Transcript · September 29, 2024 · 14K words
Ep. 12: Midjourney Mysteries Unveiled: Your Questions, Our Answers (Part 2)
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.
0:00:00Intro
Play from 0:00:00 Hey, hey, hey, we're back Midjourney fast 📍 hours. It must be Friday, Rory. we're back in the building, man. I don't know these, these weeks in between recording just seem like they take forever to get here. And I know you and I are just like waiting till Friday so we can get into this stuff. we had that guest last week, Chris branch. Got some great feedback on that. It was an awesome episode. Try to mix things up for you guys. And, you know, Rory and I, I think are on the mindset of. We want to try different things every week. We don't want to make this show super, super predictable or boring in any way. So we're going to continue to mix and guess we're going to continue to bring different formats to you in our approach.
Play from 0:00:38 And just keep that in mind this week, though, we're going to go back to those AMA questions from two weeks ago, because we had so many good ones and we only got through a handful of them. So we've got those on deck for today. And then, next week, I think we're also going to bring Guest and I think we got a damn good one. So, just a few notes, in house cleaning updates as we, as we get started here today, but Rory let's go, buddy, So excited, I like having my business stuff where I have meetings and calls, but this is really where I enjoy, you know, there's the content creation part. That's fun.
Play from 0:01:11 Sometimes it takes time. Sometimes it feels like a chore. But this is the, this is the sauce right here. We'll get to just jam out, talk about whatever, and then, you know, let her rip for an hour or so. So yeah, man, I'm like, uh, drew and I just, you know, guys, little insight into the background. You know, we have like a little bromance texts. That's just like, dude, can't wait for Friday. Like, you know, week's going rough. Baby's not sleeping, you know, getting spit up on. I'm like, can't wait for Friday. Just want to talk about this stuff, let loose and have a good, entryway. Yeah, bro, because weekends are different when you have kids, especially young kids, because they're not, they actually feel like the work week and the work week feels like the weekend, which in theory is good because you got five of those, but at the same time, it's not like a relaxing thing.
Play from 0:01:57 So it is different. but yeah, man, it's, it's great to be back here. And honestly, I think with the AMA questions we had, like, I just felt like that was such a, Great episode. And we only got to a few of like the key questions. And so we've got some really good ones still to get through. before we do that, I was just thinking to, you know, office hours, updates, like just as a heads up, I didn't really hear anything major. That really warranted a huge update there. they're still talking about things with, video 3d coming. They're still talking about, you know, sort of weighing this option between, you know, if they're going to roll video out with, like, if they're going to wait for seven and the capabilities there, or if they're going to sort of give this like a trial run with maybe some users with six, I don't know.
Play from 0:02:41 But again, it's all hypothetical. These things change week to week. So I don't think there was anything of note that it was just like, okay. Changes the game, so I guess let's start there. But anything else before we get into the questions or do you just want to jump right in now? I think the one thing that I noticed recently just so everyone stays updated is that there was that you know The I think if you want to call it the biggest change is a really small one was that little personalization tab They had a website. So yeah If you guys aren't familiar with personalize or you haven't done it yet, they've made it a lot easier. Now it's really literally just right on the navigation bar on the side of the web browser. You just hit personalize and then you just go rank images. yeah, maybe it's easier if we pull it up for us.
0:02:54Photorealism Techniques
Play from 0:03:22 Yeah. I mean, it's just, it's this tab right here in the navigation and yeah, like Rory said, I mean you can essentially. Click this and get right into it. Level eight. Look at you. Damn, dude. You went, you went hard. Look at this. Yeah, I Don't might be a record You know like dude, I mean there was a time where it's like every day I was doing that for for some free hours Cuz I was fiend and so Yeah, man, that's how I got like I'm embarrassed to pull mine up. I think. I to pull mine up now.
Play from 0:03:53 Yeah, dude, that's how I went from like, you know, uh, the 60 hours to like, you know, virtually 120. It was just like every day I was relying on that extra, that extra hour. But, yeah, this isn't like, it's nothing really big. It's just a, it's just a tab and then again, like, I don't, I wouldn't even go into this tab really other than to see this part because really, I would much rather go into tasks and then decide which one I want to image rank, Is it, is it the aesthetics?
Play from 0:04:20 Is it the prompt accuracy? Both of these, I think, factor into your personalization. and then you've got this curate community front page. I think we talked about that last time, uh, before Chris was on. And then you can do the same thing with, with Niji there was a, one of the thing here, Rory. What did I miss? Did you complete, did you complete the last survey that was here? There was a new survey that dropped. Was it the one with the psychographics where we, where it was like, it got like the results with like 17 different profile analysis? Is that the one or was it a different one?
Play from 0:04:52 how many surveys have you done so far? I think two. Okay, so I think this was the third one. Oh, I missed it. I might have to go do that one now. Go, go check your tasks and see if it's in there. I will warn you. It's like another 150, 160 questions. So it did take me a while. And actually I was trying to do it on my phone and, uh, I got about halfway through and I had to, like, I was busy. I was on the run and then I pulled it back up and it reset everything. So I had to go back and do it all again.
Play from 0:05:21 So that was double self analysis, but it was cool. Yeah. Like I did get some, there were some insights that you get right away, like you're talking about, where it was just like, you are in the 90th percentile for this, you're in the 10th Percentile for this. Right. So, I mean, there were some interesting things there. I mean, nothing that was super eye opening in any way. It's more or less how they're going to use that information. But that's there. And I think if you complete that, you get four free Fast Hours. So that's probably worth it.
Play from 0:05:47 All right. Cause I'm, you know, I'm going to be honest right now. I'm kind of like scared to touch the personalization stuff. Cause I really like my code right now and it's working and it's exactly what I want. So I just don't know how much it's going to shift it. Just so you guys know, these are the kind of the internal battles as well. I like where my code's at. I don't know if I want it to change. So I'm like, do I go rank? But also at the same time, I probably am just overthinking it. It can only get better, but you know, there are certain things that I like, you know, I don't use a default still. Like, which is weird. I mean, I don't know if you still use a default, or if you're using a default, but.
Play from 0:06:23 manually. Yeah. Oh yeah. No, I just, I have it on default now, but I mean, I think it's better probably unless you're using it exclusively to leave it off and add it. Right. Because otherwise you're just going to have to turn it on and off every time. It's not like you can indicate that in the parameter, right? It's just going to throw it in there. Do you ever feel there's a random question? This is a personal question. Do you ever feel like. using style ref or something like personalized and things like that. Like you're going to lose some of the, uh, lose some of the sharpness on your text prompting, like just from just from like experience and doing it a million times.
Play from 0:06:55 Cause I, I feel like half the time I'm like, let me start with my tax prompt. Even if it takes a long way, like, like just from like an atrophy skillset standpoint. Like, oh yeah, dude, I, I would say my text prompting has atrophy to some degree because, because now, right? Like it's just, there are so many other ways to influence the end result, It's not so much just about the text as much as it was previously. I mean, remember when it was just text and then it was just image ref, Like it was just. That was pretty much it. So I think like in those use cases for sure. And then I think over time, Like when you get the style ref and then you get the codes and you get the personalized, like there are just so many things that you can do that don't require very long, specific prompts again, unless you've got very, something very specific that you're going for.
Play from 0:07:44 Like if I'm unlocking those style codes, for example. Any prompt that I run with those is usually like four words or less. You know what I mean? Like I'm just trying to get a sense of what it can bring in and leaving it open to the creativity of both the style code and what, the model's going to give me. A hundred percent. And also just shout out for everyone. If you guys are going down the style code rabbit hole and want to unlock the treasure chest, that is drew Brucker's Twitter. you know, experience, please go check him out on there.
Play from 0:08:12 Cause the style codes you've been dropping for the last two weeks have been awesome. So even they can, you know, I'm scrolling Twitter and they catch my eyes significantly and I have limited my following on there. I don't follow anything that's like not AI on that account. So you're in the, you're in the echo chamber, baby. I like it that way. Yeah. I, well, I mean, so here's the story behind that, right? It's just, We're both busy, dude. I mean, I want to keep providing some value and, you know, I don't know if the audience knows this, but like also like I took on a very big fractional client. I'm pretty much 30 hours a week with this client. So a lot of that extra creation time has really consolidated for me.
Play from 0:08:51 So it's all about like, how, how can I add some sort of value as quickly as possible? And so I've just spent so much time with the style codes. It just became like, well, shit, let me just like schedule these out in X. Because I've got so many of these and X is a platform I'm trying to get more active on anyway. Like I'm trying to kind of shift a little bit away from LinkedIn, not like totally remove it, but obviously my posting cadence on there has gone down. So like, I don't know, it's kind of serving that purpose for the time being, but the codes, yeah, like I'm trying to share really dope codes.
Play from 0:09:22 Like I'm, I'm curating this, I'm running through all the codes, picking out, a short list of those. And even within that short list. maybe using like a two or three out of the 10 of the shortlist. Right. So it's like, it is heavily curated and I'm trying to provide styles for everybody, not just for me. things that I think other people might like too. So trying to test that out, but it seems to. It seems to be a fun little thing for a lot of people, you know, just to kind of test that out.
Play from 0:09:48 It's like like I said, it's like a break off civilization from from major. I feel like there's like people that are still in here and then there's ones that are like just in their mining Just like every day like I just want style codes because you see it. I mean look guys we do stuff on social media The, the idea is to get engagement sometimes, right? You know, sometimes we just like what we're doing. Sometimes we're just looking for some engagement every once in a while. Being totally honest there. But you know, style code stuff, I love it. I just love seeing it.
Play from 0:10:15 I don't personally do it all the time because I'm just kind of like, I need to go in there and do things and like get things done. And the style codes are kind of random. So when I do have time though, I love to go in there and just kind of rip and just like, R4, you know, repeat 40. Let's go. Let's see. Yeah, but it's cool, man. I really like the, this is just a little superlative. I really like the, uh, some of the stuff you've been putting out, especially some of the illustrative, like more vector based stuff. That's got me like, that's the other part about this, usually I'm much more photorealistic based with style codes.
Play from 0:10:48 I'm kind of looking like for me personally. I'm really looking for those like illustrative gems, dude, like the 3D gems, the illustrative gems with like dope color palettes. And I'm like, those are the ones that, um, I really like. So I've been playing around a lot more with like different styles in that realm lately because of the codes. it's really solid, man. Like, I think that's, that's the one thing that I've really liked about the style codes. Is that it forces me to go down a path that I wouldn't go down normally. Right. I, it's just something that I, I, I have like this aesthetic preference in my head that I like. And I always do.
Play from 0:11:21 And they like shift that and they shake it up. Cause I'm just like, Oh, that's really cool. And you think about that or like how to make something look really radically. Different just from the same prompt to, we've talked about this a few times. I just love blending them in there and like dropping the style way down just to give it a little effect. So even like the illustrative stuff is really cool, depending on what the I know, it's fun. I mean, it's, it's so much fun because like, obviously like just different use cases in general, but, it's just like a whole different sort of challenge, but opportunity at the same time, Like illustrative doesn't have to be this, this perfect thing in a lot of ways.
Play from 0:11:54 Yeah. I think one fun thing, if anyone wants to go play around with it too, is if you upload pictures of yourself into there and use them as character references and then run style codes on them with high style weight, you get like, you can like just, you know, make avatars of yourself in these wild styles, which I, you know, I always find fun. It's always just a fun experience. I need to play around with that. That sounds fun. Actually, I should do that because it's just ridiculous. You know, I did that the other day and threw one into runway. Like it was like a, it looked like a, like an animated, like flip book style.
Play from 0:12:25 And like, you know, remember how the old cartoons, like no one walked correctly. It was just like two legs, like kind of like shaking and like moving to the side. Cracking up this stuff. So dumb when you see what you can do and it's, you know, whatever. But that's why, you know, I think it's, interesting as because the rolling back to the whole text prompting thing, because I, still find myself writing, you know, exactly what I want, taking my time on it. And then, you know, if I can't get it, then I'm like, all right, style reference. Let's go. I just say, I don't know. There's something like yeah. like manual about it's like sketch drawing still like I still like to sketch every once in a while and that even if it's not good it just like it like keeps my tools you know my skills sharp in that sense we're looking at a couple questions here that probably have some sort of cater uh some sort of
Play from 0:13:10 relevance to what we're talking about the amount of good questions awesome just like so insane that some of these questions were I was expecting really basic questions to be quite honest, just because whenever we ask this stuff, no one really, especially when I ask personally on my posts, I never really get the most, you know, detailed questions. They're more like beginner or mediocre questions, but we've actually got some that it's like, okay, you've obviously been in the tool for a while and troubleshoot these things. So that to me is more of like that advanced stage where, you know, you know, of a lot of the things that exist. But you're just trying to see if there are other ways to do it.
Play from 0:13:48 are we going to pop into those? Should we do that now? Yeah, let's, yeah, let's jump into those. 📍 let's start with a nice little softball one. but I think it is a question that a lot of people have. This comes from. Spirit Novel, who also provided a great question last time, but it's, this is a good question that I've seen asked quite a bit. What is the best prompt recipe you use to get extreme photorealism? Is that a specific SREF combo and or parameters? Photorealism of people and products specifically.
0:14:15Character Consistency + --cref
Play from 0:14:17 Do you want to go first? Sure. Yeah, I mean Yeah, go for that one. Yeah, let's just stack it. this is definitely A big question to unpack because I think there are a lot of different things that you can do. Okay. So with extreme photo realism one thing I would start with is indicating the type of photography that you're going for right off the bat in the prompt. That was something that we've talked about before. And I think you even asked me, you're like, how often do you do that? And it's like, if it's photo realism, I like to do it almost all the time, because it's just a great way to indicate exactly the type of photography I'm using.
Play from 0:14:49 And what I mean by that is like, is this product photography? Is this commercial photography? Is this editorial photography, fashion, street, landscape? You know, like any of those things, I think then puts your image into this nice category of this is photo realism without having to say photo realism. So if you've ever played with those photography types, you know, you get back different results based on those things, So that's, that's the first thing I always think of. The second thing I'm thinking about is, to have something like style raw on. I don't know how impactful that is. We've talked about that as well, but I would have that on, especially if you're going for style realism, then you can get into, you know, things like keywords and tokens, this is where it gets into like things like film stock, which you and I have both talked a ton about.
Play from 0:15:34 We've both made a lot of posts about, so if you indicate things like the type of film, this was taking on, take it on the type of camera that was used. The type of lens that was used, maybe an F stop. So like an F 1. 8, for example, that just means that your lens is wide open with an aperture. that's taking in a lot of light. And so what that is good for those lower apertures is a lot of the times like portraits or photo realism. Also, I would be thinking about. Things like, you know, some of these hacks that you hear, like with, Hey, if you put like iPhone photo or Snapchat photo or Instagram photo or something like that, like those will get real real, Where it almost feels like I took the photo on my phone kind of thing where it's not even like a polished type of look.
Play from 0:16:21 So I think there are levels of realism, That you need to consider as well. But those I think are the main ways that immediately come to mind. And I'm sure I'll think of something else while you're going to dive into yours, Yeah, I think there's the it's all comes back to like how the tool works, if you think about how these models are trained, like they're putting in training Images from somewhere, right? Wherever they're sourcing them from, then they're tagging them. Like, how are they going to describe an image? So do you want to, like, when you're prompt, basically you're going into the tool, it's like going to retrieve those tokens and then kind of like mash everything together.
Play from 0:16:56 Right. So it's always like, how do we, how do we sort of look at that now? That's what drew said perfectly is about like the, the photo category, So you have like editorial photo, documentary photo, any of the, those types of, you know, fashion photography, product photography, they all have a, radically different aesthetic, even though they might seem the same. But to me, that's really important. Then like when I want something that's like super sharp, I'm always focusing on, you know, again, what, what is like the hierarchy of what I'm looking for? Like, do I want the person to be the main focus? If I want really subtle details, I want all of the journeys focus on the person.
Play from 0:17:32 So what drew is saying about the background, Like an F stop the less detail you have in the background or the less detail it has to go into the background, the more it can put into the face, So when you're looking at something like that, person smiling in the foreground, blurred background, like you can do stuff like that. So it'll really just focus. You can even say, you know, person in the, you know, in the foreground, smiling, looking left in sharp focus, you know, background is blurred. So it's a way to like, kind of, I don't want to say hack the system, but really like puts all that generations, memory or like ideation into the person or into the object or into the thing you're trying to generate.
Play from 0:18:07 So that works really well. For me, I typically like to put something like, 35 millimeter. I just like the way it looks. It tends to come out really well. Another great token. Yeah. Yeah. That one. And now I've been adding this a lot lately is just at the end when I want it to be that it's a high precision photography. I don't know why, but I use that a lot and it comes out fairly well. In addition to you think about as just as many photo realistic tokens or categories that you can put in there sometimes to like really dial it up. When you're talking about contrast, or you're talking about texture, or you're talking about film grain, or you're talking about whatever.
Play from 0:18:42 You just think about where those, when you put that token in, where it's going in it's brain to like, retrieve that. It's like, oh, it's going to all the things that are super photorealistic, cause an illustration wouldn't have. Some of those, you know, some of those, it's like I say, 35 mm on an illustration tag, right? It's not gonna be there. So you're just putting yourself in the right category to get the right, like random access generation there. but yeah, I think looking at basically those things that we talked about, that's really, you know, we spend a lot of time on photorealistic stuff, but it felt like hacking a long time ago.
Play from 0:19:13 But yeah, Yeah, it's not like now, now it's a little bit like even with the model, or the version it tends to go more photorealistic than it ever has previously. Like that was, that was the big jump with six, it went from a little bit more of that styling. to more of like a little bit of a photorealistic jump. You know, I remember that specifically with when V6 launched. It was like, okay, you know, the, the, the prompt coherence is different, but also, I'm getting like images that look much more real. Out of the box without having to like necessarily say it.
Play from 0:19:44 Yeah. Quality two also works can also work too well. Sometimes I will preface that, you know, quality two, if you're adding that parameter dash dash Q2, that'll give you more fine grain detail, but also at the same time, sometimes it's too much, like sometimes it goes overboard. I noticed that a lot with, with my, if it's a closeup of a face, it gets like super saturated or super. contrasted to the point where it's not cool anymore. Like it looks overdone. Like I overused a filter on Instagram or something. also I think one thing probably should mention here is the stylize. if you're going to use stylize on these two, anytime you go up over like to me, two 50, 300, anything above that range, you start to lose some of the real, you get some crazy details, but you also start to lose a little bit of that.
Play from 0:20:28 finer detail of like skin and blemishes and things like that because it just starts to put that gloss on and it gets way too you'll notice it when you're doing it's like the higher you go on the stylized the closer in it'll get to the person's face and like the more Like AI, it looks like chiseled jawline, huge lips, perfect eyebrows. So this comes back to that caveat with personalization to, you know, like how, well trained is your personalization to a particular style? How many images have you ranked?
Play from 0:20:59 your own personal preference. It was a lot easier to tell you guys without personalization, sort of those Barriers and thresholds of like what we like to use for, to achieve that photo realism, Like that was the other thing that we used to mention a lot is like, okay, yeah, if you keep it within this range, it's going to look a lot more photo realistic. Now it could be different. It's hard to say, right? Because I'm not sitting around here with 50 other people's style or personalization codes, but you know, it is much more on a case by case basis.
0:21:15Mood Boards
Play from 0:21:28 Now, Hey, let me throw in this question too and get your take. Cause it's like a, uh, secondary. Follow up to this question. This is from my guy, Noah Davis, shout out to Noah. Good dude. What do you feel is the best approach to getting the most realistic and cinematic images in Midjourney for image to video output in say runway or Luma? Totally. So The one thing that you want to focus on with this is really like, you have to direct the motion that you're trying to achieve in runway. So what I mean by that is, we've, I've talked about this example before. If you want, if you want to like a image, let's just say we have a car that's driving from, you know, right to left.
Play from 0:22:09 That's what we want. We want the car to be, we want like a shot of the car like driving. Now, most people would just generate the car and then not think twice about it and just be like, Oh, I can put it into runway or Luma or Kling. And like, it'll animate the wheels and it'll look like it's driving. But when the wheels are static, that like signals to the still like, or static, meaning that like the, the rims are visible, they're in like a freeze motion, so there doesn't look like they're actually moving. It's just going to like slide the car across the image. Like it doesn't, man, don't put me on blast like that. Remember I was telling you about that first time I did that with the car. Yeah, exact same thing. Just like. The wheels didn't move like I was expecting it to move. It just fucking slid and it was like, Oh shit.
Play from 0:22:48 It's like, this sucks. But the trick for something like that is like in your prompt, you want to put spinning wheels. So it has the wheels already spinning. You're like giving yourself a leg up in terms of generating. So something like that. Really important to me also with people, because sometimes it can get like runway. I know specifically can do this. Like the motion isn't really natural because like you start with some people that are just in like a static pose, meaning like, Oh, it's like a fashion post. And then, like, that's kind of all they do, and maybe you get, like, a little head movement or something. A lot of times I'll do, you know, natural dynamic motion in both prompts.
Play from 0:23:25 I'll do it in Runway and in Midjourney, because sometimes you'll get, like, I like people that are, that don't look like it's a standard pose. Like, a hand might be flailing up in the air like this, because that's just halfway done, and then that, you can let Runway take over the rest of it. Right. So it's like starting kind of like freeze frame, like in a, in some sort of candid style pose, or just, you know, use, I use cheering a lot, which is a weird token, you know, to get people in like this kind of like interesting looking, uh, position.
Play from 0:23:53 So they're not normally just, you know, stone faced arms at their side, looking straight at the camera. Cause you're not giving runway much to work with or Luma or Kling, uh, much to work with. So those are sort of the ways that I'm looking at it. Also. Quick tip. Upscaling your images can be really helpful. So the more detail that these tools can pick up the video editors, the better it is. specifically, if you have a person that might be in the mid ground or maybe the background, And they have a fuzzy face. If you put that into one of the tools, it's going to morph the face to whatever it wants because it can't see it.
Play from 0:24:28 But if you upscale that by, you know, four X, maybe in Magnific or something like Krea or even just two X, and you get, you know, good facial definition. Like it'll stick, like it sticks pretty well with some of this stuff. The masking has gotten much better. So that's a really detailed answer, but I think that's kind of the way to approach is like, how do I build motion into my images first so that it gives the tools a lot better opportunity to succeed.
Play from 0:24:54 From that standpoint. So that's long winded answer. No, that's a good, that's a really good answer. I don't really have much to add other than a very, very small note to this. You asked about getting the most realistic and cinematic images. So going back to that point about. The type of photography, love it. I love to use cinematic still at the very, it's the very first token that I would put in before subject, before action, before even camera angle. I mean, so if you put cinematic still or movie still, you're going to get something that feels cinematic.
Play from 0:25:29 So I think that one's relatively easy to achieve, you know, from that standpoint. But. Another one I've been holding on to has been, uh, IMAX as a token, that one works really well. IMAX, interesting, haven't heard that one before. For whatever reason, yeah, that one tends to work very well, especially for big, like, landscape shots or, you know, just epic scenery for whatever reason. it's worked a bunch of times for me. It's specially works really well with like two by one aspect ratio. So, I think it's a really good token to use just to keep that one in your back pocket.
Play from 0:26:01 Ooh, okay. I max. Yeah, I like that. I might have one. That's interesting like that, that I haven't really thought about in a while. You were just talking about IMAX, you know, one thing that I didn't like, I didn't think about for a while is if you use stills archive, and then you use a website, dude, it's incredible. I mean, like, it's almost like a free Frickin hack. I mean, this is in V six too, but I mean, these almost look like the exact shots from TV archive, netflix.
Play from 0:26:34 com like, look at these, I mean, there's no prompt other than that, That's wild, man. Stills archive, HBO, what does this look like? That looks a little Game of Thrones y. This is for entertainment only, YouTube. Don't take us down for this. I mean, that looks very much so little G. O. T. action right there. Yeah. So, I mean, like, you know, you can play with these. That is King Joffrey to a T.
Play from 0:27:02 Right. So, I mean, like, you know, this is a little Hacky, Like, okay, like, look. A little Jon Snow. Stillsarchivehbo. com. I didn't even prompt Jon Snow in there. There's Jon Snow, and then I think the next one's Stranger Things. hundred percent. I mean, that's it, dude. I mean, so this is like, uh, it's an interesting one. you could put stills archive. I mean, you don't need this, but like, if you put these. coms, which I've played with a little bit too, like what's the difference between Netflix CBS, you'll get those slight variations.
Play from 0:27:33 And, and it's funny because the shots start to look like you could tell exactly what network they're on, which I think is just It's like, wow. Okay. Like I can look at this and know exactly what network it's from. so anyway, that's like a, that's like a very interesting thing that I've noticed with that. That's a really good token. And I want to go play around with that now. Now I kind of want to see that I was playing with one the other day. Actually, it was just funny and don't, uh, you know, I couldn't get what I want guys. So even if I show you this, just regardless, don't, you know, don't hate me for it.
Play from 0:28:05 This one was. television still like I was trying to get like a wheel of wheel of jeopardy, but like you look at the people, I mean, how isn't that ridiculous though? Like that's what I'm saying. Like, so I think this guy It's cracking up, man. I was like, there's some here that are so funny, like it's so stereotypical, like television still from wheel of fortune, but obviously it wasn't creating the wheel of fortune that I was looking for, even with image referencing style, referencing, but how wild just couldn't do it.
Play from 0:28:35 But the television still, that's a good one. cause that one, for whatever reason works really well. Now, the other thing I think to, to show here is let me find the boots. This was an example of what I was talking about earlier, like natural motion, right? Like instead of two boots standing side by side static, because I have this boot that's already in motion, it like runway is going to pick up on it or Kling is going to pick up on it or Luma that it's like, okay, we're putting the foot down and then we're going to take the next foot.
Play from 0:29:03 It's like, all right, we're already halfway through. Yes. Right. Nice. So Something like that is going to work way better than, let me see, maybe something like this too could work a little bit better. It's like the heel of the boots already up. So it's like in motion. Yeah. But like if that heel were flat on the ground, yeah. Different story. Cause it might just, you get like these little like vibrations and they don't move sometimes because it doesn't know what to do. So yeah, you're giving, you're giving it just an extra bit of information and knowledge and context to this one for it to become more predictable. Yeah, yeah, like this one not going to work.
Play from 0:29:39 Like this is a, this is a bad example. This is an example of what not to use because it won't, sometimes you'll get this and it'll just start walking backwards, right? Like you, depending on even if you prompt it, you know, walking towards the camera. It won't say it won't do that because it does like, doesn't know. But like with something like this, it knows that we're like moving forward because that's the way, like the motions already built in. So if we have that one, there's also, let me see if I have, this is good. I mean, like if you're just getting into runway and Luma too, this is great knowledge to have going in because otherwise that's probably going to take you a little while to figure out, All right. So here's an example, right? Like this one is what I'm talking about, where like we're, we're starting to like bake some motion in here to what we're, we're looking at. So you get something like this, right? You can already tell that there's like motion in the background.
Play from 0:30:27 There's people, this guy's got, you know, he's, he's kind of got all the, what the lighting is already like painting half of his body. So there's like some motion baked in something like this too. Where you can see like there's hands in the air. There's like no distinct poses. Like everything is sort of in motion already. Now, if we go down and take a look at some more of like these cheering style images here, where I'm talking about, like we have, let's just say, or just motion in general, like you can tell something like this it's in motion. Like our hands are odd positions. Our hair is all out of whack. There's people in the background.
Play from 0:31:03 So you're just going to get. Because the hair is already out of place, it's not going to stay static because it's like picking up on that as being motion, Or even something along the lines of when I scroll all the way up to here a little bit more. this specifically is like a perfect example, right? Like her hands might be jacked, but you can tell that she's dancing and her head is in a weird position and her arms are up and her hand is like kind of in the middle of the frame.
Play from 0:31:28 Right? So that you might look at that and be like, that's not a great image to share. It's like a pot finished product, but I might take a look at that and be like, that's a great image for, you know, putting in a run in runway or luma. Cause something like this, again, her hair is moving, her hands are in a weird position. There's some motion in the background. So it's already halfway there. You don't have to tell them to do much. So when I say stuff like that, like what I'm, that's what I mean.
Play from 0:31:53 but dude, I have to go, no, I have to really go check out this. Sit like still library or still still archive. That's a good one. That's a really solid one. It's crazy I like that was and that was from a while ago, too I like I just remember like you don't even need to put anything other than that I mean, I didn't even try making those complicated by any means. I just wanted to see what was gonna come out it's silly. it almost felt like cheating is like yo, what is this?
Play from 0:32:18 I'm jealous But that's a but cinematic. I mean, yeah, so like that's that's a great example of how you do something like that you also got to think about how those images like are those images going to end up on the web, going back to the training model, that's how they're grabbing these. They're also going into all text image descriptions. Right of for, you know, trying to match keywords and tokens in that way too. So that's, that's interesting. let's go to the next question. This is from Liam Broadbent asks, I'd love for you to dive deep into developing an image style for a brand from strategy to deliver well, Liam.
Play from 0:32:54 Uh, we don't have time to do all of that, but, you know, I will say this, like this was something, that I posted about before. This is actually part of, a recent talk that I gave as well. I don't think it's, it has to be overly complicated. So I'm just going to like, give you more of a foundational level of knowledge here. I think there are a few things that you want to do. So if you're attaching yourself to a brand, whether you work there full time or you're an agency or your contract or whatever it is, you want to know. Who your target audience is, What they look like. So their demographic, their age, their gender, their ethnicity, what they typically wear, where you'd want your photographs to be shot at.
0:32:55Brand Styling
Play from 0:33:31 Like think like a photographer in that sense, the camera angles to tell your story, besides people, what other elements are you trying to incorporate? Are they tools that this person uses? are they just objects that reflect, you know, that particular industry, you know, things like that. So I would almost create this list, this category of list of things that are going to apply to you. Same with your visual signature, like who is your company, what's the personality you're taking on, what are your core values, what's your brand color palette, So I think that's one of the key ones. If you're trying to make something feel a little bit more. Custom and tailored to the brand. And so I think if, and then the last thing I would just say is like usage, where are you going to place these?
Play from 0:34:13 thinking about like image dimensions and placement and so forth. So I think if you map that out, you've got a very good head start front loading the work of all the types of imagery and maybe nuances that you're looking for. And then you can take like, this is where you can really get. versatile with the nature that you do it, right. You could find like just an S ref image in general and use that, along the way to sort of drive, the styling and color and aesthetic by using style weight, people ask all the time, like, do you use style codes for this? We've talked about that as well. I think you can, the caveat there is like knowing what the right style code is.
Play from 0:34:51 That you could potentially use for your brand, but either way, when you're using a style reference, it needs to come down to style weighting. There's going to be an appropriate amount that you're going to need to apply. So that's where it comes down to, you know, sort of the experimentation part of it. And then, you know, I'll just show you a quick example here they'll, they'll showcase the use case pretty well. let's just say, you know, I want to, so I've got like a, a brand that is, their colors are pink, it's a nice pink gradient, for example, like they've got two different colors of pink. So I just took like the color palette or the swatch or the gradient, use that as a screenshot. Use that as the style reference. And then I can prompt anything I want, throw that in here
Play from 0:35:36 and then just style weight it, which I did here, which is like a style weight of 40. And it's going to start to infuse that color. And that's without me even indicating anything specific that needs to be pink. You don't see pink in this prompt at all, So it's going to automatically start to apply it in different ways. that's going to depend on the weighting and all those good things, but that's a good head start. Now, if you want certain things to be a brand color, you would obviously point those things out, but this is just like to a really simplified example of how you can start to do those things to infuse brand colors, to, to get more specific, Because if you know who your ICP is, their age, their gender, and you've got your color palette and the type of environment or scene they're in, you're already most of the way there.
Play from 0:36:19 So, that's more of a quick class in like a three minute nutshell. Yes. It's pretty good, dude. that's pretty solid. I think the other thing maybe you could potentially do is depending on what the, what the end result is, right? If it's your current brand that you're utilizing and you already have brand guidelines or you're developing a brand, maybe it's something new or you know, you're working in an advertising agency and you know, someone pitches you ideas on, you know, they want to do some sub brands or some unique projects that require a whole like style build out.
Play from 0:36:47 Um, you know, you can do mood boards on Midjourney fairly Easily. Will they be perfect? No. Can you edit them? Yes. So something that even just to get like a unique, you know, thing going, let me see if I can actually like, let's just, let's just do something live here. Let's see if this actually works. So, like right here, right off the bat, right? We just did this little skincare example here. Like here's a mood board. Like maybe this is the aesthetic that you're going for. You know, now we have some color, we have some color palettes here, some color swatches. We have sort of like this look and feel about everything. We have this extreme minimalism and you know, this might be a little bit more up the alley of what you're looking for. I like this one a lot. Like this one to me, pretty cool.
Play from 0:37:26 Cause you can also go in here and, you know, if you need to go in and edit this stuff, you can go into editor and just like wipe out a box and something else. And you know, the other thing that you can do too, is like stacking what we just talked about individually, you could take that swatch or that, you know, gradient, for example, use that as an S ref within the mood board. If you're going for an exact color, Because Midjourney, at least. While we're recording this doesn't have any association with like Pantone or Hex codes, so that's the way that you could sort of you know Do that without having to take a guess at what the right hues are But that's yeah, I think if you stack both of what we just said together would be a great way to That would serve several purposes, that would give you several scenes that you could pull SRS from that
Play from 0:38:14 would give you different ideas for the types of shots serve as inspo, like you mentioned. So I think that could be cool too. I haven't really done that, but I just might try it. It's a good idea. Cause then I like, even something like this, right? Like we can take this and then just play, right. Instead of muted, let's go, you know, we can use our little, permutations here. So let's go with pastel colors. Let's go dark mood, and then we'll go with like ultra clean, And you can just play with the colors here and probably get some really interesting stuff, and then maybe you can start to blend stuff, but this is how I would like attack, even just like from a concepting stage of like what you're trying to get, because you can get a lot of different looks and images out of this.
Play from 0:38:54 Let's see, which one was the ultra clean? I kind of like this a lot. Just from like the way that it, it constructed it from a color. Yeah. Textures. It's got a lot of things going on. I like that other one in that group too. That that last one or the first one in the group? This one. That one. One one looks pretty. That one. That last one. Love the textures. Yeah. Love the textures. Love the shadows. Love the color scheme here. when we go to something that's like a dark mood. It's right. Like it's going to change. Now we have a little bit more contrast, a little bit less like this color scheme is different.
Play from 0:39:26 Looks pretty cool. This one is more in the, you know, here's the pastel or the neon colors. which looks like just about every other brand product at this point, Now with the pastel colors. Yeah, right. And then, you know, even get down into these, uh, something just a little bit more soft, So this is a fun example of how majority can be useful in terms of like ideation for this stuff. And then again, Like drew said, you could take this and then run with it a million different ways, but I love mood boarding on there. You know, that's like my, my typical, there's like a Turing test for, for the image generators for me, that's like what I like to put in with every new model to see, like how much crazy detail can it get?
Play from 0:40:04 Cause it's a lot of concepts in one place. Going back to this, like, I know we've talked about, I'm not sure if it's come up on the podcast, you like to use the mood boards, but also like, we've talked about style codes. That was an interesting thing that you brought up a while back when we were talking about style codes is using the mood board, multi panel. shot approach when you're, unlocking style codes, To get like a different vibe. Cause it's almost like you're generating multiple images in one, without, you know, spending the extra fast hour.
Play from 0:40:33 So I thought that was like brilliant when you mentioned that, like, So if, if you are doing style codes or you are, you haven't thought about the mood board thing, like highly, highly recommended, it's just like a different way to, to get inspo and concept what you're doing. And I think it saves you some time too. Totally. I'm curious, with these mood boards or multi image panels, it seems like aspect ratio comes into play here too. What are the, what are the aspect ratios you like to use the most for these?
Play from 0:41:03 for the mood boards, I typically like four or five. So even if we can show this right now, like I'll just do use this, use this prompt again, four or five. I also like, 2x1, I like 5x4, and then sometimes you get a little bit crazy. You want to go like, you know, 3x1 or 1x2, which can be very interesting in terms of like how it just generates. I like to do this a lot of times, just to like kind of see which ones come out, because they all kind of get structured differently, and a lot of times they end up looking different, So like here, this looks way different than it did in the original prompt because sometimes the mood boards get A little bit, uh, they just like end up rendering different. So would you do like a two by one? A lot of times you'll get more like this scenery style, if that makes sense. less like, like paneled mood board,
Play from 0:41:52 which is what I'm looking for in these like five, four, or these, uh, one by twos. you know, that's where the big difference is. The wider you go, it tends to put things more like in a scene. Now, this is like an interesting mockup that it did here, but really it ends up a lot of times looking like this just to be quite candid, but these other ones, I really liked the longer ones. Cause you can sometimes just get some really cool stuff, the way that it's positioned or just the look and feel of it all. Aspect ratio of portraits probably like trained on single objects a lot more of the time, right?
Play from 0:42:21 Because it's portrait style. You're usually focusing on a person or a product or a thing. And so, you know, it seems like with those four or five or those portrait style aspect ratios, you're getting almost like single objects, Where the focus is on them. Whereas, like you said, with more of like a three, one or something like that, you're getting more of like that scene element. Or even like a 16, nine, probably like a scene element. Liam. I think we crushed that one, bro. I think we crushed that one. So digest that. that was a good, I think, example where we've both got like really good things that we can offer with how we would answer that. So hopefully that gives you something to chew on. Let's go to the next question. Rory, what do you think about, let's throw in this one.
Play from 0:43:02 Which Midjourney feature do you find the most challenging? Character reference, probably, I would say that I would agree. I would agree. I don't, I don't even think we need to like really spend a lot of time going into why, but we do have character reference questions. So maybe this is a good, good segue for that. So we've got a couple character reference questions. So let's just kind of like combine these into one. And, I'm going to defer to you here. Cause you are the character reference guy, Lydia Benzik, probably murdered that last name. I have no idea. Sorry about that.
Play from 0:43:35 When using the C ref function, the character is not always similar to the previous graphic. Do you have a trick for this? And then you've got Dan Phillips who asks, what's your take on achieving product consistency in the context of MJ for industrial design? C ref gets some of the way there, but what else can we do? Right. So that kind of gets into maybe more of that, like character object reference question that we kind of got into in the previous AMA.
0:43:41Body Proportions
Play from 0:44:01 But, you know, I, I know that you have a very. unique and I think thought through way to go about this. It still feels knowing what you're going to talk about. It still feels like it's, it's very a roundabout way, which the journey is going to have to, to clean up over time. But having said that, if you need to do this right now, and this is something that you want to do. You've got the right guy here. Rory's, done quite a few of these. So Let's do this. So I'll just show it as an example, right? Like the idea here you're almost like training a mini L the goal is to get, if you can only put one character reference image in, to me, like, let's just say you have a limited set that you can put in.
Play from 0:44:42 How do we get as many images of that person in as possible that are consistent? So, theory, if it's me, I have a lot of pictures of myself. Right? Like, I can just go and string together a bunch of pictures and use those as character references. For a generated character that might not be the same. So I like to use these character sheets. This works really well for me, personally. Because the idea here is if we're going to attach one image, right? If we're going to touch one image, I'd rather have four different faces and four different views of that one person. It's just more data in for Midjourney to figure out like what you're trying to get out of it versus one single face.
Play from 0:45:18 So typically I'm looking to get like, this is really good for me. Cause like I get different, angles of the face, you know, you have the same consistency of the character in this one image. Same thing, like with something like this, right? Like we have the guy looking up, he's looking straight, he's looking to the side, so like the dimension starts to become easier, Even something like this, like this guy here, you know, we got to look in left, looking forward, eyes closed, different facial expressions. So typically I'm using like a very similar prompt here, character sheet. And then I'll have like the person. What they're wearing, you know, at studio setting, because really, again, going back to what we talked about earlier, don't want to pollute sort of the data set here.
Play from 0:45:56 Just have everything focus on the person, the details are all about the person, not about the environment. Then, that flat background. That's really where that gets into it. And then I have, this is the important part of the prompter image split into four different pictures shot from multiple angles. So the goal is to get the different, like, you're trying to create like almost like a 3d model with still images. Cause you can add. Multiple character sheets, right? Like I can just hit either very subtle on here, or I can rerun this with the same seed again, just to get as much data as possible. Cause the goal is to then translate it.
Play from 0:46:27 So like this again, like this is another good one where I have multiple images of the face from multiple different angles, right? Like that's what I'm looking for. This is a good one where I got multiple shots of her face. I use this image as the C ref. So if I'm going to use one C ref image, you can use multiple character sheets if they all look the same, but I have now have five pictures in one image. So we're hacking around it a little bit. So then it translates really well. Like her face translates very well from here to here. The eyebrows are the same. You see the hair has the, uh, Different coloring between, you know, black and gray, which is good. We have sort of like this same sort of like wispy look to it in the front. That looks like the rest of these images.
Play from 0:47:02 Her hair is tied in a ponytail. You get some of that here. You know, you get it in this one as well. So that one can be really helpful. And you're really just doing, you know, you're really just describing the woman the same way in the prompt as you did in the other one. Now, you know, if you do this, you can get really consistent. Like I think this one tends to be very consistent, right? Like this is pretty damn good across the board with her, but it works in, you know, a number of different ways, right? Like this guy, same thing. You know, we have the same sort of red jacket here, a red shirt with a black hat that translates. Well, this one to me is probably one of the best ones that I've done here, which just is very simple, but it does have her, her eyes in multiple directions and her head from different angles, but it really translates into this image.
Play from 0:47:45 Well, so to me, using character sheets, that's the best way to do it. When you get into multiple characters, it's a hackathon. Oh, Jesus. That's what I, that's what I can't, that came to mind for me. You did that a lot. You did that. Uh, that LinkedIn post, I think I still have it saved too. Like it was just, it was just so like beyond where it's like, man, like you could spend a lot of time doing this, but holy shit, like it's possible.
Play from 0:48:11 Right. And so, you know, Rory did that with a couple characters, then mixed them into the same scene where it was just like, yo. So anyway, you know, like this is where we're at. Possible still, you know, sort of a roundabout way to get there. yes, there are other tools outside of Midjourney that can do this probably better than Midjourney at this point. But again, it's just like, how many tools do you have the capacity to, to learn and want to experiment with? So that's why we always try to anchor it to Midjourney if we can. But again, with Midjourney going into that route of. 3d video, these things have to get better in order for that to happen. So it should be solved.
Play from 0:48:49 That would imagine like V7 is going to be a huge upgrade there with the character consistency too. We'll see. Well, you know, again, we're talking business, right? Like they're losing people there. Like, that's why people are definitely trying to, to go to other platforms. Let's, you know, if we're isolating variables as to why people leave, that's definitely one of them. They want, yeah, existing characters. That's why I think they're hiring this storytelling team or storytelling products that they're about to launch. I keep forgetting what it is. Um, but at the same time, You just, you need the consistent characters. If you want to tell a story, right? Like if you want to do anything with this, not even just professionally, like if I want to go just to have fun and do this, you know, like create my own little animations within video.
Play from 0:49:31 Like I want to have a consistent character. I want to look different every time. So yeah, I have a feeling they'll work on it soon. It just, it can't not, I do too. It's gotta be a priority. Yeah. It's gotta be a priority. It's not one of those things where it's like a nice to have. They definitely need to have that. So keep your eyes peeled for that. Let's you want to do one more question. I think we have time for one more. let's do this one. I think this is another question that I've seen a few times just previously. So I think let's, let's go here. Vera or Vera. I know who exactly who this is.
Play from 0:50:02 She's great. She's fantastic. She comes up with the question of when I create a full body shot, the face is often too small and thus a bit creepy. If I do start with a closeup, then zoom out, I usually get unnatural body proportions, right? For example, legs, thoughts. between you and I, we do have a few things we can point out here. I'll let you go first. Yeah. So sometimes the answers outside of Midjourney, and I am quite honest on some of this, like if you have a good image and the face is small, like I'll just use Magnific to go in and add the details there.
0:50:25Upscaling Methods
Play from 0:50:36 And, you know, if you're not using Magnific a lot of times, you know, you can use any sort of upscaler to get more details. I think Drew would probably go into some of this a little bit more, but there's You know, Magnific, I really love it for that standpoint because it saves a lot of images, especially for me that I'm trying to, you know, produce and put out. It'll go. And if you dial up the fractality a little bit, like maybe one, two, I don't, I don't go three, three gets a little bit crazy, but dial up the fractality a little bit, you will fine tune those details to give the face like some structure and not make it look like nightmare fuel. So that's one thing.
Play from 0:51:09 with this though, the way that I've been doing it and getting much more success is because now when you use the editor and you zoom out, you have like the ability to alter the text prompt as well. So I will start with a closeup of the face and then I will go to either a medium or a, uh, medium closeup or a medium shot, but I'll edit the prompt to say medium shot and then drag it about down to where, you know, the mid chest.
Play from 0:51:34 Should be then the same thing with like, you know, either full or medium, full shot and drag it down to the waist or to, you know, the mid thigh and then going step by step, it's not going to save you any Fast Hours or any sort of brain cells. It's just, if you really like that close up of the face and you want to preserve the details and you want it to be that way, that's the way to do it. Or you're going to have to drag it out into an aspect ratio that fits a full body. Prompt it and generate it and then use the editor again, and then just like keep, you know, working through that. There's no simple fix. I think on this one, that's just like out, you know, magnific to me, simple fix. I go plug it in.
Play from 0:52:15 Yeah, use the, use it and go, but you know, it's working in steps. That's kind of what we've always preached here is one time prompt sometimes or not the, not the answer for everything. Man, you just reminded me of like a couple of really good images I created back in 5, 5. 1, maybe even 5. 2. Which were like in 16, nine, you know, and everything was just, all the body parts were more like, I need to go back and run those through magnific because it was like, the image was fucking amazing.
Play from 0:52:45 If the proportions were right. Yeah. So actually I, I do think that would be, that would be really smart. Um, and I don't really have much to add other than two quick things. Like you touched on aspect ratio. So like, you know, start with a ratio that is complimentary. To a full body shot. if it's more of a scene, do something like a three, two, and I would even say like a three, two versus 16 by nine for faces is a pretty big difference because 16 by nine sometimes can put the face too far back in the scene.
Play from 0:53:16 And then that's where you get the problems with the mush face. still think like with the editor and now the flexibility you have there. Okay. It now is easier to keep things a little bit more proportional, versus just having the, you know, the 1. 5 or two zoom out. that was sort of like a little bit more trial and error, but I also really love that idea that you just came up with, like editing the prompt as you go. Ooh, I like that. That's so good.
Play from 0:53:43 I mean, I do like that. Like the other thing that I would. Express and maybe this rolls into something else I wanted to bring up anyway, is creative upscaler and Midjourney, can fix a face. It can fix a face. So if you've got something that's a little mushy, try creative upscaling. It actually can fix faces and it can fix words too, which is, which is interesting. Now, the wild card with a creative upscaler is you're going to start Take a little bit of the roll of the dice on how it's going to interpret the image.
0:53:55Creative Upscaling
Play from 0:54:15 Right? So it could change some things that you actually like. So it's not perfect by any means. However, I will kind of like dive into a quick example here. And Rory and I were talking about this right before we jumped on, but I hadn't played with creative upscaler in a while. I just got into the habit of like, if I wanted to upscale. I'm just going to do subtle, Because for a long time, it just kind of seemed to like clarify the image. But I do think over time, these upscalers might've evolved a little bit. You know, I think subtle upscale now has a little bit more of the smooth polish to it, the creative upscaler. It almost adds a little bit more tonal contrast here. So let's do like a side by side. Here's a. Air Jordan one shoe, And this is zoomed in like I'm zoomed in already upscaled subtle right here. So it looks pretty good. You know, like you got plenty of detail here.
Play from 0:55:03 but there is like a little bit of smoothness and then obviously like Air Jordan, where it says this, like it did not get those letters, right? which is like, okay, damn, you know, like, am I going to fix that if I really wanted to put that out now, if I creative upscale that same exact image. it's almost got this exactly right and you can see this tone and micro contrast here in all the details. So it's like I think I've been sleeping on, creative upscaler. And so I like I started asking myself these questions. I'm like, okay, like, Is there something to this? So I ran a few different experiments where, it was just like, Hey, ChatGPT. I want you to, I'll even give you the guys the prompt.
Play from 0:55:42 I want you to closely analyze these two images. They're the same image, but we're enhanced with different upscalers. I'd like for you to explain specifically, but simply how each upscaler seems to be working in order to achieve its result and output. Tell me the differences you spot in the two and tell me what you prefer. Okay. And so upscale one, this was the creative, seems to be focused on enhancing sharpness and contrast. It accentuates the outline, stitching details, texture, adding more clarity, the logo is slightly more defined, blah, blah, blah. Two, prioritizes smoothness and a slightly softer appearance. So this was a theme that I noticed with all these images.
Play from 0:56:19 It kept saying the same thing, creative, clarity, and definition, finer details. this one smoother, almost retouched look, And so when we're going into this like photo realism thing, and you don't want that plasticky elastic look, I think we need to be using creative upscale here, folks. Alright, I just, I'm just, I don't know if anybody else is the same way, but like to me, I, I used creative a couple of times. I wasn't impressed. You know, it was just like, it changed some things that I didn't necessarily want it to change, but it's definitely got the use case here, especially when you're talking about people or these finer details, to add in.
Play from 0:56:55 So, Just something I want to point out in addition to your question about fixing faces and such creative upscaler, let's not totally sleep on that. I like that one. Cause I've been using it probably a lot more than ever now with, stuff I'm doing with runway. Cause I want just higher res images. And sometimes I just don't want to burn, you know, magnificent hours. Right. And so, and this is all in context, exactly. If you want to do it in the tool and you're going to do it in the tool.
Play from 0:57:22 to me, they've gotten better, like I, I, that's just me personally, I've noticed even some of the subtle ones have just been getting way more pop than what I used to get, which was like you said, the polish, the softness, the retouchness, which I just don't want. That's why, you know, typically to be candid, that's why I use Magnific because it keeps and adds a lot of like fine grain detail. And you know, with this, now with creative, I've gotten some really cool, Stuff and it's a lot less, uh, you know, it's a lot faster than some of the other upscales up there.
Play from 0:57:50 So it's nice. I like, I like it, you know, I try to run, I don't know, does everyone do this? I, I run both every time basically at this point. Do you really? I just click both 'cause I'm just like, I don't know what I'm gonna, I don't know what I'm gonna like. So it's like a, it's like a style code in a sense where I'm just like, let's just pull the slot machine, see which one ends up better. And, I tend to go with the, I tend to go with the creative one more, I think. And then the subtle one though, sometimes just really nails it. It depends on what you're prompting though. I agree.
Play from 0:58:18 Right. Like, so, you know, if this is something that you haven't played. Around with deeply try both. I mean, there are definitely use cases where you're still going to want to use subtle, you know, and then definitely use case where you want to use creative. And I think that's the point. But the wild card with creative and why I originally stayed away from it was the fact that sometimes it is going to change some details that you may, that you may like, right? So you might lose a little bit of that control. So just worth noting, man, we, we ran through the questions, anything else you want to touch on before we.
Play from 0:58:49 Sign off man. I feel like this flew by and I feel sad because we're not gonna get to do this for another week, bro Yeah, what the hell? I don't like that. I don't like that thought. Yeah, there's a lot I mean, there's probably a lot going on You know, there's there's some crazy stuff that happened in the last week with everything AI. I mean, you got, you know, how about how about sound the alarm bells? We got James Cameron, you know, joining a advisory board. Yeah, stable diffusion. I mean, if you weren't following the trend, if everyone's calling it that now, I mean, what are you doing? Like, that's you want to talk about somebody like the person that's been the most visually ahead of his time in terms of like pushing boundaries and also, you know, for avatar too, I think they had to wait like five years because the technology didn't exist that he wanted to shoot certain scenes.
0:58:58Industry Developments
Play from 0:59:36 So if you're starting to see a little bit of investment there, I mean, you start to look at it. It was a curious case for me. That's like, it was stability AI versus some of the other AI tools that are out there, like, you know, you see runway Luma, which is interesting because like stable video has kind of just been like an afterthought for a while now. So I think that is just something worth like at least touching on because that to me, it just, it just signals like that's what, that's what's going on. Like people are interested in that they are, you know, really, if the, you know, if Hollywood, if you didn't think Hollywood was, somewhat interested in this, then you don't see someone like James Cameron joining a board of advisors.
Play from 1:00:14 So it might be a small detail, might be nothing, there was like this really dope quote about him in the article too. It's just like something along the lines of James, cameron doesn't think about the future. He lives in the future or something. I was just like yo put that on a t shirt I mean what i'm stealing that you know, like and I don't even think that was exactly it The the actual quote was even better than that. I was just like damn. Okay, that was good You I mean, because you see some of the stuff that he did with avatar and we can almost do it now, right? I'm not guys. Yeah, I'm not saying we can do avatar quality But the same theory is there like the video to video or the animated diff, right? Like that's basically how you put motion capture on someone's face on or on their body You turn them into renderings, you reskin them, that's kind of what's going on with video to video. I've seen a lot of cool shit
Play from 1:01:02 people are doing now on, you know, mostly Twitter, where you have 3D artists that are making just these, you know, renderings, not necessarily polishing them and putting them all, you know, full production value, but taking like a baseline model rendering, putting it in the runway, then generating a scene like a, you know, skin and scene around it, which is Kind of wild they have to be thinking about stuff like this because James Cameron, you know, it cost him how many? Does it cost him like a billion dollars to make avatar or something like that? So if he can slot in a couple of cost savings, I'm sure the production studios would be happy. And that was a while ago. That billion dollars was a while ago. I wonder what that is today.
Play from 1:01:40 Yeah. They're not saying we want to like get rid of every, you know, don't jump on me three, you know, 3d people in there. I'm not saying get rid of anything. I'm just saying like, there's going to be some creative ways to use the tools that probably the AI companies have not even thought about considering. He's developed a technology. To, to film his movies. I wouldn't be surprised if we see something even radically different from them, like the medium that we see it now, because he gets a mind like that in there, he has all the problems, right?
Play from 1:02:08 That's what I think is interesting. That's right. He has all the problems. It's like, I, this is what happens here. This is what sucks here. I hate doing this. And then it's like, Oh, here's what our tools can do. What's put the minds together. Can it do this? Maybe. Right. Like that's, it's just, it's another tier of perspective and. experience offering than just somebody that, does filmmaking in general, I mean, this guy does Hollywood filmmaking and in very, upper echelon, there are levels to this. So to bring in somebody like that, to your point, I think is fascinating to think about because yeah, he's, he's seeing things at a much different level than other people are even going to I mean, there are blind spots for a lot of people. So yeah, it's fascinating to see what that is. And you know what? Also going back to our guests last week with Chris, he did a,
Play from 1:02:56 LinkedIn, audio, maybe yesterday, but he had somebody on there that's got access to Sora and was talking to him and he's like, Hey, like, so what's the deal? Like is, is it as good, as they originally said, or are you like having all these challenges within it and all that stuff and Are they updating it? And he like, he couldn't say a whole lot with like how they're updating it specifically, but he did say, yeah. Okay. It is actually worth the hype. Like, he's really impressed by what he's been able to accomplish in there. And this guy, I was no slouch either. I don't remember his name, but his background, I mean, he was certainly qualified to be talking about it. So yeah, it was, it was insightful. I still don't know when, when that sort of thing's going to happen.
Play from 1:03:35 So. Who knows, but yeah, Like we anchor this back to Midjourney, but this really is about, creating assets from scratch to tell your own story and whether that's static or dynamic in nature. It all, it all blows into one. So yeah, man, I think this is a good spot to wrap it up, break it off, you know, unfortunate, rip the bandaid off, go back, let's rip it back to work. Yeah. Let's let's do like adult stuff, I guess. Crap. But yeah, it's something we're not qualified for anyway, guys. Thanks for joining us. This was another great one, Rory. I know next week we're going to have a guest on or at least that's the plan and don't miss that one. Because again, like going back to what we did with Chris, I think this is going to provide a whole nother layer.
Play from 1:04:19 Of context and perspective of what's possible. And I know this guy's got plenty of great stories to share as well. So, hope you catch us on the next one. Rory great chat with you, man. Yeah, dude. What's, uh, let's get everything in motion for next week. I'm excited. Perfect, perfect way to segue that, by the way, different perspectives, every guest we're going to have on here is going to be doing something probably radically different at a certain point. We'll also, you know, try to get as many cool and unique and random people as we can. So you guys can just like, you know, let your minds run wild. So yeah, dude. Good job. Yeah, dude. All right, guys.
Play from 1:04:54 Thanks for joining us. We'll catch you in the next one. Like and subscribe. See ya.