Transcript · September 15, 2024 · 14.2K words

Ep. 10: Midjourney AMA: Answering Your Most Burning Questions (Part 1)

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.

0:00:00Episode Introduction

Play from 0:00:24 Hey, you know, with that said, though, I think we're going to make it up to you guys. We've got an awesome episode today. We're doing the AMA. So if you follow us on LinkedIn, for example, you already know that we ask for your questions. What's your most burning question as it relates to Midjourney? What are you stuck with? Is there anything that you don't know about? Is there anything you wish you knew more about? So we're going to, you know, go through those questions one by one. We've aggregated all those questions for you guys. So we're going to tackle those. Before we get to that, we're going to update you guys on some of the office hour stuff.

Play from 0:00:52 Since we've been out of the loop for the last two weeks, make sure you guys are on the cutting edge of what's happening over there. But first things first, let's catch up with life. Uh, how was, how was the trip, man? So, you know, tell us about the, the international trip over to Budapest, your speaking engagement, how'd everything go? I'd say the speech was like. The highlight of it. And the thing that went the most smoothly, which is really the opposite of how I expected it had a couple of little travel snafus KLM.

0:01:08Budapest Trip Recap

Play from 0:01:20 I'm shot. I'm shouting you out here. Cause you guys screwed up a few times, but, um, lost my bags on the way from Amsterdam to Budapest where the same clothes for a couple of days, got on stage for rehearsal, realized that, um, you know, they told me I could do my presentation in Figma so I could have that format because I really needed some firepower to design this whole thing. Turns out, that wasn't the case one day before the presentation, so I had to reformat 83 pages into PowerPoint over the next like five hours. Did that. Got up on stage. You know, typically when you get up on stage, number one, it's fucking, it's a total mindfuck to have couple thousand eyes looking at you, and then also just, you know, to be up elevated.

Play from 0:02:04 It's a weird, it's a weird feeling. Right? You do hit like a flow state, but at the same time, I like a little bit, you know, I like to have, when you get up there, typically there's like a, teleprompter in front of you that has your current slide, your notes, and then the next slide. So, you know, what's going on for whatever reason, the notes weren't there and the next slide wasn't there. So as soon as I got up on stage, I'm the first one to go, two, 3000 people looking at me and there's no notes. And I'm like, well, we're just, we're just going to let this one ride. Luckily I was, I was feeling pretty good that day and it rehearsed a few times. So it just kind of flew through it.

Play from 0:02:38 You had that juice. Dude, befuddled looks on people's faces was cool. It's also awesome. This is a very Small detail. I don't know if anyone's spoken on stage before, but when the entire audience like lifts their phone to take a picture of slides, it makes you feel really good. It makes you like a confidence boost just like starts to keep going and going and going. But, um, yeah, that was awesome, man. The cities were cool. Amsterdam was outstanding. Had a really good time there. Budapest also super damn cool. Uh, love the people, not overly crowded, really big opulent city and food. Incredible. I was obsessed with the food. I could not get enough. I was, you know, if you can get me to eat stew.

Play from 0:03:19 95 degree weather or like stewed type food. So we're talking, you know, chicken pop ash or any of like the goulash, things like that. It's 95 degrees and I'm eating soup. Like this is the best thing I've had in a long time. So it was, uh, it was awesome, man. And then, uh, they lost our bags on the flight back, had a really short connection. Had to sprint through the airport, like about a mile and a half in Amsterdam with a 40 pound bag on my back. Got on the plane right as the doors are closing, you know, sweating, disgusting, pulled back. Love that. Yeah. So big, big recap there, but, um, presentation was a hit. So I was killer. Hey, at least you were, you were running without the kid.

Play from 0:03:57 So that's cool. Right. You weren't running with a baby in your arms, so that's important. Yeah. Also, I, did you, did you say the word opulent? Did you like that one? I do like that one. I feel like we need to break away a separate podcast just for our vocabulary, you know, we're pretty good at that though. How to pronounce words, how to define words, you know, leveling you up with your dialogue and daily conversation.

Play from 0:04:22 It just seems like a no brainer, but I couldn't figure out a better word for it than that. When I was there, it was like the first word that came to my mind, you know, really big, like archways, statues, everything just seemed grand. And I was like, Budapest, man, and it's everything. You know what? I love the USA. It's an old city. Yeah. Can we get, can we get our streets as clean as any street in Europe? I mean, like not even a cigarette, but anywhere I'm from New York guys. So. I'm used to trash and stuff everywhere. But like, seriously, going to those cities and just seeing how clean it is.

Play from 0:04:53 It's insane. Go to the red light district at Amsterdam, watching the chaos unfold, seeing all the shit on the ground, walking through the next morning, all gone, all clean. I'm like, what the hell? It's like, how is this possible? that's my complaint for the day. That's all. That's all I got. It's great to have you back. I missed our weekly get together last week. I mean, that keeps me a little bit sane. So I've just been over here. You know, chomping at the bit to have some sort of conversation about Midjourney for the last 13, 14 days.

Play from 0:05:22 So engine just like, yeah, yeah, man. So I'm glad we're here. I mean, we've got, you know, a ton to talk through today. We've got, we've got a massive amount of questions to get to, to first though, let's go through the office hour updates. as usual, Rory, I'll just kind of talk through these piece by piece. Feel free to jump in where you want. We can have a little bit of a discussion about it. so first things first personalization. They're focusing on increasing adoption, uh, of the basic personalization system that's already in place right now. For some reason, still not a lot of people using that. I would guess that maybe has something to do with, with newer users Maybe they either don't know about it or they haven't gone through the process of ranking, but that's been on their agenda to try to drive that adoption and usage of personalization

0:05:55Office Hours Updates

Play from 0:06:07 and Midjourney. I don't know why you wouldn't at this point, but that's my speculation. So they're planning to make personalization, a default feature in some modes here shortly. They're also developing a new batch, they're calling batch eight. So instead of your four image grid that you get back with a prompt, getting an eight image grid back, I don't know if that's replacing it, All together, just like a full swap out. Or if that's just an option, my guess, it would be an option because some people are stuck in their ways with the four, I can still see some value in that.

Play from 0:06:39 Uh, but that sounds cool. Would obviously love to have more images in the fold immediately, cause sometimes I don't know about you, but I find myself kind of. running the same prompt two or three times on purpose. You know, like maybe even throw in like a slight number of chaos just so I can get a full spectrum of what I could expect back. Yeah. jumping back into personalization real quick, just a real, just a quick question. Do you use it on every image or do you, is that like your standard like set now or do you go and, uh, you know, kind of flip it on, flip it off as you need. It's default now. Yeah.

Play from 0:07:10 All right. I mean, I I've, I've run a couple where I've turned it off just to kind of see. And I don't know if it's just because I only did that a handful of times, but I wasn't impressed with what I got back versus what I got in personalization. So I was just like, ah, okay. Like not so weird. I mean, so we're just, I did it the other day. I turned, I flipped it off and was like, this can't, this feels weird. Like it doesn't feel right. You know, like it felt like it was the majority six is better. Or 6. 1 is better than this, but it was just when I threw my it's just because it's, it's biased, right? Like my personalized code. So I'm just like, yeah, that's exactly how I want it. But it's interesting. I didn't realize that they would have less adoption.

Play from 0:07:45 Number one on that. I thought people would be amped and utilizing that like pretty consistently. So that's kind of shocking. But, it's always, they always like to adopt, you know, get more users on their new features, which makes sense. Hey, listen, guys, if you are not using personalization and you're listening to this, Definitely spend some time on that. So how you get to use it is essentially ranking images. Again, there are two ways to do that. That's by going through the ranking system. So all this can be found on their website. it's under tasks. you can, you can basically do an a or B, you know, which image looks better, but you can also skip, right?

Play from 0:08:18 So if neither of those images appeal to you, skip it, be very specific or methodical or strategic about the way that you're ranking them. The other part is the explore page. The images that you like on the explore page also factor in I don't know if those are weighted the same not sure but that's how you get it and That personalization is just so much better than the the default at this point plus we talked about this I think a little bit on the last episode but with personalization in the fold that also affects stylization Yeah, right and so You think about if you don't have personalization on your defaulting to Midjourneys aesthetic when you dial that up, Or when you, you know, you take that down and it sort of takes you off path from your prompt in the text. So I think it's just, you need to play

Play from 0:09:03 around with it and define that because that's where the future is going with Midjourney. And one of those things that kind of makes it a little bit more unique over the other tools. Yeah, I've been really a fan of it. I didn't think I'd necessarily, I just didn't know what to expect and now I use it on every image, but sometimes I will say, I will say though, if I have one gripe with it is that sometimes it does stray from my text prompt a lot. because it's like trying to do too much almost like my, maybe it's just my personal style, but when I have a more detailed prompt with more detailed things and I use personalization on it.

Play from 0:09:35 It's like, okay, I can see where some of the faults are, but, you know, that being said, just for standard image generation, it just firepower is next level for at least for my code. That's just my personal opinion on it. Yeah, I've heard people that don't like theirs. They're like, I hate mine. I think there are just some like pessimistic people out there. Yeah, maybe, you know, like maybe. Okay. So same with what we're going to talk about next, which is image editor. You know, like some people were, were shitting on that. I don't, I don't, I don't really see that at all. I mean, to me, it's gotta be top two most powerful tool. I think a Midjourney right now.

0:10:10Image Editor Features

Play from 0:10:10 Crazy good. I mean, it's so much, it's so damn good. And then, so, so here's the update there, right? They're testing a new editor mode for external images. This is another question that's been asked over and over again. Can, can you take a, an image outside of Midjourney and edit it? No, but now you're going to be able to, when is that going to happen? I'm not sure, but obviously with people bringing in outside images, their moderation is going to have to be a little bit different as well. They're probably going to have to, you know, have a little bit more of a heavy moderation, as it relates to those images too. I just think it can be the way that David put it in the last office hours.

Play from 0:10:49 It can be used for good. It can be used for evil. I think that's the great way to sum it up. So. that's exciting. That'll, that'll obviously change a lot of maybe the way that people think about images in Midjourney, because that adds a whole new element of how you can go about creating something. So that's fascinating, but, I'm not sure how much I'm going to use that yet. What about you? I could see where something, it would be. It would be useful. I really don't have a use for it to be quite honest just yet, maybe I will. I mean, if I could, I take that back.

Play from 0:11:17 If I could put a PNG of a product photo in there and then generate around it and it actually worked, that'd be cool. Cause that'd be, you know, for advertising purposes, for digital marketing landscape, like, yeah, there you go. That would work for me if I PNGs and then generating background. Now that's cool, but you know, I don't know if that's the way that it's going to function because another, thing that you could now take into Midjourney from Photoshop, Yep. adds another element to that. I think that's a good, good point. That's a good use case. Also, I think because we haven't had it, we haven't really thought about the use cases too, right? Just like with anything else, these new features roll out. And then once we dig into them, we're like, Oh, you could actually use it for this, this, and this.

Play from 0:11:54 It's interesting. I still think what would be really cool is if they had the, smart select feature, I'm not going to use that in my everyday stuff, but if I put a picture of myself And just wanted to like outline myself, delete the background and then regenerate something in the background or just highlight the background and generate from there. That'd be useful. I'm not going to say that I'm going to use it every day, but you know, something along those lines. Cause if anyone's tried to do this in Photoshop pre smart select, you know, it's painful, it takes forever. The doing a thousand little pinpoints around yourself and it takes about 40 minutes and painful.

Play from 0:12:29 Just thinking about it, but hair is the worst thing. Ever. If you try to do a woman's hair, and there's like, you know, like, a couple different strands, like, hanging off the side, and you have to go, like, in, out, in, out, in, out, in, you, I mean, guys. Well, there is a better, there is a better way to do that. There is, I'm sure there is, and I'm forgetting at this point, but I just remember those days. 10, 15 years ago, just like kill me. But at the same time, like back then it was like, it was so cool that you had the ability to do it.

Play from 0:12:58 Yeah. That you didn't care. Right. Just kind of like what we're where we're at with Midjourney and a lot of these things, because we were so early, we knew they weren't perfect. Something still aren't perfect, but it's like the trade off is, you know that you're a part of something that's going to keep improving and it's something bigger. Right. And it's going to fundamentally change the way that we go about doing things. So I think that's, The trade off you get, you know, with some of these tools in the, in the early stages. I am with you 100 percent on that. We put our eggs in this basket. I'm pretty happy about it so far.

Play from 0:13:28 All right. Well, Hey, let's go to the, let's go to the video model stuff. So. Improving that weekly likely to release before the year, the year's out. Three months left, almost three and a half. So soon we've also, been told that it's compatible with both Niji and non Niji styles. It's going to be more effective with illustrative styles and photographic ones, interesting. and then they're aiming for a lower cost than upscaling, but more expensive than static images.

Play from 0:13:56 Okay, that's interesting. I mean, it makes sense though, like, uh, whoever here has played around with runway, luma, whatever the illustrated style or like the 3D Pixar kind of style stuff always tends to animate much better for whatever reason. I think it's just it doesn't have to. We're just not as used to looking at. You know something like that and picking out the inconsistencies like when you look at a human and you see their facial movements and you're like that doesn't look like a human because we look at faces all day recognize a billion points of interaction in a face like something crazy like that so that makes sense to me I'm sure it'll improve though as illustrative to you do have more.

Play from 0:14:30 liberty, For inaccuracies because it is illustrated too. So, yeah, I mean, you see, you've seen anime and stuff like that, where like the hair changes, like almost like every frame, like it's just like moving and shifting and like, but you don't care because it's not, you know, you're not trying to make it look real, right? Exactly. So whenever they drop it, you know, we're going to go nuts. And then guys, we'll be here emergency podcast. Like let's, you know, maybe we'll just do like a, maybe we'll just do like a 24 hour live stream where we're just, we open it up to everybody and we're just like a bunch of junkies going through and trying to break the damn thing.

Play from 0:15:04 I don't know who says, yay. I I'm in, but like bloodshot eyes, like just, you know, what? Three, three months before end of the year. Still could possibly come out before Sora. Yeah. I mean, it'll probably, they'll probably drop it on Christmas Eve because that's what they do. Um, you know, we'll all just be trying to, to hang out, chill with the fam. They'll be like, we dropped video and I'll be like, yeah, guys, guys, I gotta go. So it's, so it's either, it's either my family or Midjourney at this point. So yeah, it's kind of both. They're kind of like integrated now. And also the family has to just, you know, they know, they know what the deal is. No, they know what the deal is.

0:15:15Video Model Discussion

Play from 0:15:42 3d model. the goal there is to convert Midjourney images to 3d. We'll allow camera movement within certain limits, targeting 60 frames per second rendering. And they're considering a camera path system for pre rendered video output. Oh, okay. That's interesting. I don't know how they fit all this stuff in, but I'm, if they do it all in the next three months, I'm impressed. Cause it's a lot that even just like these small updates that seem, I don't know. It's like, Oh, it's just like a minor update on their end, like working on it. For real, if you're gonna try to do version 7, video, 3D, get this all out in 3 months, call me impressed. Cause Sora, we still can't see you, and it's been 9, 10, 11.

Play from 0:16:27 When, when, when was that announcement by Sora? That was like February, March. Dumbest thing ever. Really the dumbest thing ever. I get a lot of people that are like, can you use Zora? It's like, no, no, we can't because they, they just like, you know, pump faked us on being able to use it and then disappeared for nine months. So, oh, it's the election. I don't know. I know how many times I generate and runway and for a 10 second video, I might get nine seconds that are awesome in one second. That sucks. If you're trying to generate a 60 second video, I can imagine that it's not as consistent as they either thought it was going to be or something like, cause.

Play from 0:16:59 Look, that's a lot of frames. That's a lot of things to get right. we know what we seen in those, in those trailer clips, and we all know that that's not the first generation of the first time they ran those prompts and it just pumped out like that. So I just got a feeling there's a little bit more under the hood that, you know, they, they, they jumped the gun maybe a little bit or. They just got really excited with a couple things and wanted to, you know, boost some market share. But who knows? I mean, I, that 3d system is just so fascinating to think about because this is going to be something that lets you essentially enter an image and it's not like anything that we've seen so far with, I think some of the other stuff that exists out there.

Play from 0:17:40 this is a brand new format, So I think it'll be similar in my mind to like, uh, Like a nerf, like a neural radial radiance field. I think it's, is the name of it. Um, or gaussian splatter. Yeah, I can't, yeah, I think that's it. but nerfs essentially are a technique used to generate photorealistic 3d scenes. but I think this is the way that this is being explained and described. This is, this is going to be a parallel, but different format altogether.

Play from 0:18:11 And that goes back to what we were saying, I think, before, which is, Midjourney is kind of in its own category. In some ways, this is going to be the exact same thing. And again, Midjourney is in its own category because yes, they do AI image generation. And to me, they're still king of AI image generation, but that's never been the end goal. Right. So, so you're viewing Midjourney is just this, but we've already been exposed to parts of the roadmap that will tell us that they're not just going to be a static AI image generation tool.

Play from 0:18:44 Right. So are, is flux going to turn into a video model, right? Like are, are these other tools going to turn that way? Right. Because you've already got incumbents there. Luma runway cling. Uh, I don't know if you've tried this mini max yet. Uh, I haven't had a shot that was dropped while I was away, I think. So, I mean, this, this stuff's going to get, pretty competitive very, very quickly here. and then I think also our minds are just going to be blown with, with what this is going to, this is going to be again, one of those moments and inflection points, just like we had when we first experienced midger and we're just like, holy shit.

0:19:153D Model Plans

Play from 0:19:20 Holy shit. What do we do? You know, it's nuts. it's all has to go along the same track, right? I mean, it's almost like the same path of how this stuff developed in the physical world, Where it started with, okay, illustrations, right? Then we went to photos. Like you can actually take a photo sick. Then it was like, Oh, wait, we'll make those photos move. That's, you know, then we get our, our film. It's like, Oh, let's go into 3d now. Now we can actually go into the, you know, into an image and like manipulate things within that. Then we get 40, right? And that's like your Apple vision pro. So. I have a feeling it's just gonna be the same way.

Play from 0:19:51 It's just gonna be, you know, start with our illustrations, as the generators were great at, then it was photorealistic, then it's movies, then it's 3D, then we're gonna go probably into 4D, 5D, whatever the hell that even means. But, you know, probably put the Apple Vision Pro on or some sort of headset or hardware. Go in and just start, you know, creating like we're in Minecraft or something. Okay. Did you see this? Cause I know you were traveling as well, but did you see the, I can't remember if this was like a bar or a restaurant or if it's like a place, but essentially they have this experience now where you can watch like they were watching a football game and it's almost like you're in the stadium.

Play from 0:20:27 Oh yeah. So, so like that experience is brand new because you're thinking, okay, like this is kind of the experience I would expect. From like a VR headset. But what if you actually didn't need to wear anything to get that experience? You know what I mean? Like, so, so that is a whole new sort of dynamic as well. And so you're starting to kind of like see. Some of these things coming together, I guess, like, that's my point is this is going to absolutely blow our minds.

Play from 0:20:55 I think in the next year, two years to see where this goes. How about, how about this? What about if instead of like VR, we can do like, I don't want to say AR, but in a sense, like where they're talking about hardware, who knows what that hardware is and what it means. I'm thinking like projection, like those holograms and stuff like you ever seen the projections of like Tupac, like performing like a concert or something like that, it'd be sick if their little hardware thing could just like project whatever you wanted, like into, you know, like into a room or like on a wall.

Play from 0:21:24 To me, that'd be, I'd hell yeah. Let me just create in like real time. If that's going to be 3d too. Now I'm really pushing. the envelope there don't even know how that's possible. Don't know how it made. I'm just saying weird stuff guys. Cause whatever. I don't think anything is weird or out of the realm of possibility at this point. Yeah. Saying, saying the things that we've seen come in the last two, three years, but I could be wrong. We'll see. Is VR going to take off? Is VR ever going to take off? Like, do you know what I mean? I feel like it's had like its moment for, Like 10 years and still kind of not like adopted personally I don't really want to live in that world, but I could see how freaking cool it would be to play around in it You know, yeah, I don't know man.

Play from 0:22:05 That's a great question It really is because like on one hand it was is it just like it was early or? Is maybe the medium of it different, and it still happens. Meaning like maybe the idea was that you had to have this headset on, for example, but maybe there's a different way that, that, that comes to fruition, it's still sort of that experience, but maybe you don't have to wear a headset. Maybe it's something else that happens. Maybe it's hardware. Maybe it's shit. Maybe it's a contact lens, you know, like, I don't know what it is, but like, it could be totally different.

0:22:32VR and AR Future

Play from 0:22:37 Yeah. Cause that's where I think, you know, me personally, I don't want to put the headset on and sit in my apartment all day. Like that just seems, that just seems like not fun. So I understand why people are probably hesitant towards it, but it does seem like that's the net like spatial is like the, it's like where it goes, like we're still in this 2d screen now, right? Like we're still in these 2d phones, 2d laptops, like does that become interactive at some point? Does this, do I no longer need a 2d screen? Computer or a laptop and I can just like, you know, project my screen up here and start doing whatever I want another thought on 3d, there are a lot of people that, that may listen to this, that are in 3d, like they, they know that's a moat. That exists still right now where it's like you you need to be you put in hours thousands of hours You know

Play from 0:23:23 hundreds thousands of hours to really be good at something like that, right? I mean somebody would really have to spend a lot of time doing it and i'm just saying this because i've never I've never really done anything with it, but i'm just imagining that's the case and all of a sudden AI is now coming for 3d, the democratization of 3d, right? Again, another thing that, you know, just kind of like video to that extent of, unless you were a cinematographer or producer or anything, like you just didn't have tools or resources to do stuff like that.

Play from 0:23:53 right. It's crazy. It's going to be wild. It's I mean, 3d is just, it's so much work, That role in general, just any role within the 3d environment. It's just crazy, a crazy amount of work. And I understand why they probably. You know, hate some of this stuff because I mean, I, I totally get that. Um, I would probably hate it too, but probably, I think, I think the, the healthy thing is embraced, like be proactive enough to embrace it, to know what it is, how you use it, and if it can integrate into your workflow, Because it's, it's almost better to befriend it than it is to completely enemize it, we've seen this take place with, these grumpy guys, on LinkedIn and Twitter, or X, you know, talking about the good old days of photography and videography and this and that and the other.

Play from 0:24:42 And I get it, But I, I see that as sort of being the unhealthy option of something. Inevitably taking place. So you might as well be on the forefront of it. Yeah. It's like self fulfilling prophecy at this point. Like for me, that's what it is. It's like, if you're figuring it, if you're just keep saying you're not going to do it and AI is going to take your job probably, but you're just like, refuse to do it because of pride. Like that to me seems like a self fulfilling prophecy. Like I saw the writing on the wall with some of this stuff and I was like, I got to learn it or else. Screw it. Like I'm done. So, you know, there's, there's a time to take action. There's time to, to sit and wait. I, you know, I felt early on this was time to take action on this because it's not, it, this isn't like we're, we're not suddenly going to get moral and say, you know, every big corporation in the world's going to be like, we're not using AI because we're better than that.

Play from 0:25:26 BS. Like everyone's going to do it. Everyone's going to try to cut costs. We know how, you know, big massive corporations were. Yeah. Yeah. Like it's, it is what it is. Like, I hate to live like that, but that's a great point. Because, you know, here's the other deal, right? These bigger companies are, are really dragging their feet only because of, they want to see how this plays out with regulations. They can't take some of these risks that smaller companies or individuals can take in some ways.

Play from 0:25:53 So. When that dam breaks and it's going to break, then you're going to see that full on, capacity. And I think that's really, what's going to open or maybe potentially change the mind of those people just because that's, that's typically the way it happens. They need to see The big guys do it. the other problem with that is to with these with the larger companies, especially the ones that I've talked to, right? It's not just about getting like, you know, getting this stuff in the employees hands. It's like figuring out the entire operation. It's totally different for a team of two people like you or me to figure out an entire operating plan based on AI.

Play from 0:26:26 We could do a report. you know, Microsoft, or, you know, let's just say someone like Coca Cola. I mean, God, how many thousands of employees do they have? Think about how many accounts they'd have to set up, how many different operations that have to set up the guardrails that have to put in place. Like it's almost to me, it's the best of the first time small businesses ever had a leg up to where we could get something and like really blow it out before they even have a chance. I mean, they have access to things. Could they do it at full scale with every employee? really dialed up on into it. Like, I don't think so. So, and maybe, and maybe certain companies stay away to a degree of, they're just going to hire an agency or person to do it, Because then it's not, they're not so much on the hook for some of those responsibilities.

0:26:43Hardware Projects

Play from 0:27:09 It's not going to take as long, you know, to build a program internally. Hire the right people, put those operate operationalizations operations in place. but you know what, you know what I mean? Like, so, so I think that's also another option and route that they could go, we'll see, Hey, real quick, so let's run through the rest of these updates real quick. So we can get through to the questions, Midjourney V seven. All you need to know is it's coming before the end of the year, more than a month, probably two months. Going to happen in Q4, explore page. They're making some improvements there to the algorithm, diversity of images, yada, yada, yada.

Play from 0:27:47 Storytelling tool. this is something that we don't know a lot about, but Midjourney does have a storytelling team that they've hired. This could also maybe be part of these secret projects that they got going on kind of thing. I don't know that for sure, but we haven't really gotten any clarity being on these office hour calls with what the storytelling team is responsible for. Does that mean marketing? I don't. Don't know, Don't mark it now. Don't mark it now that you've gotten this far. Don't do it. Don't even try it's better, better this way. Keep us, keep us guessing, even though I want more info. Yeah, me too.

Play from 0:28:23 the other thing is the hardware projects. So they've started hiring people for hardware roles. that started a little bit while ago, but they've started to put more emphasis in bringing this up because they have multiple projects in development. More than two is what they say. They're going to make an announcement with that. By the end of the year, that's probably also part of these, secret projects, That we just don't know about when I'm thinking about announcements that are going to happen at the end of the year. Also, demographic and aesthetic preference study. They're doing another study there with the information that's come in through the past survey, as well as putting together a future survey around, continuous psychometric.

Play from 0:29:03 Data. So they're trying to understand the connections between personal values and aesthetic preferences, similar to the last one. But I think it sounds like it's just going to be an iteration of that to continue to build on. maybe the only other thing I'll just mention. They did launch a Niji personalization feature. So if that's something that you use or interested in, you need to rank 200 images to have a personalization code on that. I spent some time doing that.

Play from 0:29:28 I will say I don't use Niji that much, but there is a specific look when I think Niji that I do like, and so to go through and rank those images, it did take me a while because I skipped a bunch. I skipped a bunch of them. And so, you know. Probably give yourself maybe a good 20, 30 minutes or break that up. If you're going to go about it that way as well. But I think that's it. Do you have anything you want to cap off before we pivot to, uh, to the AMA questions?

Play from 0:29:58 No, I think we can get right into those bad boys. All right. Cue the cue, the awesome music 📍 now, Okay. let's, let's go ahead and get right into it. I'll go ahead and read these questions out loud. Rory, you and I can just talk about, Hey, do you want to take this? I'll take this. Maybe there's somewhere we both want to jump in, but in the effort of trying to get through as many as we can, let's just kind of see where they land here. So first question for you, other than language data references in trying to get wacky or more fun results, are there any combos of prompts plus settings, or I guess maybe even parameters you have that you just can't Just keep getting amazing results with and this is a question from Sherry Horowitz.

0:30:00AMA Time

Play from 0:30:38 Thanks for the question, Sherry. Awesome question. I really like using a Extreme style code. So style code that's maybe like super unique, not like someone's got vibrant colors and textures and maybe on the photo realistic side, not like very simple, not a minimalistic one, like a maximalist kind of one. And then I'll utilize that with style weight 50. So just give it like some pop maybe between style weight, 20 and 50. Add a little bit of chaos in there, maybe like chaos 10. And then I will also throw in potentially depending on if it needs anything else. stylize of maybe like 200 to 400 there.

Play from 0:31:16 So it's like a, I wouldn't say it's a recipe, but it's something I find myself doing often. It's like, Oh, let me take some of this style code so I can apply it. So it gives it some little pop. Then I'll take some of the, you know, style weight, or dial the style weight down, so I get like just a little bit of it. Chaos, so it varies it, and then stylize to kind of like pump up the creativity level on it. So, that's, that's a weird one I've been playing around with. What about you? You definitely have to have some good ones like that.

Play from 0:31:44 Yeah, I do. I feel like I could give you a lot of different answers here. I'm just not great at recall. I think the first thing that I'm I'm thinking about is just low chaos values in general. I love like using just chaos under five because it just diversifies the grid. enough, I guess to me to get to the end result faster. I think a lot of the things that make my results more fun or wacky is just doing things like that or running permutations.

Play from 0:32:14 You know, like I still think permutations are So underrated, you know, and so, like, you can really, I guess, get the wacky or fun stuff by running several variations of any key parameter or any key word or anything like that. Right? So to me, it's not only an efficient tool to get to the right images faster, but I think it really just kind of Scratches the itch of like, did I get everything I wanted to get out of this?

Play from 0:32:41 did I see different stylization values on this? Did I see this prompt in different aspect ratios? Did I see, you know, this type of prompt with different colors or a different subject or a different style in the mix? You know, so there's so much variability and flexibility with permutations that I just think that's such a, it's maybe not a super sexy one, but man, it's so good. You know, so those two come to mind. Maybe one other one is, you know, just combining image weight and style weight. You know, there are times where I'll do both, right. Where it's like, I want to take a certain weight of the image and then I want to take a certain weight of the style, And then finding that balance.

0:33:10Creative Techniques

Play from 0:33:19 And again, using permutations with both of those, because you don't know how much of the, you know, the image that you want to take, and you don't necessarily know how much of the style you want to take. So you have to run several iterations. To land at the sweet spot and without me really like getting in the weeds or leaning over somebody's shoulder about how they go about prompting I just don't know that a lot of people spend that much time doing it, The other one I'll say it's multi prompting and it really depends on what you want Like if you want something if you want like supreme control I'm not using multi prompting for that.

Play from 0:33:50 I know people think it works that, you know, great that way, but personally, I don't think it does. That's just my opinion. Um, if you want control like supreme control, to me it's within the text prompting and the parameters. But if you want something super creative that you just can't, envision, naturally, I think, multi prompting is the way to go. Because multi prompting gets some of the most unique results. That I've seen it not by, not by token, but by breaking the prompt up into two or three different sections. I was going to say those differently, Rory, for those that don't use multi prompting or don't know about it, quickly explain that to them.

Play from 0:34:23 essentially, multi prompting would be like, let's just say your prompt was spaceship. You think about just spaceship as one singular word. And if you generate that, it'll give you something that looks like a NASA rocket or, you know, like a UFO, whatever. If you wanted to separate space and ship with a multi prompt, it would be space, and then, or like, the word space, colon, colon, and then ship. So it'll separate those two things so you'll get space as an idea. And ship is an idea. So it might be like a sailboat or something like that, and then it'll blend it together.

Play from 0:34:55 So it almost essentially like generate two images and then blend it. That's at least how I conceive it. But when I'm doing something like, documentary photo of a spiritual healer in Nepal, you know, with red and Ruby or orange and Ruby color scheme, you know, Things like that, I might multi prompt that out to 70mm, documentary photo of a spiritual healer, colon, colon, then, red and orange color scheme. Right. So it's going to generate those two things apart. Cause I might get like this really crazy pop of red and orange now that it's separated out and it's like very heavily weighted as if it was on one individual generation blended with that image, instead of it just being like a part of the cog and it might be a subtle piece.

Play from 0:35:38 So then you can also dial it up because then you can wait it. So the same way you can wait style codes, you can wait your prompts like that. So then it could be, maybe I want, you know, three to one on the red and orange and I might get something totally wild, or I might want more of the, Documentary photo or I might want more of the spiritual healer. You can go and do this. like again There's no end to it But to me, when I've, when I work with shorter prompts with bigger, ideas, that's how I get the most creative stuff.

Play from 0:36:04 A, that is a cool one though. I, I like that you brought up multi prompting. The other thing that you made me think of too, that I like doing from time to time is taking just two images and blending 'em straight up. Yeah. Like you take a material or a texture and you take something else. Let's just call it like a character or a subject, and you just blend that. I mean, that's how you create some unique stuff that isn't hard to do. You don't even need to text prompt, right? You can just throw those things together.

Play from 0:36:30 You can obviously text on top of it. Mm-Hmm. , but. To me, that's like a super easy way to get creative and create unique images, right? Because you can take things that just didn't exist, things that don't belong together and put them together. I love that. I mean, blending is the forgotten art of, uh, you know, what, what used to be on discord. Like, Oh yeah, we do this all the time. I feel like people don't do it as much anymore. Great for like juxtapositioning certain things like just, is it juxtapositioning? I think you got juxtaposing. juxtaposing.

Play from 0:37:02 Is it juxtaposing or is just a, no, we'll just go with juxtaposing. Cause why not juxtaposing sounds juxtaposing. Where's this one? Yeah. Yeah. Yeah. Juxtaposing. Let's go with that. But you know, like perfect example, I think I just put it on my Instagram story, but I had, a flower vase, a glass flower vase sitting on a stone, So glass. Stone. You don't really think of those things as the same. They're opposites in a sense, right? Because Any glass touches stone, it's probably breaking. Like, so another unique way to use something like a blend is to take juxtaposing things, mash them together. You know, like I think about it like that as a creative, because that's a whole style, people make livings off of juxtaposing styles.

Play from 0:37:48 Right. And so that's, that's a super cool, I think, wacky, fun way to do that. Totally. That's a good one. Juxtaposition just in general. It's a great recipe for Midjourney, like, I think of style codes, that's how I do all of my style codes, when I want to add multiple, I probably do one that's really dark, and then one that's really vibrant, so, you get like a mix of both, that's just, that's the way I like it, I also like doing photorealistic and illustrative mashed on the style codes, so you get like this cool integration of both, so style codes is really fun to mix and mash and play with, you know, then you can wait them and dial it up, dial it back, but the juxtaposition, It's really like just a, if you want a great, just theory about how to create Midjourney, just take what drew just said and just do it for everything, like in every way, shape or form.

Play from 0:38:35 So that's a positioning. Wow. Name of the episode, maybe people searching for just juxtapositioning because if so, we might have a winner. Opulent. Just juxtaposition. That's a, that's the title right there. Yeah. We can, we can own that phrase. Definitely. Well, Hey, Sherry, that I love that question. We spent a lot of time on it, but I think that's probably one of the more meaty questions that we could get into because there's not a.

Play from 0:39:02 A black and white answer, right? It's a very creative and subjective thing, but I think that opens it up to everybody that could potentially be listening to this as to You know ways if we said anything that you haven't played with play with it So let's let's move to a different question, which I think has your your name written all over it So this is from my friend spirit spirit novel Sounds like a fictional character. This is a real person. I love her. have you found any techniques to get a nearly exact match of a pattern? Specifically clothing? Could be floral embroidery on satin or a specific pattern, plaid, or an abstract. Any tricks to get crazy coherence?

Play from 0:39:42 So from image to image, or Do you mean like make the pattern like perfect on, on one image or transfer the pattern from image? How are you reading that? I, I feel like this is sort of like this fashion related question that I see pop up there. There's a lot of fashion folks in AI and they're trying to, to take a specific item of clothing. Yeah, or an accessory, for example, and place that in Midjourney. Is that possible? An exact match?

Play from 0:40:13 No, no, that's that I think is the keyword that you mentioned exact match. No. now you could write, like you could get very similar results using something like, like you could take an image of that item, And maybe you go even in in painting. Paint out whatever they have on let's just call it. It's a dress. You're you've got a girl in a dress You in paint out the dress and let's just say you want to throw this pattern in there Then you could drop that image reference In your editor or your update and then also describe it and you're going to get something that looks pretty dang close But again, not an exact match.

Play from 0:40:55 I think for something like that Correct me if I'm wrong, Rory, but you're going to have to still go into something like Photoshop or one of these other tools, there are some like very niche tools with AI image generation that do a better job of matching exact clothing in those circumstances. So as much as I like to stay in Midjourney, if I'm, Asking this from like a fashion or product sense. I'm probably going about it that way. If you're looking for exact keyword.

Play from 0:41:22 Yeah. I think if you want to translate a pattern, you can do that. No, it's not going to be, might not be exactly perfect. If you want to take a fully designed article of clothing and put it on someone, probably not going to be exactly the same. If you want to take a, if you want to generate a pattern in Midjourney, then use that pattern on other things. Yeah, you can do that. So that's why I was like a little bit like, uh, you know, staring off into the distance there for a second. I was trying to find, I have examples of this where I basically took patterns and then placed it on different items or put it on dresses and things like that, because you can do that with image prompting.

Play from 0:41:53 All right, so let's look at this. This is a really old, I think this might've been like five, two or something that I did this in. Yep. Five, two. But I generated this pattern, right? In Midjourney. And I was like, let me see if I can actually take it and put it on close. You can put it on close. Now, is it going to be exactly the same? perfect, no, but it's like, you can do this with certain things that do this a lot with, um, with faces and in a lot of different ways like this, very similar pattern, you know, taking this and then translating it and putting it on a face, You just have to image prompt it and then play with the image weight to do that. So you can foreseeably do this and then you can go and utilize it in multiple different ways. You know, you can take some of this stuff. I see if I have a. Example here where I took an art piece and then blended it into the background because you can take, yeah, like, some of these,

Play from 0:42:38 you can take a pattern and then like apply it to a room, And like, it'll, it'll come and it'll translate. It won't be absolutely perfect, but you can take some of that style and transfer it over, which I think is totally fine. And it was, I like to like to really with some of this stuff, like we always say, build in layers. So I'm generating this pattern first, something like crazy like this, I know just going to translate well and then it'll apply it in very low levels. You can see here image way 0. 5. Not a lot. If I went higher, you're going to get more of it, but just taking it a little bit of a step further. Now, this might not be totally answering the question, but just showing you kind of how to think about the entirety of it because Midjourney can Handle these patterns a lot better.

Play from 0:43:17 This is, I'm going back to 5. 2, find this stuff. Cause I haven't done it in a while. That's what you can do. You can take a pattern that you have, or, you know, if you want to generate it in Midjourney or externally from Midjourney, and then apply it with a image weight, our image reference and image weight. So, so, so two key things there, two, two key things. One exact keyword can't do it. Not in Midjourney. Second component is if it's Midjourney generated images, you can do just about anything. It's, it's when you start thinking about things that are external. To Midjourney. So like an exact product or this or that, that it becomes much more tricky to get that exact replication, So to Rory's point, if you've generated a pattern Midjourney, like you could go ham with that, right? And if you've got that creative Liberty to, to move forward with it, like it can take on several different forms

Play from 0:44:07 or shapes or. You know, looks within it, but if it's that exact component, that's where it's like, nah, you probably gonna, you're going to have to get into another tool to do that. So, cause the other thing you can do is this is like kind of hacky and it doesn't work all the time is if you want to take like a product PNG, let's just say you wanted to have like a, pillow and he wanted to place it on a couch. Like you can use that pillow, a PNG image of a pillow. Use that as a character reference, and then just try to describe where it is at in the image.

Play from 0:44:37 So, I can actually find this now if I think about it. Right over here, let me see if I can find the actual where I have it all together. Like right here with this one, right? Like this was the source image. I just generated a PNG of a couch to see if this was possible. Now it works with other things that you pull from outside a little bit tougher, but at the same time, like it does work. So I use this as the source image created this one in Midjourney. Then I created this image of an open for like format in Midjourney.

Play from 0:45:07 And then I put the couch in there using C ref. So you can use C ref, use the couch as the C ref image. Now. Is it perfect? This one to me was pretty impressive that I used this couch and then I generated a pillow there too, and use them both as character references. And then they ended up. In here. So, you know, with a image reference as the background, a character references, the couch and the character references, the pillows. That to me is pretty close for, you know, the couch is not perfectly consistent, but the pillows are pretty close and it's something you can play around with. Again, if you're looking for 100 percent accuracy, it's never going to work, but it does work with things like cars really well. But yeah, that's just typically how I would go about it is if you're looking for something like that, character reference can be a friend.

Play from 0:45:52 that's another good point is the fact that you bring up character reference. Possibly being used effectively in some regards as an object reference, right? Because we don't have object reference in Midjourney, but You can sort of play around with, Hey, does like character weight kind of solve that for me and it possibly could, Like maybe it's not super consistent and I think that's maybe just. Hey, look, while we're talking about consistent and exact, where we're at in reality with all these tools is inconsistency. Okay. And so I think everybody just needs to breathe, relax. Like you're not going to get a great image every single time you put in a prompt.

Play from 0:46:34 You're not going to get exactly what you want every time you put into a prompt. Like, so I, I feel like some people may give up too soon prematurely. Okay. And maybe get frustrated, possibly prematurely, but I think, again, you have to really be excited about what's happening with this space, the options that do exist, like seeking problem salt, solving solutions, like we've talked about in the past. Like if you see a problem, see if you can solve it. Okay. And, and you know what, if you can't solve it, let's revisit in six months. My build solve it then. Right. So, so everything's inevitable matter of time. So if you can't do it today, it doesn't mean it's, you're never going to be able to do it.

Play from 0:47:12 Just not today, So just throwing that out there. I feel like some people need to hear that let's go to next question. I feel like we wrapped that one up pretty well. Yeah, this one's from dr. Beva, jessica thiessen She asked how to work around the content moderation limits in Midjourney. For example, you can't say quote unquote same sex couple, but you can say same gendered couple or pair. Same for skin, breast, child with parents in bed. Is there a cheat sheet of do's and don'ts when it comes to words that get past content moderation essentially?

0:47:30Content Moderation Workaround

Play from 0:47:49 Yeah, it comes out of text prompting. That's just, that is the fact of the matter. if you say same sex couple kissing, they might not give it to you. But if you say man on the left, kissing a man on the right, there's no problem. You can do that right away. I have images like from 10 minutes ago that I was doing for someone else. Right. So you can do it. You have to be really specific. I think this is the same thing with a lot of the, I don't want to go into the hot button issue of where, you know, everyone is upset with the, biases in prompting, but really it comes down to you, like you can prompt it if you want to.

Play from 0:48:22 And, you know, if you want someone to be a little bit overweight, you have to specify it. If you want someone to be a man specified, you want somebody to be a woman, you want somebody to be by, you know, uh, you know, binary with every, anything like that's what you have to prompt for. If you want to have a child in bed with, uh, you know, with their parents, you can do like, you know, uh, a, Text prompt where it's like child on in a bed on the left parent in the right, you know on the right That's how you have to be really specific and think about you're like you're explaining this to someone who's blind who's never seen before Hey, that's my line.

Play from 0:48:53 Is it yours? Did I take that from you? No, I'm just fucking with you man pirate. I was like wait, did you tell me that did you say it on another podcast? No, I mean Because you have to be very matter of fact the way that I look at it, you're absolutely right, because we have to look at the text prompt first of all, because the image upload seemed to have more moderation to them. Yes, but, but also like, I'm looking for that hot keyword. That hot triggering keyword. So with your examples, sex is it right? So how can I say that without using the word sex? How can I say breast without using breast? so it's like. Do I have a synonym for that? That is not, polarizing, that doesn't carry that weight of the other word. Right. So I think that's

Play from 0:49:40 the reality. Now, is there a keyword list out there of, of words that are moderated, not to my knowledge, right? I think it just, it's trial and error, but, Yeah, it's just like being very matter of fact with it and taking a little bit of the emotion out of it, almost being a little bit more cold, you know, the way that you describe it, Yeah, it's unfortunate. It's the way it is. I mean, like, you know, these, they don't have a brain. To be quite honest, the image generator doesn't have a brain is saying like, I need to be biased. Just it's just whatever data in data out. So if it has more of one specific type look of a person, that's you know, that's why everyone put it this way.

Play from 0:50:18 It's why every woman came out looking like a model for the first couple iterations of Midjourney. Because if you think about the whatever they were scraping off the internet, whatever is the most readily available photos, It's like, you know, there'd be model photos, fashion photos, brand photos. They're all edited, polished, skinny models, perfect faces. It's not like they scraped, you know, 1, 000 Facebook accounts and just took all of our data. Maybe they did. I don't know. I'm guessing more so they went through Google Images, whatnot, public source, or, Stock sites, things like that. And people just aren't typically not beautiful in those spaces.

Play from 0:50:53 So that's why everyone looks really good. So you have to prompt outside of that majority of time. Majority has got so much better with that. regardless, you get more variation now. Because I think the data is the training set has just gotten bigger, but it's tough when it comes down to text prompting when you got to do what you, you know, what you can there and, uh, hopefully it'll change soon. But the other, the other one that's really interesting is like, you can never write like bikini, which is interesting, which, you know, regardless. You can just put like environment appropriate attire. Yeah. Swimsuit environment, appropriate attire gets around everything for me when I can't do it. that one also, whenever I ask for women in fitness attire, they give me a bikini.

Play from 0:51:36 I was just like, that's not fitness attire or athleisure. That's a bikini. Just a little hack for anyone out there who's looking for, you know, another example, like remember when they started moderating the presidential candidates, like Trump, there was a period of time where you could prompt for Trump, but they've now changed this, but the way people were getting around that before it was like 45th president of the United States. Yeah, or, or describing what he looked like and then the president of the United States. Right. Like, so like, or former president of the United States and then his descriptors. And so it's all just, I think you just need to take a step back and again, just like be more matter of fact about it to get around it.

Play from 0:52:11 it's interesting how you wonder how they tagged the images as they're getting uploaded, Like I'm sure they just have to get tagged with certain, you know, to, for token retrieval. Right. So how else could it be tagged other than bikini? You know, swimwear, whatever, what are the category you talk about? Sometimes it's just sometimes the last chat, GPT stuff, like how, how else would you say this if you weren't trying to say, bikini or you weren't trying to say same sex couple and it will just give me like a bunch of different ways to say it. And that's how I go and try it. So yeah, there's a ways around it.

Play from 0:52:40 You got to get, sometimes you get a little creative. I know that's not the best answer, but it is. Well, we got, we got, we got creative folks on here. So yeah. Well, here's another question for you. What's your Midjourney plan? Are you on the pro you on the mega? And are most or all of your generations private? That's a question from Saber Talili. I hope I said that last name right. I'm on the yearly plan for the, whatever the biggest one is. Mega then, I guess? Yes. Whatever I did the, just did it, paid for the year. That's the biggest discount I could get because guys, look, I'm just going to use it. that was kind of like an investment question to me.

0:52:45Subscription Plans

Play from 0:53:18 Am I going to, am I going to spend the money up front? And I'm going to use it for the year? Probably. So I'll just save the, save the discount on whatever the monthly subscription is. what was the second part of the question? Are most or all of your generations private? All public. All public. I don't even know where I can see them. They'll never see them on the discover page, even if I search them. So I don't know where they go, but I'm not in stealth mode. I have that turned off. Yeah. But yeah, so also on mega, because dude, I mean, we fly through hours, um, especially like permutations are the death of me, but they're absolutely necessary.

Play from 0:53:51 So I just have to do it. Been on stealth for a long time. I thought about what Rory told me a little while ago that he wasn't, and I was just thinking about it. I was like, you know what, maybe I don't need to be. I've never been found. I don't know. I've never, I don't even know how, like, I've never seen it. Anyone else? I have been, here's the thing, right? Like, so here's the followup question that this person also asked. And this was like a little bit why I was wanting to be in stealth. And there are some real reasons. I think. That exists out there. Like, for example, if you're an agency and you're doing work and you don't want other, like, okay, stealth mode for sure.

0:54:00Explore Page & User Discovery

Play from 0:54:25 And which, which means that you need to have the pro or the mega plan to use that. the other thing was like early on, you could actually, I don't know if a lot of people know this, but you can click into somebody's profile from the explore page. Have you ever tried this? Really? Oh, yeah. Yeah. Yeah. Yeah. So this is what that guy's talking about because he, he also asked, if we're in love with one portfolio in the explore mode, is there any way to reach out to the portfolio owner for collaboration? Okay, so here's where we're going with this. If you take any image from the explore page, let's just find something like this. Okay, so I get the prompts, but I can also click their name. When I click their name, I can see every image that's public, So I can go,

Play from 0:55:11 I can scroll through and see, and this is what this guy's talking about. Like, let's say you find somebody that you like a lot of their images, but there is no way to follow them or connect with them or see who it is or know anything about them or no, to answer your question, reach out. And so. That's sort of been an interesting component to think about, because if they wanted this to be a little bit more social or collaborative, maybe that's an option. Maybe they do a better job of bubbling up these, profile names, for example, in different ways, like leaderboards or, you know, make them a little bit more visible and not so hidden, but this is absolutely a thing.

Play from 0:55:45 I mean, you can get some inspiration if you find somebody that's, that's got really good photos that fit your style. Right. So, yeah, I mean, I can go back and look at boy in square. I can see exactly what he put in. He's got. a multi prompt here, Rory, which we were just talking about, So analog, Leica and photography put a stronger weight on that than he did world of water and then also documentary and travel of magazine photography, very well composed, unique places, pics of the world, water nature. So it's very familiar.

Play from 0:56:11 Yeah, I mean, so, so this is absolutely. A thing, And so like the reason I stayed stealth is I just didn't want, at least initially I didn't want everybody seeing everything that I'm creating, possibly taking things, recreating it, but, same thing, Like I could click into this guy's stuff. He's got a bunch of these characters, right? So it's saying, oh shit, you know, kind of like this, like, let me, let me heart this. So yeah, I mean, this is like a great. Place to get inspiration to build on things that you already did. This is sort of where it gets in the lines of like, look, if you find somebody who's stuff you like and you start creating the same shit, then you might have a problem, right?

Play from 0:56:47 So can you find an explore where if anyone has liked your stuff, like that's outside? No, no. Right now. So, so there's no notifications on if anybody's liked your stuff the way that I've. Understood how Midjourney, elevates its images on the explore page are based on those likes, right? But there's no formal system to go about liking images that existed before this new thing that they rolled out, which now you can rank. Which image is that you think should be on the Explore page?

Play from 0:57:20 Uh huh. So now if you go under Tasks, here, let me just show this while we're here. Please do. This is, this is all brand new to me right now. Missed a lot in your two weeks, man. I know, how it goes. What the hell? So let me, let me jump over here. So the other part of this is now tasks. If I go over here, you've got, they've sort of redone this whole thing. So you've got the rate ideas, you've got the surveys, Which I've done them. So just as view completed ones, and then you've got these four.

Play from 0:57:46 So now you can rank images just off aesthetics. You can rank it off of prompt accuracy. You can rank Niji for personalization. And then now you've got this thing, which what we're talking about, which is curating the community front page. So now it gives you this big spread. page by page and you can select one image from this grid of what is this? Three by four, 12 images, So let's just say I want to vote for, I don't know, which one should we vote for?

Play from 0:58:14 I don't even know. This is so, let's do this one. I kind of like this one. Now it's going to take me to a new page. Same thing. I pick one out of these 12. If I don't like any of them, I can skip this, go to the next one. But this is now, I think how they're trying to curate. What's shown on the explore page. Whereas before completely based on natural discovery, I think probably some things within discord that I probably didn't know about would be my guess too, because I don't probably everything with discord, but yeah, I mean, it was almost like, that's how the velocity had to happen for you to get that.

Play from 0:58:49 Uh huh. All right. Well, that's. Good to know now, I didn't even know that was a thing yet. So thanks for catching me up, dude. Yeah. Yeah, brother. I got you. Well, Hey man, we've probably got time for one, maybe two other questions and I think what we should do, we've still got a handful is split this into a second part. We did go into office hours a little bit with this one, which ate up some time. So I think, you know, that second part of this episode, We can just attack the questions, but there are a lot of good ones we still got to get to. so let's take one more. Let's take this one from Sarah Hart.

0:59:24Style Transfer

Play from 0:59:24 Sarah says, I can't seem to get right in Midjourney how to turn a real life photo into a cartoon or a cartoon into a real life representation. Okay. Great question. Okay. So we just brought up Sarah's question. she left us this great visual example here. So let me just set the scene and then Rory and I will dive into this. I think we've got a couple of different ideas, the way that you can attack this. Essentially what Sarah's doing is, she saw this really great image that she liked.

Play from 0:59:53 She took the exact prompt, including the style weight and image reference to try to replicate it one for one, right. Or to try to get this realistic Charmander. In her image, And in doing so, she couldn't get it. She was still getting this cartoon result of Charmander. So yes, there are a lot of different ways that you can go about doing this. Rory, why don't you kick us off? I think we've got several. So Sarah, if you're listening to this, get out, get out your notebook. Cause we're going to give you several different ideas of how we troubleshooting this. Particular example here.

Play from 1:00:28 is there an image weight in there? Not an image weight in there. So let's just assume it's one. Okay. So also not a style weight. So let's assume that's a hundred. Okay. So we have Charmander as the image reference, right? That's an image. That's an image reference, not a style reference. The illustration. Okay. So that number one is where you're going to want to drop that. So dropping that to probably anywhere between 0. 5 to 0. 8. We'll give you just less of the illustration. So it's probably going to come through longer, you know, more there, but also if you want to pump the style weight on that as well, which I think we're talking about as well, it's just like pump that up a couple notches.

Play from 1:01:06 Cause I don't think it's there. So it's at a hundred. So maybe one 50 to like maybe. 200 is where it start. And then if it needs more, keep going. So you're trying to balance. If you're thinking about this, trying to balance how much of that image you want, how much of that text prompt you want, and then how much of that style you want. So oftentimes if you're getting too much illustration, right, what's influencing the illustration, we have the image reference. So we need to take that down. So we need to take that image weight down. So that could be between, like I said, 0. 5, 0. 9, maybe something like that. Now, if the style is not coming through then, then you got to, you know, how do we get more style dial that up?

Play from 1:01:43 Also, if it's not in the text prompt, I can't read it. Does a seventies live action scene of Charmander? Yep. Standing heroically in a foggy forest with a small flame on his tail. So it's basically describing Charmander talking about the lighting and then talking about his expression in the scene. Style you could also type in dash dash no and illustration or something like that or no cartoon or no, whatever Yeah, there's another way to do it So the image weight is the first thing that could that that comes out to me because the image weight now Seems like one.

Play from 1:02:15 I don't know if you think that's to like image weight one has way more firepower. Oh, yeah We've talked about that for sure a hundred percent so That's what's pulling through. did I cover too much there? No, no, no. Okay. So just to recap what you said. No parameter you could, you know, basically rule out things like cartoon, illustration, for example, you also mentioned style, weight, dropping the image reference, agree with all those. Right. And so this is the great part about Midjourney. This is the fun part, in my opinion, because you and I troubleshoot a lot of these kinds of things.

Play from 1:02:47 there are multiple options that you have. So if one of these doesn't quite work for you, try the next one. So those are all great. The other thing I was just thinking about, and you mentioned this is remix, which is probably one of the easier ways that you could go about doing it, but it depends how closely you're trying to get to this exact image, So that's the other part about it. Um, what I would probably do. Is just take this image that they created and use that as your reference, For both, probably both style and image, or you could try one or the other, like on its own, you know, that's the way that I would go about it and then dial in. The weighting appropriately.

Play from 1:03:26 So that's where your permutations come in, right? Where you maybe want to run different values to see what gets you the closest. But to me, this example, right, because you don't need, you don't actually need that cartoon image weight because you don't want a cartoon, I would imagine he's actually using that just to tell Charmander looks like, But he probably didn't need to do that. Honestly, drop that like Rory said and run a higher style. And probably get the same result or just take that image, And go about it that way, which to me is, I would say the more common way that I typically do it.

Play from 1:04:03 I'm not saying that's the best way to do it. It's just the way that I do it. It's a good image of Charmander, which I understand why, like, whoever did that probably used it. the illustrated image, because it's going to give, like, the right body proportions and give it the right context. So, that's probably That's probably a good on the initial prompt, right? But like trying to recreate it, it might not be the same thing, but I think what you just said, perfect. Just take that image, use it right. If you need more guidance, then add the image, the image, uh, reference to sort of guide the character. You could probably also edit this down to now. So if like, you don't like the way that Charmander looks, the other thing is the, is the aspect ratio exactly the same in these, is it four or five?

Play from 1:04:39 Yeah. but if you want to edit this, you know, you go in there, you highlight Charmander. And then either one utilize this image up there as a, as an image reference or. You know, play again or play around again with something like the style, the style weight, the image weight, to see if it'll just change the chart matter. If you like the, if you like the environment, you can use the editor feature now. So you could, you could also try character reference for that. Yeah, actually reference. You know what I mean? Like, so, so multiple ways, but I think your answer lies in the power three.

Play from 1:05:09 What do we call on those, man? We need a name for those three character reference. Style reference, image reference. We get, we need a cool name. Somebody, somebody come up, there you go. Somebody come up with something cool for us, but, but to me, yeah, the answer lies in those three, right? Like I, you could probably get away with just using an image reference and waiting it appropriately. Okay. If that doesn't work where Charmander starts to not look like Charmander, then I think you get into something like, Character reference, As a stack on top of that, because that way you're keeping him closer to what he actually looks like. depending on the rest of the style of the image and how much that mattered to you, that's where maybe throwing in a higher image weight or throwing in style reference for that matter might come into play, but I think your answer probably lies with image weight, starting there, maybe adding character reference.

Play from 1:05:58 it's like, very interesting to look at, with that set with her images, right, you can see that the, the image reference transfers into the body positioning on the, on the Charmander, the one that she's trying to recreate, that one doesn't, it might just be like a random gen, like you said, it might just be like a random, like, one off that, you know, if you guys have been doing this long enough, you know how many random gens you get that are just low key awesome, and you're like, I wasn't going for that, but hell yeah? You know, so sometimes that also happens. Happens to me like once a day, minimum. Oh yeah, yeah, yeah. that was a really interesting question by the way. Thank you for that, really cool to just kind of like, get to break something down and appreciate you putting it together. Yeah, like to show us.

Play from 1:06:37 I think these are the things that we probably like, We, I think we think about them, but we probably don't like slow down enough to like, think about them enough because you and I have done so many like carousels, but it's like, it's, this isn't the best way carousels. Aren't the best medium to show that it's on video, right? It's us like being in a room and talking about her doing something like this. So I think, yeah, great. Great question. We got more great questions. We're going to get to in part two of this. So Rory, let's wrap this one up, man. I think we, we did a pretty good job. We had some awesome questions for round one.

1:07:00Episode Wrap-up

Play from 1:07:10 Thank you to Sarah Saber, Jessica. Spirit, Sherry, great questions, appreciate you guys putting those things in the chat for us. Rory, anything you want to wrap up with? No man, that was good. I really, I really appreciate everyone who threw comments on the post. They were awesome. I texted Drew last night and said, holy shit, like we have a lot of good questions. Like not just Not just, all right. Yeah, we can, we can rip through these. Like we have to talk about 'em. So that's why there's a part two, because they're, the quality of the questions was outstanding. So thank you for interacting and thank you for providing those. 'cause it gives us content too, so thanks.

Play from 1:07:47 that being said, solid episode again, dude . Absolutely. Absolutely, man. Feels good to be back, back in the saddle. It does. And, uh, we'll be back with, uh, part two of this, and we're gonna get deep in the, in the weeds on those two. So. Thank you guys for joining us. We'll catch you in the next one. Later.