone style so I can just replicate it across. And then I go, can you just create 20 images of like, you know, like successive images of like this dog, just like a golden retriever in a scene with his with his owner who's fishing and like that was it. I just said, create 20 images. It was like because it's working locally and there's not like the ChatGPT limits. Like it just does 20 and it did it really fast. And, you know, I felt like some of these were super.
Play from 0:33:12 I mean, it keeps it all. You see it from. zoomed out like very consistent from a color scheme perspective like color grade perspective. ah I thought that was was awesome. You know, just the way that it was able to sort of pull that all together. I love this image. It's perfect. um That one, you know, we're not doing like mind bending stuff here, but the fact I can do 20 at once, I could ask for 40. But the cool thing that came out of it was I went to I asked because you see everyone's doing like these these style guide.
Play from 0:33:43 things are like, know, the stuff with Sea Dance where it's the animated sort of storyboards, right? So was like, what happens if instead of me, like, you know, can we create like style, like style reference from Midjourney, but in like a reference sheet? Like what exactly we want to do here? So I took those images, right? I just said like. Can we create like if you were like asking chat TV, if you were receiving one image and it had to have all of the things that you needed on it to just like replicate a style that would be transferred and all I have to do is say like, you know, dog and human like in the back of a truck, dog and human like camping, right?
0:34:05Building visual style reference sheets
Play from 0:34:23 And it keeps everything the same, the same way Midjourney style reference works. So it came up with this like document, which, you know, there's some like errors here. but it has the style locks. gives me... I love that it has the exposure curve here, like midtones, shadows, like core tokens, things like that. ah You know, source, like a graphic image as, know, where these sort of like, these were pulled from the video itself.
Play from 0:34:49 ah You know, camera, sort of camera logic, color grade rules, and then prompt nucleus, which I was like, this is cool. Like, it's just like a style card. And then just applied it. And I was like, just like... I just said here, generate six successive images for a scene in brand content, outdoor lifestyle, fly fishing. Like that's it. And then bang, like using that same sheet, right?
Play from 0:35:15 Like we're keeping everything consistent. Like I don't have to prompt for that anymore. So I thought that was, this is with Nano. This is with GBT. guy in the character consistency? okay, you've got a separate, okay, I got you, I got you, got you. But, you know, this one is the only one that messed up his the color of his jacket down here. Actually, this one messed up the color. It's probably in the prompt. um But just again, like from the looking at the.
Play from 0:35:44 also do the character sheet in GPT Image 2 as well? Or I guess in Codex 2? Yeah, that was a Codex one, because Codex, these character sheets are great because you can get like, you know, the full view, you get the the face, you know, the facial reaction shots and then like the action shots, which is which is good for this stuff. Yeah, so I was like, this is this thing is super interesting because I used it with some I used it with some other stuff.
Play from 0:36:10 I use it in a more of like a really hardcore illustration style in in free pick. And it worked really well. Let me see if I have, that's my 2D vector projects. god, spaces. Oh man, we can come back to this. This is a, because I using, where is this?
Play from 0:36:36 Let me see if I have. So I was doing some stuff with like a really intricate style and it's hard to like describe it. Like I don't know how to describe it, but of course I don't make anything easy for myself. um Let's see if I have this here. No, it's going to be in free pick, but there's here. Let's open this up.
Play from 0:37:02 No page one. No, this is not, this is just dead air. Sorry everyone. While this stuff loads, should have had this prepped, but I... Didn't think I was gonna remember it. Now I do. where is this little sheet? This thing right here. This is super cool. I've been using it for these illustrations. I have this geometric pattern that I wanted to use and also make it look like ruffled tissue paper with a very specific texture and color palette.
0:37:36Codex and textured illustration systems
Play from 0:37:36 But I asked again for these sort of references to give it sort of a muted color scheme, how everything should look. And then this is the cool stuff. It's how good ChatGPT is on some of these images, how detailed and textured it is. If I was to open this. uh If you go all the way in, look at all those micro textures on the...
Play from 0:38:02 It's getting that sort of tissue paper look, but seams, and it's very, to me, very impressive that I can do that. Because photorealism, yeah, it's good, but some of the stuff is not great. If I probably look better on Figma if I show this. But you go all the way in, the depth and the detail on all of this, each little ridge, each little texture. This is crazy.
Play from 0:38:28 So that's what I was saying, like, and I've been using it to sort of create, keeping this on the download, because this is a surprise, but I'm showing it on the podcast because, you why not? uh But that's just another thing to take into account is that we can, you know, if you're going to use, we don't have to use text as much anymore. Like this is a good reference now for generating whatever I want from it. This is also a good reference. So it's getting more intuitive to the point where you could build like a style guide in a reference image.
Play from 0:38:58 I guess is what I'm trying to get at here. And that's basically like a mid-journey mood board in an image. Like it doesn't have to be in code. doesn't have to be in like 86 layers of JSON. Like it can be within an image and it can be, that can be a little bit more dynamic instead of having to text at it all the time. Right? Like I think that's might be a little hack for everyone to try there. I'll also try to like build something out for like a prompt that'll just build that.
Play from 0:39:26 I was working with Codex to be like, what do I build? How do I build this? What's good for you? Like asking a million different things. I'll ask it for a prompt to do that so we can all benefit from that. yeah, I mean, that's true. mean, yeah, it's. Man, we've really come a long way with prompting, haven't we? I mean, the early days of mid-journey, you're just typing in all these keywords that sound cool and interesting and see how they play together inside of Discord.
0:39:34The evolution beyond text prompts
Play from 0:39:57 you know, then you're thinking about structured prompts and the coherence is good where you're thinking about, you know, all the elements of photography and having to describe those things and you know, obviously like the effort it takes the effort and energy it takes to just Yeah, come up with those things and then how do you you know package that up and sort of like Speed up the process and become more efficient, you know, you got prompt structures and you know templates that use and this and that you know, it's just like this the evolution of this has just been really interesting to see and you know, just typing in an empty box was never, it was just never the easiest way.
Play from 0:40:40 And it was cool from like an exploratory standpoint, but we quickly outgrew that. it's just, yeah, it's continuing to be really fascinating to see how it evolves over time. It's all context, right? I mean, it's all like what you provide. Like that was, if I had given it one image and said like, oh, build a style reference off of this, like it probably wouldn't be as good as if it had like the full, you know, video file and extracted, you know, 50 frames from it to make it more consistent.
Play from 0:41:10 So it's just something that I find, like I've been trying to, I've been always been kind of trying to recreate. This is also like a prop to Midjourney, like how hard it is to recreate style reference. and like make it work like it does. It's not easy. It's not as easy as attaching an image and making things look exactly the same as that one or mood boards, same thing. Like coming up with that, that's, I've been trying to do that for these other tools and they do it, but you see the like the lengths that you have to go to to get there.
Play from 0:41:38 If it was right, like that would be a lot easier if I didn't have to do that, but I'm hacking around the system to try and figure it out. But I do like this more, I do like this more visual approach. that we're coming to, like the same stuff with C-dance. You know, like, I don't know if anyone, if we did this last time, but I'll just put this up because I feel like it's, it's, it's worth, you know, worth everybody seeing if they haven't seen it.
0:41:50Seedance storyboards and visual prompts
Play from 0:42:05 Um, like these little things with C-dance, like being able to now give visual reference and like choreography on a certain like movement pattern or camera direction or whatever. And listening to that, because ChatGPT can create these images, now things can become a little bit more interesting. I think this now opens up a lot of stuff for us.
Play from 0:42:33 and going back to this thread, this point that we're making too about text prompt, we're getting away from text. If you see an example like this, this is Rory's example, right? What if I wanted to create something similar? Do I need you to give me the prompt? No. No, because it's it's been done and it's got a reference, so you could take that into any LLM and say, hey, look, I want to do this same thing, but for this, right?
0:43:01Reference images as reusable seeds
Play from 0:43:01 So let's do that for this and you know, use think of this as a template that we're going to use over and over and over again, right? And so it's just like that mindset of it's almost like once it's been created. Once. That seed is locked and you can do whatever the hell you want with it right that that even goes for iterations on it. It's not like I have to treat this exactly the same. I could.
Play from 0:43:27 take things within that and say, I'm also going to add this, this and this, or I'm going to subtract this, this and this. the point is that visual aid is the prompt. It's the strategy to create the prompt. It's not me sitting and saying, hey, Rory, what's the prompt, man? Can I get the prompt? We're trying to take a guess at it. This is how you do it. That's the way I always look at these things too, is when I see something new and interesting, I'll bookmark it and say, OK, if I ever do anything like that, now I've got a reference that I can work from.
Play from 0:44:02 I'm not starting from zero, ever, ever. mean, unless I'm just purely just creating to create. When you get in these situations where you're doing client work, you're trying to be efficient in anything that you're doing, I you're not starting from zero. Typically, you're trying to rely on something you've either seen that you can build on or something you've done before that you can build on and adapt. So it's just a mindset. I feel like that's worth reiterating in case anybody's still thinking about this differently.
Play from 0:44:29 But yeah, I mean, the whole storyboard to video stuff is fascinating. And we're getting there, man. mean, we're definitely getting there. see the seed dance videos that are 15 seconds long, which is, again, like the longest these things have been for as good as they are. Even in that 15 seconds, I do typically catch one or two things. At least that still aren't completely there.
0:44:51AI video still has an audio problem
Play from 0:44:56 But this video thing is is a hard thing to crack. I mean, it really is, but it's getting there. Also, one other just let me get on this pedestal for one second. The downfall of these video stuff still is still the audio. Still the audio, no one, no, no one has got no one's gotten the audio right yet. It's the same voices, it's the same giveaway.
Play from 0:45:22 Usually the audio level is too high immediately. It doesn't feel like it's taking the space into consideration. So if this is in a room, what would her voice sound like in that room? Do you want to see some examples of how to counteract that? a studio. mean, maybe you could prompt and text it that way, but I'm just saying that if audio is gonna be a part of, I mean, the video part, but also audio is just as important here.
Play from 0:45:53 And I'm tired of the same voices, the volume's too loud, it doesn't sound like it's the right, the volume's in the right space, acoustically. And I'm not even in like an audio snob. just making an obvious point that's been a huge like issue in my mind. So let me sort of show this to see if we can actually, if I have the...
0:46:13Audio reference hacks in Dreamina
Play from 0:46:19 So this was a little hack that I did. Let me see if I can find this. I'm in Dreamina here with a... All right, so this is... If you're working in C-dance or you're working in... This is Dreamina's platform, but if you're working in any of the node-based systems or anything that has Omni-reference... from Seedance, the best hack I've found is to upload an audio clip of whatever you want the voice to sound like.
Play from 0:46:46 And it can really do a good number here on mimicking the voice and the spatial depth and everything along those lines. Like, let me see if this one's here. Can you hear that or no? So I have to hold on. Let me stop sharing. We gotta figure that out though. We like to show videos and we're not getting the audio coming through and I'm just thinking like, I haven't.
Play from 0:47:14 like the window itself. Maybe that, no. Window. Also share audio. There we go. Allow. God, it's not gonna let me do it. Huh, all right. Let's, this one, is this it? Okay. Now you tell me if you can hear this, because I put, you know, this has like been a little bit of a hack. Instead of doing like uh 11 Labs voice in Dreamina. I've been going and you can do this with any Omni reference.
Play from 0:47:40 I've done this in Weeby. I've done this in nodes or not nodes, spaces is you can add just like an audio clip of your voice and it does account for spatial ah depth and things like that. It's actually really smart in that way, but. Like, so this is, could you hear that or no?
Play from 0:48:07 Okay. It's too loud though. We'll have it at max volume here. But I mean, like, for the video, like, I don't know, there's something, there's just still something that isn't right there. Let's see, let's pull this one up on... I have some errors and I was working on this. I'm talking about the bigger problem in general. Let's pull this up and just see. I'm just curious of what you think about this because I think it sounds like me in some some sections, but doesn't sound like me in other sections.
Play from 0:48:39 Like I was trying to do a dialogue scene just to see how it would. Do you remember the old do you remember the old video of like the YouTube video of the surfer where the guys go and like, you know, so pitted. All right. So I used I use that guy as an image reference and then used his voice reference for this dude in here. It's just to be funny. But let's let's see. very Rory Flynn thing. I like that. I don't know why came to mind, but I was like, you know, I might have seen it on like Instagram or something.
Play from 0:49:06 I'm like, let's let's use this guy's voice and see what it's like. mean, I don't even know if he's alive anymore, but let's say so. Yeah, again, open to any and all criticism. don't I was just making this as to like to see how it like see dance could handle like a conversation. I'm not a cinematographer or know how to do this stuff, but this is what it sounded like. So let's say.
Play from 0:50:21 So you can hear like the spatial depth is wrong in a lot of that stuff, right? Like it's interesting that it's still able to, it's capable of like, here's a seven second audio file of me saying whatever the hell I'm saying and a seven second audio file of that guy saying, so pitted. And like it somewhat keeps it consistent. But like, again, I'm doing this manually trying to figure this stuff out. Like it's not necessarily like production grade. Like this is just me like playing around in the background.
Play from 0:50:48 Yeah. like that part needs, I would love for that part to catch up. um just crazy. Like the seed dance stuff with the, with the Omni reference, if anyone hasn't played with it, it's really awesome. Like that you can just, instead of like having to create the perfect reference image every single time, start frame, end frame, like you can create the character sheet as the reference. You can drop in the voice as the, know, as another ingredient, you can drop in the environment as another ingredient, the secondary character, like all these different things that you can use as references and it adheres to them pretty well.
0:50:52Omni-reference for video control
Play from 0:51:20 It's just like that's. That to me is such a better experience than like the first frame last frame thing. yeah, too many things can get. Yeah, if you want someone to walk into a scene, right? Like, I want to start with like, I'm panning through a location and then then I walk in, right? Like you need that Omni reference feature to be like, okay, here's the person that comes in instead of it just like starting on. you know those prompts can get very long too, right?
Play from 0:51:48 after a while, so what, you know, is there a character limit? What's the coherence? You know, and with the Omni reference stuff, you're saving characters, right? Because you don't have to describe, I mean, that's one of the benefits, obviously, not all the benefit, but one of the benefits is you're saving characters on having to go into descriptive detail about it, because it's already got the information, you know? um just, it's just, I feel like the overall example or the point that I think I'm trying to make is we're moving into a more like visual space of like things are not as much like write a giant script and then like hopefully it follows it.
Play from 0:52:24 It's like use this exact thing, use this exact thing, use this exact thing, which makes it a lot easier when you can start to create good images, right? And know how to create images or at least ask G chat, GBT or Google how to make a good image. They'll go do it now. Okay, what are you? Yes, and I agree. I agree with all that. What do you um let's let's switch the Midjourney for a second. Have you been using Midjourney lately?
0:52:37Midjourney V8.1 updated take
Play from 0:52:50 ah What's your updated take on 8.1? I know I've ran quite a few images lately. So I've gotten more reps in with this new coherence interpretation of things and blah, blah, and just getting more of an understanding as to. you know the shots that I'm getting and what I would expect.
Play from 0:53:17 I've been seeing some really good things in general from from my personalization. That said, ah in some I have started to see the old blue and pink you starting to kind of make its way in sneaking in the back door and ruining all my photos again. um It's not as ridiculous as it was.
0:53:28The blue and pink problem returns
Play from 0:53:44 But it's to the point where I'm going back in and the personalized and ranking more images that have no blue and pink cues to try to offset it again. You know, like. I don't know. I don't know. It is so annoying. EXP was was definitely part of the problem. Yeah. But let me see. was noticing that I was definitely getting more blue and pink hues when I would go higher on the exp scale, but I haven't I haven't seen that in a while because I haven't been going super high levels.
Play from 0:54:17 It's mostly low levels now that I use show it show me show me some though I want to see it. I want to feel nostalgic. This isn't like the most perfect example, but this does illustrate the point. uh So I'm just going to show you this on the screen. This is Ferrari, which has a very recognizable red hue.
Play from 0:54:48 And look at the hues on the red. muted. Vivid. This is blush. This is like, and look, you can see the cool lighting on all this. So like I saw this was definitely one of the culprits where I was just like, what the hell is this? You know what I mean? Because it's like, OK, well, screw that.
Play from 0:55:15 not even going to mess with that one. Let me see if I've got some. though but black and white it's so good that's just like But like even these, you can see all the blue and pink and blue and, you know, blue and pink and... Even the flowers. it's just, it's just, you know, it's annoying.
0:55:37Midjourney still owns realistic faces
Play from 0:55:41 It's annoying. So that's my one gripe. That said, ah I mean, the crispness and the reality, like the realistic look of these coupled with the creative interpretation that Mid-Journey is famous for. Best image model still, best image model still. mean, that being said, it depends on the use case again, where I'm not weighing this in terms of the edit models like Nano and GPT.
Play from 0:56:13 That's not a fair comparison because those are different things. But in terms of just straight up image gen, mean, bro, what are we talking about? Let me see if I can like just jump into, I mean, You know, the detail on these is just ridiculous. mean, just absolutely ridiculous. was just doing, I was messing around with some little Wayne stuff.
Play from 0:56:39 uh Introduce [name] to Lil Wayne last week. yeah, got a big thumbs up of approval on that one. Yeah. Now just like I could ask for Lil Wayne now when we're in front of the TV. Lil Wayne? Okay. That might make an appearance in the thumbnail.
Play from 0:57:05 Might have to. Look at that. I mean, dude, there's nothing better in the game for shots of people. You can't convince me otherwise. If you think otherwise, I don't know. Maybe this is an open challenge. Come at me, bro. I mean, I'll go head to head with you on creating real imagery of faces. And this is just unbeatable at this point in time. I mean, this is...
Play from 0:57:32 And for unique people, like real people, not like, you know, perfect people. Yeah, mean like go ugly, go whatever you gotta do. You ain't beaten that. Like the texture and the realism of this has been nuts. again, like we, I think we touched on this when we talked about 8.1 releasing, but if you do a lot of the photorealism stuff, you know, probably worth having your own personalization just with photorealism.
Play from 0:58:03 So you don't muddy the waters. Like. I didn't know that upfront. So I've kept on with this profile. And just again, like I'm being super picky about what images I select when it comes to photorealism. If it doesn't look real or could convince me that it's real, I'm not selecting it. Not selecting it. And now we've got, right, that personalization on the training data where this is just, you know, like, this is William Defoe.
Play from 0:58:31 Let me see if I like. With the knife too, look at that. Nice. Yah. You know, like these are, it's just so good. It's so, it's so good. It's the... you know, even the focus is incredible. Like, just... it does the fall-off better than any other tool, too.
0:58:56Reworking old prompts with Describe
Play from 0:58:57 Like, if you're trying to... Yes! prompt from V7, which I put in to describe, uh to use as in, that's kind of what I've been playing around with too. I've been trying to like create images I've created before, but getting them to the better quality now, cause I'm also just trying to like update my site and some of that stuff. So I'm taking, I'm assuming that, you know, maybe the older prompts. aren't all ideal for this new coherence.
Play from 0:59:23 So basically what I'm doing is I'm just taking an image dropping in and describe. Then I'll go over here. I'll run so maybe maybe I'll just give you an example, right? So like just drop this in here. Let's go all the way up. And then, you know, all I'm going to do is I'm going to go here. take this as an image prompt.
Play from 0:59:49 I'll maybe throw a little bit of a EXP in there and I'm gonna run just like an image weight without it, without an image weight and one with one because I'm also trying to understand how close it's gonna get just with its describe functionality versus giving it the juice, right? And that's kind of like the way that I was just quickly updating some of the work. So if you have to update work, because like, I don't know about you, but like anytime I show stuff that I've done, a year ago, even if it's good, I don't want to show people old stuff, you know, because it's just like it can look so much more real now.
Play from 1:00:23 So this is just sort of like a little hack I was doing is just kind of like taking images I like, dropping them into describe, you know, then using them as an image reference, right? Doing the weighting, I would copy and paste sort of the XP and IW so I could just quickly like boom, boom, boom, boom, boom, um Cute. Oh, let me here, let me run. run something.
Play from 1:00:49 So take this. And like I said, if I was going through this to save time, I would just do this, copy this. And then all I have to do is go boom. color's so good. It's crazy.
1:01:10Luma Uni vs Midjourney color
Play from 1:01:14 I've seen a lot of people comparing Luma-Uni to the best one. uh that too. We should touch on that. But the color is so good. it's... I can never get away from the color. That's always something that feels like it has like 10x the shades of what the other tools do. It always gets this unique blend of like, so good.
Play from 1:01:45 Yeah, and I mean, not every image is got that super ultimate crisp as a selector. I mean, maybe if there's a knock, it's that, but you're not going to get that in any model except for maybe like a Nano. But even then, it's like the people don't, you know, there's just no creativity with the imagery. unless you really go for it.
1:02:11Midjourney 8.2, 8.3, and V9
Play from 1:02:12 Which again, it's like the trade off of time and you know, like it's just those constant like things. anyway, like I think that was kind of like what I noticed uh in general, like just playing with it a lot more recently. They did mention something about 8.2 coming and I'll be honest, I didn't sit on that office hours, but apparently that's gonna help with personalization accuracy.
Play from 1:02:39 Produce more exciting and varied outputs. and reintroduce bolder and weirder visual styles. I like these three bullets. Let's go with three. Like these three bullets, I'm tired of seeing like the coherence, tired of seeing the quality, But like these three bullets, that's kind of exciting. I think they said something about this launching by the end of May, which, you know, we're two weeks away.
Play from 1:03:06 So maybe, you know, that's, I don't know, it's interesting. Here's also what they're working on, style balance, weirdness levels, personalization behavior. Where's the edit model? We got a V9 reference there already? Mm-hmm. Yeah, let's just kind of quickly go through this because I haven't gone through it in its entirety, but they're talking about weirdness a lot here. Let's Allowing more edgy and unusual generations.
Play from 1:03:31 Again, another thing that sets you apart from other models. Yeah. Avoiding outputs becoming unusably chaotic. Thank you. That was weird. Weird a thousand, you know, back in the day. edit model development. we go. Actively being trained and tested. Plan capabilities include object replacement, in-painting, out-painting, multi-object editing.
1:03:46Midjourney edit model questions
Play from 1:03:57 Day one release may ship with fewer capabilities in long term. It's always their plan. does that mean it's like an editing model as in like the editing model they have right now? Like not like an actual like we're gonna take this image and do something with it? I mean, look at this though, separate edit models. And then it's also, you know, like I have heard on past conversations that like the nano similarities, you know, like not being exactly like it, but in essence being something similar to that.
Play from 1:04:30 So that's the way I was thinking about it. It better not be just. is object replacement, inpainting, outpainting, multi-object editing. That sounds kind of like what's already there. That doesn't sound like an edit model. That sounds like it's the edit tour. I hope I'm wrong about that one. You know, what you know, what's?
Play from 1:04:59 There's there's some ambiguity in these notes. Because like you know, I think a better way to describe it. This is more feature based like. This is something like swapping replacing, but I like adding subtracting. uh You know color change it. You know, like I would rather lay it out that way. Then this this sounds like just the functionality itself.
Play from 1:05:24 which I think lends itself to some ambiguity there, to your point. But it's reported in the movie slowly internally. Yeah, I don't see a date, but... Early V9 work is underway. Let's see, focus on larger scale data improvements. And they're talking about an 8.3, new data improvements, additional tuning work.
Play from 1:05:50 Then nine, focused on larger scale data model improvements. David noted, bigger models are still needed for major gains and prompt understanding, coherence, structural consistency. But aesthetics alone can improve substantially before V9 arrives. Huh. Interesting. That whole wording of that note on the edit model, now, it'll make me feel too warm and fuzzy on the inside.
Play from 1:06:15 That makes it sound like we're just updating the editor, because they haven't updated it since like V6 or whatever that was. They haven't done anything with it. maybe that's... I mean, it's just like... What I've been doing though recently, lot of stuff I've been doing in C-dance is like using mid-ger... Because it's ingredients now, I just... can like actually build something that's cool or in Midjourney and then like here's the environment, here's the character, right? Like I can do that now or just like take the environment, take the character out of Midjourney, create the character, she would GPT, throw them in there and then it's like here's the ingredients, which it's nicer now than having to build, like I said, that first frame, like I don't have to build that necessary first frame, I can just draw a sketch of how I want it to look and also that, you know, put that in there.
Play from 1:06:57 So it's becoming more, I forget who the dude was recently that had the the like the 20 or 30 minute video that's incredible and it's all it's all mid-journey and like see dance. um Trying to think of who exactly it was. that... God Goblin guy. Yeah. Gossip Goblin? No. Is that it? like I think that might be it.
1:07:13Midjourney plus Seedance films
Play from 1:07:23 Because I think it looks so good because obviously it has the mid-journey aesthetic. one, Pat here, I'll just pull this up real quick just so everyone, if anyone hasn't seen this, I would definitely say go check it out. Here, let me share my screen real quick. Yeah, go for it. I think that's probably all the major things I'm seeing in these office hours.
Play from 1:07:48 Anyway, I think it's more on these models. I am curious about the edit model date. Do we have an ETA on that? mean, sooner the better. What's funny is you can, yeah, what? Hope so. Yes. But like anyone who's used Midjourney for a long time can look at this and just go immediately go. That's Midjourney. You know, you can just say you just has the vibe. It has the look. It's very like this one was. Yeah, this one was great. Like I thought this was really, really solid work and interesting that they used Midjourney and it wasn't like the standard, you know, talking points of it's nano.
Play from 1:08:22 It's whatever. It's like, yeah, good. Because this this is what it could be. Yeah. that's just the reality. It's a slower process. And this is what it could be. You know, like this is what it could be with them. If they wanted to just like have some consistency like this, you know, and we only we talk about this from a talk about this from a place of love. If they had, if they had the capabilities of just like being able to build within a world that's not necessarily just like one shot images where you can build it in a world people would have see a lot more of this stuff that's coming from them than it would from you know, nano or ChatGPT, like, because it's just better.
1:09:03Midjourney’s strange lane
Play from 1:09:07 Yeah. wrong, I'm not here to say, but they're not catering to enterprise. They're not catering to sort of the API route, which also has some correlation enterprise. They're just kind of like in their own lane. David always likes to sort of anchor it with saying they're a research lab. And to this point, I...
Play from 1:09:33 I haven't really seen what that looks like. And I think we won't until we know whatever the hell they're doing with the hardware secret projects that they're doing, which they've been talking about for a long time. And I mean, I think this is fair criticism. We haven't heard nothing about it. I I thought we were gonna hear something about it a year ago. Then I thought we were gonna hear something about it before Christmas. And we heard that, we heard that over and over again. We haven't gotten anything.
Play from 1:10:00 And I don't really know, just mid-jury has got its own lane. but we don't really know what that lane totally looks like. We can't even envision the roadmap yet until we know what some of these other projects are. I mean, we kind of did talk about it a little bit with the video thing. sounds like they're kind of maybe moving backwards on that being really a priority, but then they've also unlocked the ability to kind of go back and start researching the 3D thing again.
Play from 1:10:28 So I don't know, there's still a lot of things about mid-journey. know, we're both huge fans of Midjourney that I think the average person doesn't really know. A lot of people think Midjourney was a thing, you know, it's not still a thing, you know, and I just think there's so much TBD with them that we really don't know. mean, Midjourney's always been playing the... Yeah, and it's like, know, big thing for me is presentations.
Play from 1:10:57 I give a lot of them, right? Whenever I'm doing presentations that need imagery, mid-journey right away. Like I don't go to Nano to create imagery for like supporting images. go mid-journey because it's awesome and it looks really good and it looks different and it doesn't look like stock ever. And it's like, you know, it just has a different vibe to it. And I will always go there for that regardless. But. You know, for something like this, when people are going to start to make content like this, it's going to be easier.
Play from 1:11:24 I've seen more 20-minute stuff from the AI space than ever. And it's like, if you just want it to look a little different and don't look like everyone else, you go to a film. I've seen a lot of the film festival stuff. And the aesthetics is the same, because you can only use certain models to make consistent stuff. But start to dial that in, and that's what our content is going to become. Again, we say this at a place of love mid-journey.
Play from 1:11:51 like, what's world-build? If we want a world-build, if we want to carry, want to story-tell a little bit, just like make it easier. Yeah, the world building thing kind of. Hushed hushed a little bit and maybe they pivoted on to some degree, or maybe that's just more of the hardware stuff, but I just remember that coming up a lot more often. You know, it's like a. As sort of like a future destination for them, but um hey man, we're we're going. We're getting close to the end here, which probably wrap soon.
Play from 1:12:19 I I do want to bring up one other thing that we should mention here. We've both seen it. Some of the folks may have seen this. But this is great. This is great. This is this just is the perfect chef's kiss for all the hate for the AI sloper, you know, troll crowd out there. mean, you know, if you've got any experience with putting your work out there, doing anything with AI, making sure to call it out and then it's done any sort of real engagement, you're going to find a troll hiding underneath your post guaranteed.
1:12:28The Monet AI social experiment
Play from 1:12:58 You know, and if you go if you pass sort of that viral threshold, you're really going to get it. um So this this to me was amazing because if I got my story correct, this guy here, let's give him a proper shout out. uh SHL zero Ms was basically like hey. I generated an image in the style of a Monet painting using AI.
Play from 1:13:24 Please describe. in as much detail as possible. What makes this inferior to a real Monet painting? Like, just the wording of that is just the perfect bait here. And ultimately, uh you know, you get these great comments. know, it's just, it's Monet-ish, yes. Certainly pretty, but at the risk of sounding pretentious. It, in my opinion, would not hold up next to the real thing seen face to face in a museum.
Play from 1:13:50 The first time I really saw Monet's painting up close, it was astonishing. The brushstrokes here aren't as precise. Less subtle variation in the shapes of color. The over effect is smudgier. You know, you get all this stuff. The color looks like high school. Yeah, there you go. Looks at high school level art 101 work. Not even close. No cohesion of elements. I'm disappointed I have to even point this out. There's no cohesion in depth. Blah, blah, blah. I mean, just, you know, thing after thing. And then, you know, 24 hours go by.
Play from 1:14:17 This guy's collecting all these responses. It takes off and reveals that this is actually a Monet painting that he cropped into. And it's just the perfect encapsulation of this stuff, Because, you know, I think about this a lot, but I'm like, you know what? I think at some point I might just stop saying that these are AI imagery or even attaching the platform.
Play from 1:14:44 You know, it's just maybe I'll just describe what I was doing to create these images and let people guess. Because, you know, as soon as you say that, flags go up. people have opinions, they get very emotionally charged about their opinion on this stuff. And I just thought this was amazing. And I'm surprised, we didn't think of this to do this, but it's great.
Play from 1:15:10 It is so great on so many levels. We used to this with clients. We used to give them two different images. One was real, one was AI. And say the real one was AI and say the other one was real. And they always choose the real one no matter what. They're like, that one looks better. Then we swapped it, it was the same thing. So it was like, they're just saying it because it's AI. They're not even looking at it. They don't care.
1:15:13Why people over-detect AI
Play from 1:15:35 No. know, like that is a great, I'm glad you brought up that example because I think, you know, there was a long time ago too where I would do some of these things where I'm like, hey, see if you can guess which of these images are AI and put them, you know, head by head and, know, like no one, almost no one would get it perfect. So there was that sort of like mini test you could do. But then also when you get in this work, uh one of my clients, I think about one of my clients specifically on this, I was creating imagery that, I've just seen so many images.
Play from 1:16:06 I've got so much in the bank here between real photography and then the AI stuff. And I'm like, this stuff looks real, real, real, real. I'd die on a hill for it. And then somebody internally in the client says, this just looks way too AI, this, this, and this. And I'm like, OK. And I'm like, I don't see it. I don't see it. So I take it to third party. And I say, hey, what do you think of this image?
Play from 1:16:31 Or even take it to an LLM and sp- point out what they pointed out and say, hey, do you think this is this and this is this? And how certain would you be that this is an actual real photo? And boom, it's like, no, this is a real photo. It's just like, there's that bias. There's just that inherent bias that occurs on either side of the equation that people can't see through. You just can't see through it.
Play from 1:16:57 And I just think that's so funny. uh Because the minute they know it's AI, They're looking for the AI. Instead of looking at an image for two seconds, they're looking at it for 15 seconds. They're looking at me holding something in my hand where maybe one of the fingers is sort of bent like this, and that's now AI. And it's just like if you've ever captured photography, maybe with a faster shutter speed, you're going to get things that maybe don't look exactly right in a still image.
Play from 1:17:32 Mm-hmm. I don't know. That to me isn't talked about enough, like that component. I remember working with that client. was right at the beginning. I was just like, oh boy. Because there's an inherent bias here. How do I explain this to the client? There were several examples of it too. And it's just like, I think this is bleeding in here. I think because you guys know this is AI, we're looking too closely at some of this stuff.
Play from 1:17:57 And that's sort of like the. The Catch-22, if your brand, would do want to inspect the work if it's AI image generated. On the other hand, you don't want to inspect it too much, because then you're going to overanalyze it. It's also clients too, they just want to say something. know, it's like, give them even more ammunition to say something with like, it's AI, it's all, we'll really find something, you know, because it can never be good on the first pass, nothing, nothing can ever be good on the first pass, right? Like it's like, however many of you have had clients and something gets by on the first pass, like...
Play from 1:18:26 I mean, you might as well go buy a lottery ticket. Yeah, right. Like it's just like someone's always got to say so even if it's your even if it's your boss, unless you have a really good boss or you have a good team or whatever, someone's always going to find something that's always like, yeah, no, no. Yeah. How about we move it six pixels to the left? um You know, I don't know. What about the color? It's too saturated, isn't it?
Play from 1:18:52 That's what I'm saying. mean, there's just, I think there's a real problem with that, but this was a really good social experiment that I think a lot of us have probably seen, but it was just, it was awesome. It was awesome. I was wondering, mean, like, it's so, you mean, to credit, I don't know if this is a digital scan or if this is a photo of the actual ah image, but there is definitely, when you see it on a screen, it's pixelated.
Play from 1:19:21 When you see it in real life, it's totally different. there is that, like, you know, not to just totally bash everyone who... Yeah. see it in person, you can see like the textures and the bumps on the, you know, the paint where it dried in a certain way and like paint can produce color that, you know, that like digitally can't be captured because of just like the blending of it. Regardless, I'm not trying to say one thing or the other. Obviously people just hated it because they were just like, I just want to it.
Play from 1:19:48 um You know, we're not taking anything away from our people. one too? I love this one. Goes this guy goes, I think Monet would have wanted people to see the best version of his work, so I compiled all the critiques everyone had and generated an improvement. Just amazing, you know, just amazing. insane. Like, it's just like, get over it.
1:20:09AI backlash and disclosure debates
Play from 1:20:14 You know, like Hollywood, do you see the stuff with the Hollywood where it's like Academy Awards, like it can't be nominated for an award or something like that if it had AI. I'm butchering this, but the whole concept was like, like you couldn't do it and then everyone's like, why not? And then they're like, well, you have to disclose it. And the reason they want you to disclose it is because of this exactly, just because it's just like the immediate, like it's a backhanded way of getting everyone upset at it for no reason.
Play from 1:20:39 Right. And you got you got, you know, the goat himself, Peter Jackson over there, Mr. You know, war of the rings and everything goes, you know, like, what's the what's the big deal? It's just it's just like VFX. It's like he's like, I don't understand why everyone's freaking out. It's like I get it. People think it's just going to be like. People think it's just going to be. From now on, everything is only made this way. Right? Like, I think that's another problem that I have with everyone.
Play from 1:21:08 Like, just future project as if everything is going to be made this way. Like every movie is made with an iPhone or like everything is taken with digital camera. Yeah. Yeah. And, you know, like you obviously look at something I'm sure you like looking at analog film photos as much as you look like looking at digital photos, right? Like there's a place for both. Like it doesn't have to be one or the other. I personally like the disposable camera look.
Play from 1:21:34 Like I love that. You know, so it's like there's not it's not one thing or the next. That could be the whole thing at all times. Everything's gonna be synthetic. too simplified and all or nothing. know, they're just, the reality is, like, not to get too, like... Deep hours? Deep hours. We need another uh after dark.
Play from 1:22:00 oh man, I lost. I lost what I was gonna say. No, no, no, no, yeah. Yeah. it's not one thing or the other, like you said. It's not gonna be one thing, it's not gonna be the other, it's not gonna be like all in on this, all in on that. I think it's the same arc as what you saw with CGI, like where everyone loves Jurassic Park 1, but then as soon as you start introducing like, know, fake dinosaurs, everyone's like, I kinda wanna go back to, know, Jurassic Park 1 or Star Wars, like everyone likes the...
Play from 1:22:30 The flaws in Star Wars, the new ones, everyone's like, looks, everything is CGI, right? So it'll have its time, it'll peak, and people probably want to go back and they'll do their way. You'll have like your whole indie community that'll make things one way. You'll have your whole indie AI community that'll make things another way, like whatever. It's just like, appreciate that people are making stuff. Like that's it. Who cares how they made it? Yeah. You can break walls, can break in career path restrictions.
1:22:50AI as a career unlock
Play from 1:22:59 You can do anything now. What was I gonna do? I could have stayed in marketing for the next 20 years until I hopefully retire and been doing something I loved, but to love something continuously decade after decade, probably not a reality for me. I like to change and I like... know, experimentation and this and that and the other. And it's just like, I'm doing different things now.
Play from 1:23:24 I mean, I've completely pivoted my career. So have you. That's not possible without AI. you know, that has happened to a lot of people now. I mean, and so it's just a really interesting thing. mean, obviously, like the AI conversation as a whole is very polarizing. It's not inherently good or it's not inherently bad. I'm just saying that uh there are really good pieces about it. to be on one side of the fence or the other completely is just, I think, objectively wrong.
Play from 1:23:53 think it's got to be somewhere in the middle, right? Even if that is 95 % human. Still that 5%, you can't sit here and tell me that all AI is bad, right? uh I was just going to say, think today's society is absolute. It pretends like everything's our absolute truths. And the reality is there's only a few absolute truths in anything. There's like on one hand absolute truths.
Play from 1:24:19 So with that said, anything between has different pieces of it and people just like to put AI into the absolute truth bucket. Just like, get over it. I'm happy people wanna make stuff. Just keep making stuff, keep being weird. Like that's it. Like that's what this is all about. know, like I like that people just make stuff and it makes them happy that they make stuff. You know, some people cook, some people knit. You're still making stuff.
1:24:30Keep making weird stuff
Play from 1:24:45 I think it's cool. Yeah, you cook, that's a... Nah, nah don't but... I appreciate it. I appreciate when people like stuff. I would tell your, you know, say that... Tell your local. I don't know. Seam stress. We'll go with seam stress.
Play from 1:25:14 Crochet er. Is that crochet er? um I did. I think that's it. Like, I think that's the I think that's the right term. I don't know. Yeah. was not gonna come up with that. Like, I was not gonna pull that out of the depths of my brain. Like, there was no way. good one. It might be useful in a mid-journey prompt here or there, so we'll bank that one for later.
Play from 1:25:42 But should we wrap that one up after seamstress? All right. and uh as long as Rory and I are not traveling, we'll be back next week. We'll be back next week. And we appreciate you guys joining us and we'll Rory anything else you want to say before we go? you're still here, like, if you're if you're still here, you're probably subscribed already. But if you're not do that anyway, and if you haven't told someone you have to tell your local seamstress, let's add your tailor in there.
1:25:45Wrap-up and seamstress CTA
Play from 1:26:12 Is that maybe a dry cleaning person that you have? Yeah, go system there. Like go, go to you know, your local car dealership and talk to them. They might like they might car salesman. We haven't had that one. I don't think you know. think we've had a bunch of things here. you're, uh maybe, that's it, that's all I got. I'm not gonna waste everyone's time anymore. uh We're running to the end here of occupations, as far.
Play from 1:26:42 Yeah. I'm glad we're able to do this. uh yeah, we'll be back next week. We'll catch you guys in the next one. Peace.
