Transcript · June 20, 2025 · 16.9K words

Ep. 44: Midjourney Video is Here! Nerds React + New --SRef System Chop-Up

Auto-generated captions, lightly cleaned. 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.

0:00:00Thursday night chaos & surprise video drop

Play from 0:00:01 What's up everybody? Midjourney Fast Hours podcast episode 44. And uh Rory and I were just sitting here talking. We were just like, man, what do we what are we even going to talk about this week? Like nothing has happened. Nothing. Zero. Like like not like there's been a video model dropped. Not like there's been like a new SRE system dropped. Not like Midjourney is just shipping out of their freaking minds right now. Nothing. So, dude, this is I'm super pumped for this one because, you know, like there's always like a couple of those weeks where it's like, you know, small changes. What are we going to bring to the table, right? And this one, I mean, dude, I this could be a [ __ ] 4 hour podcast if we wanted to be probably.

Play from 0:00:47 We won't do that to you guys. But, we did bring our drinks. We're going to do We're going to do a little Midjourney After Dart. We're shooting this a little early. They were doing this on a Thursday night this time and uh yeah, we're going to go into the weekend strong and just kind of play with, you know, all these new updates. I mean, just so many new toys, new shiny toys. Basically, just like a emergency podcast on a Thursday. We fit it in cuz we had to couldn't couldn't wait around till Friday for this one. We need to get it out while it's fresh. And if you can hear my mom laughing in the background, that's that's what those nights I can't, but I want to. It was like I don't know what my dad just said. said, "I'm at my parents house." I don't know what he just said, but she is like borderline pee the pants kind of laughing. So, yeah. Damn. Your dad still got it then.

Play from 0:01:32 You know, he'll throw a zinger in when every once in a while. I wouldn't say that's that's 100% of the time. But damn, man. You get you get your wife laughing after that many years, bro. You still doing something right. I I you know, you work for those moments when you get it. You're just like, "Yeah, did it." Yeah. Yeah, cuz you swing and miss probably nine out of 10 times at that time. Dude, I I say so much funny [ __ ] My wife won't even crack a smile. I'm just like, man, 100%. Come on. Cuz it could be the funniest thing, but the wrong time that doesn't make a difference. If it you could you could have the banger of a of a you know, a line and if you know if the girls are like you know doing something crazy in the kitchen or like you know my wife's like changing a diaper like there's no like in one ear, out the other ear. I didn't even hear it. I'm

Play from 0:02:17 like, "Well, now that's gone into the ether forever. Never to be spoken again, but I know it was a killer." So, I mean, you know, you know when you got a good one, you know, you just want to see it, you know, just just a little bit. Just a little something something. Yeah. You know, it's uh you know, I'm pretty sure my wife's probably pissed off at me after just like stopping our day yesterday, just being like, "That's it. That's enough. Midjourney drop video."

Play from 0:02:44 Yeah. What were you Yeah. What were you doing yesterday? Because I text I I just happened to see it at the right it was like one of those right time things because you know I've I'm doing [ __ ] during the day but um like I really saw an up you know like I just saw this update and it literally just got announced like 3 minutes ago and I'm like whoa like that was unexpected and it's still cool the way that Midjourney dropped it cuz it's like a whisper in the wind you like it's like news the old way you know it's like the pony express to some degree you know it's like you get it when you get it. No blast, no big like reveal, no giant, you know, demo video showing like all this crazy stuff. Just drop it on the website and then like you tell someone and then someone tells someone and then you see it on Twitter

Play from 0:03:29 and then like it just goes the most organic way. And that's the one thing I will say about this is this is the most organic release I think I've experienced in a long time where I got excited again. I was excited about VO3 when I started using it. I didn't feel this like I I felt like awesome. I don't know what how else to say it. When I started hitting the button, I was like, "This feels like version four like sort of like the unknown. What's going to come out of it?" Yeah. Is it going to be good? And then when it hit like right away, I was like, "Oh, thank God." You know, like it just it was just like a wave of relief of like, "Yes, all right.

0:03:40Midjourney’s whisper video launch strategy

Play from 0:04:09 We're back in it, baby. We're so [ __ ] back, dude." Okay, listen to this timeline. All right, so round one video rating party June 11th. Uhhuh. Video rating party round two, June 13th. Real video ranking June 15th. You got me excited with that one. And that's when I was like, "Oh, fuck." Cuz I didn't I didn't read all the fine print with that. Right. So, let me like just kind of show you guys what we're talking about here. But you know just going through the um you know the updated notes here in Midjourney right which you can access just on updates here but I was just like oh real video ranking click here and I'm like oh dude what so like this is like pre-release ranking

Play from 0:04:56 and so I went into those and those were it was just like boom boom boom and I was like wait no way I was like this is real and then then I went back and read the five I texted I was like, "Bro, wait, what?" And you're like, "No, dude. No, no, no, no. They're just read it. Read it." I was like, "Ah, shit." Okay. June 15th. June 16th video pricing survey. And then you get the uh the style references, the new update also.

0:05:10A wave of feature updates: SREF, video, pricing

Play from 0:05:22 June 16th. Um, yeah, default model. And then all of a sudden, you get video on the 18th. Like, dude, that week was just nuts. like can can we just talk about like this is Midjourney's side project essentially like video as it seems but yeah there's a really good so there's a really good line in there which is what we've mentioned in here on this podcast a few times if they want to get to like real time world building simulation which is where they've always said they wanted to go they were like we have to do this in steps and we have to do this in layers so we said we definitely talked about this where I was like well you start with images and then you go to motion which is videos and then you go

Play from 0:06:08 to 3D and then you go to you know so then you go to 4D so they they set it like that's the timeline that's what they're building towards so this is just a step and that for what seemed like a side project right like this is this is damn good when you me dude it's funny that you mentioned that because I was just reading this right like going back to the office hour nodes goal images video real time 3D integration within one to two Okay. Long-term vision. Real time open world simulations with interactive environments. Yes. Right. This is I mean these these two lines here are pretty much everything you need to know. I'm confident they can do it now. Like after this like this to me was like a litmus test for the future. Like if it came out and just was like Sora and it was like

0:06:15Long-term vision: real-time world-building

Play from 0:06:54 all right this is hard and it doesn't work ever and it's just not fun. I was like this isn't going to be good for them. But bro, they unbelievable though. Great V1. I mean, I I'd love to like, you know, I'm I'm sure you and I just like like let's share some examples. Let's share like first impressions, strengths, weaknesses, like you know, obviously like like let's dissect this a bit. Um there's there's so much to get into. give me give me first pure raw um insight based on like your first time playing with this. So, taking you back to that time where you just jumped in, right? And then you you said there was a little bit of that surprise element of like and and look, we were talking about

0:07:00First impressions of Midjourney video

Play from 0:07:42 leveling the expectation, right? Like we had just talked about that. We were just like, well, look, I think we're both in the same camp of like we expect good but not great. like we know there's probably going to be some stuff, so let's not get our hopes up too high kind of thing. But like where's that landed with you now? Uh now that you've had some time to kind of like play with this thing. Uh there was if we go back like what was it two three episodes when we were talking about can Midjourney compete in video?

Play from 0:08:09 There was like really two things that I wanted from it and I stated there was one I wanted the Midjourney factor which was like you know I call it the Midjourney factor which is just like can Midjourney make this easy and fun number one they nailed that with autoprompt right like like that's the to me you can go on it's the only tool you can go on and type in picture of a car and get an awesome picture of a car and now you can just hit autoprompt and you can have a really cool video in less than you two minutes from with like a really good aesthetic without having to go and prompt six paragraphs, right? To me, that was massive. Like the auto I didn't know I wanted autoprompt until it was there and it worked and I was like, "Thank you for this." Yeah. I like the I like the two

0:08:30Auto-prompt is criminally good

Play from 0:08:56 variations. And I think that's we go back to looking at the video rankings when we were saying, "Why is one video super still and one video?" That's right. Yeah, you called that out. Yeah. Low motion, high motion. Right. So, this is great. And it's take Mid Journeys like very distinct aesthetic or the crazy [ __ ] that you can't animate in any other platform. And if it can do that, then like it's going to keep us all and we're all going to love it because no other platform can do that. Nailed it. Like they just totally nailed it. I went right to the like the most complicated images I could find and was like, "Let's see." And hit autoprompt and first shot.

Play from 0:09:42 I was like, "All right, here we go." For all the complaints about Midjourney being hard and there's so many parameters and there's so much stuff. Now, does it get any easier than autoprompt and you get a great video? I get some people want control. Most of the time if your images is constructed correctly or prompted correctly, it's going to autoprompt. Well, so like it's going to get basically the only thing only like a few things that it can do on some images, right? But yeah. Yeah. I after testing all these that's what I've noticed cuz like I' I've you know to where I'm at right now is I've basically tested really short prompts and mostly done you know the auto. I I've run some manual obviously like just see how it

0:09:55High motion vs low motion—how to choose

Play from 0:10:28 works. But like to your point, I haven't really gone into more of the the like very specific complicated sort of uh prompts to see how that maybe varies in in the results. But I would say like um in general, I like how approachable this is. Right. Super. It's it's not um it is not intimidating at all. Right. cuz you go in here. Um, like let me let me go back to my, you know, let's just go back to stills here. But, um, let me scroll back a little bit. I was loving just kind of doing this particular SR. I was just really loving this SRF I was doing. It's a good video.

Play from 0:11:13 Yeah. So, I was just kind of like going through and like, you know, simple animating these cuz basically these are all very simple prompts because the way that I unlock suffs, which we'll also get into, right, is like I I I use like either a phrase or a single word and I found some that like really kind of work well with different SREs to kind of give me an idea of if I'm going to like it. And so, these are very very simple ones, right? But um you know like just clicking into this and you know for the first time just seeing like hey I just got four options here right and you know low motion high motion it's easy to understand based like it for for me for you for most people I would imagine it's like it's pretty easy to understand when I would want to use low versus high

Play from 0:11:59 right so I like that idea of like little and a lot right based on the scene because like if you've got something very like still and stagnant and something like this static. It's like, okay, well, I'll probably use low motion. I might run high motion to try and see what happens, right? Just out of curiosity because we're experimenting. But like, you know, I love that. And then let's talk about let's [ __ ] talk about the fact that you get the four the [ __ ] four grid, which no one else is doing, and you get all of them at the same time. I'm not saying it's I think there is something there from a user experience standpoint that needs to get tweaked, right? because your eyes can't focus on all four at the same time. So that's the challenge, right? It's like cuz what I want to do is like as I'm

0:12:24The magical 4-grid video preview

Play from 0:12:45 trying to quickly move through these is kind of like take a step back and glance at them from far away and just kind of see follow the one that feels right the longest until it falls off and then I'm like, "Oh, okay. Not that one." Then I got to go back and kind of watch them. Yeah. Um but I mean the fact that you're getting four, bro, what the [ __ ] is this? This is awesome. Awesome. Awesome. Awesome. That was the way V2 was when I was beta testing it and I'd get four every time and it was it was amazing cuz you'd get like one to two that were like exactly what you asked for and then maybe two two of them or one or two of them was just like a little bit off um you know what you asked for, but it would sometimes a lot of the good stuff would come from those two. Now like if you're looking at the the one that you

Play from 0:13:30 have hovered over, right? like the one that has the exploding background. Like that would be the variable one, right? It's that's why it's nice to have four because then it' be like I didn't ask for that. That's what I want. You know, it's cool now. I I like that. I want to play with that. Um so the four having the four is great. The other thing I think there's that's that's interesting to note is they didn't do like a separate tab. It's all accessible within the create which is super nice to me. I like that. I like just being able to click it. It's all in the same place.

0:14:00Interface brilliance: simplicity done right

Play from 0:14:01 You can then, you know, sort by video. As you can see now, we're doing that. But that's a simple that's a simple subtle thing that like I was hoping it wasn't like editor. We're always going to be in like its own dedicated space and I was going to have to jump back and forth and then like, you know, so it's nice the way they did it and I'm happy they did it this way as they normally tend to make things simple and it doesn't get much more simple than having four buttons and that's about it. I love it. Yeah, I love it. It's a great first start, right? because um they did kind of cater to maybe people that that really haven't been generating a lot of video, right? It's like this is this is very approachable. Again, like I think anybody could appreciate this. Look, you want to go ham, you know, get into the extend the video or the you know portion where you're actually

Play from 0:14:48 um you know text prompting. Exactly. You have that option. You're looking for something just kind of experimental or loweffort, right? or just ease efficient, just do the manual. And like I've been testing that out specifically, too, because I wanted to kind of see um if I could identify certain patterns. Yeah. Um in in what comes back and I'm curious to see if you've picked up on anything so far because I' I'm seeing a couple things that I I've noticed. For example, like anytime there's like somebody like this and the motion's high, they always turn around. Yeah.

0:15:20Observations on motion templates & quirks

Play from 0:15:22 Always turn around. And then there's like, you know, like whenever you get into like some of these slow motion ones too, it seems like they're they're very similar slow motions. I think they're almost like slow motion template kind of movements. Yeah. Um because like I you know Yeah. Because like some you kind of like you see the rotation, some you kind of like they kind of jitter back and forth and then you know like others I don't know just something I've noticed like have you seen anything like that on your end or what other observations like have you kind of taken away from from what you've seen? Well, that one definitely.

Play from 0:15:57 Again, going back to whatever episode we did on this one where it was we were talking about images for video when we talked about that where it's like if you have motion baked into your images, it it's you're already have a leg up. So, like I tried a lot with the motion blur background or some sort of like you know like people are already like half running in motion or a car is like speeding and you can see like the you know tires are spinning and there's motion blur. So, it's like it just like takes it and runs with it, which is always what I've always what I've appreciated, but there's there's a creative factor to it, which is why, you know, again, I feel like I'm highlighting like 50 episodes here. We talked about this. I know, man. Sort of like Apple, right? We went back to a

Play from 0:16:43 long time. We like Midjourney feels like the Apple of this whole situation versus like, you know, when we were talking about it was like stable diffusion. Like this is that's how long ago it was. felt like the PC version or Flux now feels like the PC version, but the there's a creative edge to it where you don't have to control everything. Every other platform is focused on control. Midjourney says we do the other way like we'll just we'll just, you know, it's a great observation. Push a button and it'll work. Which is why, you know, I don't want to think sometimes. I don't want to write totally three paragraph prompts to get something cool. You know, I just hit high motion and bang, I get what I'm looking for.

0:17:00Animation consistency & style fidelity

Play from 0:17:20 Like that's awesome. So again, this is like a and and what what also can't be overstated is how hard some of these images are to animate like and how hard it is with motion to keep that style consistent beyond the initial frame. So, if anyone's not familiar with running, you know, imagetovideo on a lot of these other platforms, if Drew was running an image like this, you know, maybe prior to, you know, Cling 2 and V3 and, you know, some of the new developments, if you wanted to move off of that frame, like if you wanted to like pan left, yeah, a lot of times that style would not transfer. or Midjourney style would not make it to the rest of the video because it's just too complex and the video generators are just like I have no

Play from 0:18:06 data on this so I can't I can't compute. Runway notorious for that. Um Clling's pretty good at keeping consistency. Miniax was always pretty decent at that. But like you couldn't throw a lot of this stuff into V3. Like no no no shot at V3. I think it's revolutionary and awesome. Just some of this stuff just doesn't work in it. It doesn't have data. So like that is to me also again we were talking about where they could win having that style and being able to animate something like this right like that's where I'm seeing just massive improvement and I think if you look across if you open up Twitter if you open up LinkedIn Tik Tok whatever you're going to find people that are amped about this because they have all these crazy images that they're like this makes no sense to put it in a story

0:18:30Old stills, new life—archive resurrection

Play from 0:18:52 whatever I just wanted to see what it looked like moving and now you do that. So, it's a great point. Yeah, it's a great point. Like, this is again one of those moments in time to go back into the archives. Deep deep into the archives. Don't forget about all those great things that you've already created. I'm talking to myself. I'm talking to you. I'm talking to all of us because uh dude, that's I mean, we've we've already create we've already done the leg work on so many great images. We owe it ourselves to kind of animate and kind of see where these things take us.

Play from 0:19:24 I also really like this idea um here with like the SREF combination with the video, you know, like that's amazing, you know, like the fact that you can really, you know, get different and weird and unique because there are so many different SREF codes. There are so many preferences that you may have, you know, so like I'm going to go back and, you know, look at some of those SRF codes that I really like. And, you know, these were fun. Like I mean, these were the first ones I tried out with video.

Play from 0:19:53 So, a lot of this wasn't um necessarily realism, but like I just love seeing like this felt like a world to me, right? Cuz now it's it's the it's the SREF and you know, you got great stills, but now you know with the motion added, you're almost kind of creating this little world, right? Cuz you know, you could take any one of these and start to like connect uh the dots and and make this like its own little neighborhood or or ecosystem, right? In a way. And I think that's I think that's pretty neat.

0:20:00Using SREFs to build entire worlds

Play from 0:20:21 I I would say so. I think it's also like if you can't sort of dream up a world for yourself, Midjourney can provide a srefer random world for you. So that's another thing again like that's like personally sometimes I just don't want to dream up a world too. Like sometimes I just want to see what it can do and where it can go and I want to play off it. But like S referendum perfect for that, you know? It's like there's it's like forced creativity. Like I'm I'm not maybe I'm not creative in the moment but I can get there with some with some help. Um, but dude, the these are really just like overall first impression.

Play from 0:20:59 Expected to be let down, not let down in any, shape, or form. Yeah. Like really solid, good first look. Yeah. You know, and people think maybe that like we're being over like overemphasizing it. I think it's just from using all the other video models so much and knowing what's hard with them and knowing where like the rough edges are and coming into this and pushing one button and getting good work out of it. is like it's a lot like there's a lot of thought that goes into the video stuff to make it good and just to have one button and it works and it works with the crazy stuff you know like that's it can't be understated with and I think you saw I saw so many people back on I you know I did a post on it on LinkedIn yesterday saw a lot of people

Play from 0:21:45 that came out of the woodwork that I hadn't seen on my post in a long time interesting lot of people that were like so amped about it I think there's like I said this to someone yesterday. I forget who. Maybe it was uh maybe it was our buddy Min. I was just like, I like feel something again for AI. You're no longer numb. I'm no longer numb. I was like, I feel it again. I'm like, oo. I'm like excited. I was sitting around waiting to play with it today while I was sitting in four hours of traffic. I'm like, come on.

Play from 0:22:13 Brutal, man. Yeah. Brutal. What? Whatever. Midjourney is better than traffic. Guarantee that. Take that to the bank. So, um, you know, tell me this too because we we'll we'll kind of like preview kind of like what's next here because they did announce kind of like what what they're working on next with this, but any like where where do you see the shortcomings currently? Let's I mean, obviously we're looking at this V1 model just released. So, I mean, we're not going to sit here and blast it, but like where do you see the shortcomings currently here? um in terms of like you know short-term improvements that you would expect to happen.

0:22:50Where the current video model falls short

Play from 0:22:51 Shortcomings is a few. Number one, I think it's interesting that they chose a 480p route instead of a 720 or 1080. Personally, if if that's what it takes to keep the cost down and give me four generations, I'm totally cool with it. Like 480p I it it says it's 24 frames per second. I don't think it is 24 frames per second. It kind of looks a little bit choppy in that regard. Um, but like those can both be fixed. Like if the quality keeps up with the way that it is, the motion, the coherence, everything else, I can sacrifice to the 480 now cuz, you know, we highlighted this on the episode last week on Topaz Astra now. I can run that through, get 1080p, it'll be crisp. or just regular

Play from 0:23:37 Topaz Video AI up the frame rate, up the resolution, totally good. So, like if that that's what allows me to generate more videos here of the things that I want to play with, totally fine with that. How much how much is Topaz Astra? Topaz Astra, I'm not exactly sure. Is that is that a license or a subscription? So, I worked on that product. Um, I gave the team a lot of feedback on it. I was working on I worked on Starlight with them, which was their original upscaler. If you're not if you're not familiar with Astra, it's Topaz's new video upscaler that is specific for basically AI video, adding more texture and detail, but it's built off of Starlight, which is like the their like initial diffusion upscaling

0:24:20Enter Topaz Astra—perfect timing for upscaling

Play from 0:24:24 model. So, it was the first diffusion video upscaler. But Astra is great. It's easy. I don't know if I I think I have some, you know, being honest, I have some free credits from being, you know, tester giving a bunch of feedback and like playing with a bunch of stuff. So, I I did buy the Topaz Video AI a while back, whatever that was, whatever that cost, the onetime fee, it was like 300 bucks or something like that. Um, and I use that all the time. Yeah. But Astro Now has been really good for, you know, they're trying to be magnific, you know. Yeah. Good, good, good pivot for them. Um, yeah. So, professional products are awesome. Not just like AI stuff. Like their stuff for video itself is great. Like they they have really

Play from 0:25:11 high-end products and they're really awesome. Yeah. I mean, they they had um I mean they were they had a lot of photography tools. I mean they I mean they still do, but uh that that was how I got in introduced to them a long time ago and they had one tool in particular that I had a license for and I'm the name is escaping me now. Um, but it was particularly good and I liked it for one gig. No. Was it G? No. Something like that.

Play from 0:25:37 Oh, it's gonna drive it's gonna drive me nuts now. Um, but what it did was it it had this awesome little thing at that time which none of these other ones did. And it was basically like this micro contrast um and tonal contrast sort of like function that when I was doing a lot of portraits with my photography, it allowed me to get so much detail on the face. Okay? You know, and it just it had like applied correctly just it it took the images to the next level. I would literally spent money on just for this one thing. And um man that like love that like he used to bring out the pores and so I like I did this series too where I was taking photos of

0:26:00A personal story on AI, portraits & impact

Play from 0:26:25 homeless people with with their permission. I wasn't being an [ __ ] Um but like I was telling their stories and I ended up turning this into like this whole fundraiser thing. It's a story for another day. Actually was like a really cool thing in the long run. But like the pe the those types of people typically have a lot of character on their face. They've been through some [ __ ] You know, they're not in the best condition. They're probably, you know, one of the worst versions of themselves probably in that moment, right? Like they're very vulnerable, but it's it it shows up on their face, right? And so it's like a tool like that used to let me dial it in and pick up all these little like all these little micro wrinkles and pores and you know imperfections and it was just like man just took the images to the next level.

Play from 0:27:11 So Topaz is like somebody I've been familiar with but I haven't haven't been keeping a close watch on it but it sounds like they've really pivoted well you know with with the AI stuff in particular and then obviously like with the upscaler stuff too obvious um even with the image gen. It's good. Well, they just they just introduced Bloom, which is their competition to Magnific. I I will say it's okay. I did see Bloom, but I didn't know what it was. So, that's that's the direct comp to to Magnific. Okay, got it. Good. Wouldn't say it's Magnific, but I still think like their video upscaler is like the product for the time. And I said that to the guy that I've been working with today. I was like, dude, you could you guys couldn't have launched this at a better time.

Play from 0:27:54 You're going to have two million people creating videos and they're going to want to upscale them because they're at 480p. It's just like right product, right place, right time. Like this is Yeah. You know, Astra's and Astra again, the one thing that I will say is that it still softens things out a bit, which, you know, smooths them out a little bit too much while adding a little bit too much contrast. Um, I think if you have a relatively decent looking image or video to start with, it's going to be great.

Play from 0:28:22 If you have something that needs a lot of work, it's going to smooth it out. So, just know that. Interesting. Yeah, that's just my that's sort of my analysis of it. This one Oh, you get some free credits here. So, there's this one with Bloom. I was like, it's quick, it's easy, it's simple. Um I thought it I thought it overdid some stuff. It's like magnific uh fractality on things. Yeah, that always drives me nuts. I hate that.

0:28:40Upscaler comparisons: Magnific vs Bloom

Play from 0:28:50 Yeah. you know, cuz you have to like dial it in so precisely to not get the artifacts and ah, you know, and and there's almost like too many options then because it's like, oh, you know, I like using the God, I haven't even used Magnificate in a couple months now. Um, but you know, it's like you've got those different models, then you've got the different uh types of photo it is, right? Is this art or illustration? Is this video game? Is, you know, so it's almost like it's it's great to have options, but it's almost like too many options, right? And then you you run into the fractality of the HDR stuff and you know you have to find this perfect balance. Um so which which you know whatever. But uh I did see that Magnific uh Magnific um upscale generations did get faster. I

Play from 0:29:37 did see some news on that. Yeah, Jav was doing something too. I saw him tweeting about something. I don't know what it is, but he seemed pretty excited about it. If anyone's not familiar, Jav Jav Lopez is a great follow on Twitter. He's the the one who created Magnific. He's always testing everything. But you the other thing now that you say it going back to it just joged my mind on the sort of the maybe the rough edges on Midjourney on the uh on the Yeah. You know what? Maybe I'll share it because a few things I noticed and it's probably very similar to what you've seen with other with the other video generators.

0:30:00Visual bugs: wheels, backwards cars & physics

Play from 0:30:12 There's a ton of things that seem to be the same and different at the same time. But let me see if I find the video. Like like cars for example, right? Cars if you don't specify, they sometimes just end up driving in the wrong direction, right? So like this one, that's something that happens in a lot of, you know, other video generators unless you tell it. So I pushed auto prompt on this. I got four cars that are moving backwards. Yeah. Now, that's something that's going to happen a lot if you're not familiar, you know, with how these things work because you're not like, I just pushed autoprompt. Why didn't it work? Right? Like, that's how a lot of these things happen. Here's another one that's standard. I I went back and used an old Midjourney 4 image, but a lot of

Play from 0:30:58 times you'll get static wheels, right? This is something we talked about a lot. If the wheels look like they're static in your image, a lot of times they're going to end up looking static in your video. So, this goes back to like prompting, you know, for certain specific items. Like, you're going to want spinning wheels. Like this one. It actually did it, but I think there's basically, you know, see two of them. Two of them have no spin. It just look like the car is gliding, right? So, simple like things like that that have always been prevalent, right? Like I've always I'm noticing that. The other thing is, you know, there's some interesting there's actually some decent uh I was playing around with this one.

Play from 0:31:35 This is an old like V6 image. I was like, "Oh, can we do fight scenes now?" Not bad. I mean, like, oh, fight scenes are pretty [ __ ] to begin with, but this one I was like, "Okay." Yeah. Like, I can screw with that cuz it was a this is a complex image, right? Like, this isn't something that's super simple to animate. There's a lot going on. There's, you know, all the all the water splashes in the background. We we touched on that a little bit like just kind of like thinking about you know we saw the rankings last time I think that's where we were at right like we saw I can't remember if it was one or two maybe it was just one that first video ranking party but we did say look hands hands

0:32:00Why hands magically fix themselves in video

Play from 0:32:21 and body movement already looks better than the stills. Yeah, right. Like you're like the the physics of this are already better, which then again I I don't really know how it works, but it feels like okay, well, if you've unlocked that with video, you should be able to apply that now with statics, right? So, the interesting thing is that if you have messed up hands in an image, the hands correct in the video, which is I've never had an issue with hands in videos really. Like I don't know what it is, but for whatever reason they they correct they autocorrect themselves.

Play from 0:32:57 Like let me see if I can find the example of that. Maybe it's it's one of these. I forget which is the good one that I that I utilize, but like her hands are totally jacked up in this like in the in the image, but they they sort of correct here. Maybe that's not the one. See how they kind of like autocorrect there? Like they they're kind of they were like a messed up Yeah. find the image. See how they're Uhhuh.

Play from 0:33:25 See, they're kind of like screwed up here. Jangled and hands, but when you come and actually like look at the video, the hands sort of just auto correct themselves. So like hands again, I don't think you have to worry about them as much. This was also impressive that you could spin someone, right? And then you come back and then the faces are the same most of the time. time. I think I extended this one, but that's that's another thing to test like if the models are good or not.

Play from 0:33:53 Like we'll keep the same face as they spin or as they turn because a lot of times it'll just regenerate to something else, right? Like it'll be something totally different. So that was one that I was pretty damn impressed with. Again, it's like little tests and then there's certain things like, you know, like can I just give it this and it just takes it. Yeah. Right. I gave it a magazine cover that or not at least a mock magazine cover that I made. Like that's sick. So just again things look at look at that movement. But the movement the movement and the pacing of it looks perfect. Notice the camera shake when he uh when he hits the ground, right? Like certain things like that where it's like he steps up or uh maybe it's this one.

0:34:00Face consistency & surprising details

Play from 0:34:36 Uh no, it's not this one. I think it's this one where he like stomps. You see like the camera like bounce up and down. That's what I was That's the one right there. Yeah. Um, hey, tell tell everybody like how you're Let's talk about organization for a second because now this adds a whole different layer of how how you going to track the videos you really like, you know, along the way, right? Um for me my process now is ch it it has to change with the video thing because right now there's no upscale option. Uh so that was how I was sort of like you know separating the images that I really liked from the images that you know were just was I would upscale them and then

Play from 0:35:23 that's how I would filter them in my archive. Now with video currently not an upscaler. Important to note though upscaler and higher quality that's in dev. So I would that like that feels like the next thing that they would pump here for video. So that's going to be interesting. But how are you kind of going around and and I guess separating the videos that you like? For me, I've had to pivot and then like you know, you know, actually like heart the the videos that I'm liking and then that's how I'm going to sort those.

Play from 0:35:57 But are you doing it any differently or how are you thinking about that? I'm being sort of selective with the videos right now. Like I'm only doing the ones that I really want to see. Like I didn't just go nuts yesterday. I was like I I had like a bunch of images that I wanted to test. Like I want I knew like I wanted to test this one and see what it could do, which I was like really impressed with what happened on the viewfinder. So I'm probably just going to start to like this stuff now. Like I haven't even thought about it cuz like I I knew I wanted to see what this stuff could be, you know? Like this one to me was like this sort of entirety was crazy that I just took that 2D animation and made that right like the little bubbles flying out the back. I'm like this is great. So I don't even know man. Like it's a great question cuz this is I was

Play from 0:36:43 so just dialed into to finding like it's so weird. I was like finding the things that I wanted to go back and animate. Yeah. Yeah. Wait, just in case anyone's uh hasn't seen this, we have a little, you know, Drew is now Oh my god. Damn. Holy [ __ ] I had to get I had to, you know, I like created that video and I sort of snuck both of us in there just being, you know, talking heads, which is what we're doing now. So, you know, a little sidebyside of us just being idiots. So, you know, like again, this was used, you can do this with yourself.

0:36:45Sneaking Drew & Rory into an animated video

Play from 0:37:15 This is pretty easy. just did a ref of a picture of me and an o ref of a picture of Drew into like a you know stupid caricature image and then you know now we can run it and now they basically have and now our YouTube thumbnails are out of control dude some of the stuff you sent me is insane this is the game we have to play on YouTube guys okay so you know continuing to talk about video for a second how much you loving the fact that we got an explore page now with video stuff. I haven't I haven't admission I have an admission of guilt. I have not looked at it one time yet. I I haven't I haven't really like scrolled through yet. Uh but this is awesome, man. So good. You know, you got similar, right? So default,

0:37:30New Explore page for video (and why it rules)

Play from 0:38:01 you've get you've got images. You had random, hot, top day. You could sort this top week, month, you know, images that you like. Now you've got this option with video. There's no hot, but you've got random, you've got top day. Let me see. Okay, top week, top month, and then you've got ones that you've liked. Still don't know how they pull those in there. But regardless, the other thing is like Yeah, there we go. Whatever. Whenever is that yours over on the right? No, that was where I got the idea from, though. All right, cool. I was going to say that looks that looks very familiar. Yeah, this one's dope. Does it have the prompts in there?

Play from 0:38:39 All right. So, that's the interesting thing, right? When you when you hit autoprompt, it just uses your image prompt like as the guide. Yes. So, the other thing the other thing that we should not I don't even know if we mentioned this is that there's an the extend the extend feature is and it works really well. You can extend up to four times. I tried to just keep going today. So, you can get 25 seconds worth of a video and it's super coherent.

0:38:50The Extend feature is shockingly seamless

Play from 0:39:07 There's no jumpiness in between. You know, a lot of times you see with the other video generators when you have to extend or it's like you extract the last frame and then start again. There's not a seamless motion. V3 does it really well. Not a lot of other ones do it well, but this one there's a very seamless pickup from like here's the end of the video to then it just keeps going right away. It's super smooth. It's exactly what it is. So, you know, I kind of like 5 seconds as weird as that sounds. Like I I think eight is the perfect one. Yeah. But you know, five to me is good if you can extend this. Well, what did you what did you type in there?

Play from 0:39:48 Girl walks away from the camera into the burning fire and catches fire and screams. I've already extended this a few times. I want to see if it And you know what? Like I'm I'm I'm also interested in in playing with this. I'm sure you're probably doing the same, but you know, I'll I'll take things that originally started as a low motion or a high motion and pivot and see how that adapts too, right? So, I would recommend like if you are experimenting and trying to like figure out how this works, like do that. Um, I think I don't know if I have any strong observations on that so far, like it doesn't throw things into a loop by any means. So, I think that's a that's a positive. Um, which is really cool to see because I think like you can in theory, right, like have parts that are sort of like more even pace, slow,

0:40:15Camera direction tips for better results

Play from 0:40:35 and then all of a sudden they accelerate or vice versa, right? Things start really fast and then all of a sudden you kind of tone it down like that. Those are those are interesting techniques and it's a very simple thing to do. So, I like that how that kind of bakes into, you know, they they play well off each other, I guess, is my point. Uh, and I'd like to see that continue. Let me actually let me let me pull up one because I ran the full like 25 seconds before before we got on. Let me let me see if this girl catches fire real quick. I'm really Yeah, I'm really curious to see if she catches fire here.

Play from 0:41:09 And I think I might have one. Um, here's the other thing. I timed it today. It was 48 seconds. It's like that's insane. It's like, damn, that's how long it took on for one generation. I'm like, that's again subtle wins. That's a win. Generating at 480p, you can get it at, you know, I don't know if it's a tactical decision because everyone's like, you know, if if we get faster gens, people will generate more. Do you know what I mean? Like I don't know if it's tactical, but I like it. And I like I like being able to rapidly iterate through stuff and just like scrap and keep going. And having it at 480p allows you to do that without wasting too much.

Play from 0:41:52 Did it Did you catch on fire? Yeah, a little bit. Only one of them here. Which one? This one right here. All right. But takes a little long to happen. Let's see. feel like it should already happen. Let's see what happens here. Going back to the observation thing when you said that I found that the prompt is more coherent on first frame than it is on extend. So the one thing that you're battling with everyone just so you know this this is true across basically all the video platforms when you extend or use like a you know if you go and try to make that secondary clip you're at the you're sort of at the bay of whatever

0:42:20Prompt control vs camera behavior when extending

Play from 0:42:37 the camera or the subject is doing. So like if you're if your camera is like swinging right and you like no swing it back left. A lot of times it's not going to do that. It's just that's just not the way. it's it's like going to keep going right because that's the way it's already going and you extend and you can keep going that way. So if you feed into it like if the camera is going right like keep it going right. There's things like that where you know you can't have the camera really go up and then like drop like way down like it's not going to it doesn't work that way when you extend. Like there's certain tools you can do that with with text to video a little bit easier. But with that being said just know when you extend I think that's a really good point. Yeah. So, if it's like push in and you push all the way in on like close up on someone's face, but then you extend you're like pull out. Sometimes that'll work. That's

Play from 0:43:24 probably an easier way to do it than like swinging. Um, you know, with it. I'll show you an example of what I'm talking about cuz sometimes you just have to roll with whatever the camera's doing. And that's the easiest way to get a better generation if you're text prompting. Um, if not, if you're just hitting autogen. Hey, look. Someone took that. Definitely. Someone definitely utilized the uh my girl in a red dress with a giant animal behind it. Oh, here it is. Click on that. Click on that prompt.

Play from 0:43:52 Canon shot. Yep. That's Yeah, that's the one I shared on on Twitter with the I used a girl in a red dress with a giant Oh, no [ __ ] With a There it is. with a giant rhinoceros. And I'm like, that looks familiar. [Music] Yeah, that's kind of dope. Sharp round top. Oh, that's sick. Yeah, that's dope. What? Dope idea. Hold on. Wait. Go back to that one real quick. Click on the first frame. Uh, no, it's the uh, right.

0:44:10Wild new video prompts spotted on Explore

Play from 0:44:19 Yeah. All right. So, it didn't So, it added that. It wasn't in there to begin with. Write graffiti and red and black tones in sticker format. In sticker format, huh, interesting. Good to know that that that can be done though. Didn't even think about that's dope. that I mean that's that's I think like the value of click on that one that product photo one I'm just curious yeah to see how the like the pineapple on the top keeps keeps consistent dude this is like a it's like built off like a 3D sort of mindset here cuz those weird that those like you see the pineapple shapes these are like weird nerd things that I'm looking at right now but the pineapple shapes and the shadows they stay consistent throughout this whole thing with the broccoli and the kale floating above it or if that's cannabis I don't know what it

Play from 0:45:05 Yeah. What is that? But that's like something maybe maybe it's kale cannabis sounds awful. Vegan. Yeah. There's some good stuff here. But like this is this is a perfect thing too, right? Like new feature, new capabilities. Explore page can unlock some ideas that maybe you didn't have. Also gives you some intel on the types of prompt, you know, that achieved it, right? So like just like perfect example is this, right? It's like don't know that I would have thought of that, you know, and probably wouldn't have thought of it for a long time, but that's super dope. Love that. I did that a long time ago in runway like sketch what I did was doing like sketch to basically because I was using final frame like last frame only

Play from 0:45:51 and just having like the sketch as the end. So trying to like animate it with the hand and that worked which is cool. But like there's uh [ __ ] Yo, this feels like this feels like some you [ __ ] right here. Which one? Look at this [ __ ] chip. Look at him. Oh, hell yeah. That's me. That's amazing. That's amazing. Well, look at Let me show you the Let me show you the experience. Yeah, show me what Yeah, show me what you got. Does this one This was pretty impressive to me. I was just screwing around with this before we got on. Let's see if this I don't know which one is the right one, but I'm I'm hoping it's one of these. But like if we started here, right, just so everyone can see the initial frame or maybe not, but that's the initial frame, right?

Play from 0:46:38 Crazy. There's a lot of motion in it. You see what I'm talking about with like the motion blur to begin with and like everything that you can see that it's already moving to the right. I don't have to prompt for it. I know that that's just going to pick up and run with it. So like it takes it and then it goes and this is like the first clip, right? And then we go into the second clip where it like zooms in, curves around, and then like we swing around front and you can see the camera motion or the the camera shadow here somewhere.

0:47:00Midjourney’s edge: vibrant color retention

Play from 0:47:07 There it is over here on the left. We just keep going, right? So, this is again, this is pretty damn good for extend. You would never know where the clips extended. They just sort of bleed into each other, which is what you want. They they do bleed in really well, too, right? Because like even when I was going back and like trying to pick you know first you got to find the one that you like and then it's like okay which one am I going to you know extend and then you know you get the extensions and then you know you compare those right but like that was what I noticed you know in trying to like analyze each one of those you know I would click in and I would click to that point right before because there's a break um that it shows you and I don't know if you can see because you've got like that that share

Play from 0:47:52 screen uh button right on your thing. But what happened here? Well, well, you can go back to you can go back to showing what you had. But I was the point I was trying to make is there's it shows you the break, right? Like when you go into and so I was skipping to right before that and as it would go into each one, it was so smooth and subtle, right? because like you're like, "Oh, I like I was expecting something to change instantly right when it started.

Play from 0:48:24 Wouldn't even know, right?" So, I think that is a really really interesting and good point, too. It's like, damn, that's kind of dope. Really bleeds in. Well, yeah, this one I think this one was impressive. Just that most time you have a hard time getting turns with cars, as weird as that sounds, right? Like getting a turn is an odd thing. V3 does it well. Cling can do it now. Well, but like having a sharp like 90 degree turn that wasn't something that was really easy, right? Like this one to me really kind of like like you guys again just nail stuff, right? Like some of these motorcycle shots too. And this is again I'm talking about adding motion to your image. Having that motion blur, it just like makes a big difference with this.

Play from 0:49:07 Like some of this stuff looks super real and you know I'm sure they have a lot of motorcycle data. It's dope. You know like some of these I don't know if it's this one but like even the curving of the camera like with that like these are complex like that's not easy you know this isn't something that every image generate or any lot of variables dude a lot of variables. Yeah. And keeping consistent with the colors and keeping everything I know it's the other thing we should mention here. There is no color loss. Zero color loss from one thing to the next. Like one from image to to video. Like I was totally shocked that this didn't lose color like even on the softness on the uh on the earrings and stuff like that. Like this guy,

Play from 0:49:53 right? Like it keeps the style perfectly and then continues to animate with it. Like even all these little micro details here might not be perfectly, you know, resolute, but I was really like these are little things like that soft blurriness on the earrings. Killer. Have you have you tested images outside of Mjourney? I have not. I haven't either because I have so many. I know. I haven't either. I I asked because I'm wondering if that's if that stays true with your point around the color um outside of Midjourney images just in terms of like how it picks those things up. But like this insane, you know, like this this color should should not work like this.

Play from 0:50:35 These are really I mean it's dude that is extreme and I was like this is so cool how that looks. This one was the original one that I was like, "All right, I can take something that makes no sense and it'll work." Like, this is so so extreme from a color standpoint. And I was just like, "That's it. That's the one that that's the one that specific video I was like, "Yeah, we're we're got" And don't ask me why. People probably the dumbest thing I've ever seen. Like that's I'm just like that color is so extreme.

0:51:00SREF update — what broke, what’s better

Play from 0:51:11 I mean, dude, that's what But I think the point we're trying to make, though, is in theory, it shouldn't work. We haven't seen that type of stuff work. No. Yeah. And here's you'll see not to not to I'm sorry for cutting you off on that one. No, you're good. You're good. But the again to another to dropping back to another podcast that we talked about, there is Midjourney's brightness factor. Like you see this with a lot of other image generate or videos they they get dark because it's too much power to render all the detail.

Play from 0:51:44 V7's super bright because it can handle the detail. Now the video generator can handle the detail too. So that's I just wanted to throw that out there that we're just we're hitting on all cylinders here from from previous episodes if you guys haven't listened. Do we really are? Yeah. Yeah. Hey, there's never been a better time to go back. We say that every episode. Look at this. This one's pretty dope. Like this. Love the just Yeah. Everything about this one. Did it get hunt on both sides of the car? It did.

Play from 0:52:13 Nice. And all it had was burnt orange hunt logo. Didn't even put it in quotes, but it this was the image. So, yeah. That recognition, right? Yeah. Actually, I take that back. It was wrong. Because if it was the other way, like if you go back to the image, it says hunt left to right. And then here it says hunt left to right. So technically, if it was the same, it would be, see how H is at the end, right? Like not the not the beginning of the word. You see what I'm saying? Or no, kind of like right here, H is the first letter. On the other side, H is the last letter. So it's it makes sense.

Play from 0:52:58 It's still like impressive that it did that, but Oh, yeah. I mean, well, that's but but that's that's how you would display it though, right? You wouldn't want them both to be facing the same way. Correct. But I'm Oh, you're just saying like, dude, the fact that it picked that up. It just picked that up. Yeah, that's Yeah, dude. That's [ __ ] that's ins Yeah, I know. That's awesome. I mean, and and there's no variation, right? There's no variation in those hunts because you know even something as small as that could ruin the image or the the video and and now like dude I mean that se this just little scene all the colors popping that little neon I mean that's dope fresh the rim even the rim color was orange on the left side you know it had like the little orange trim over the black rim and same on the other side like that's it's things that

Play from 0:53:43 you want like that cuz that's that's stuff that wouldn't have worked in the past in a lot of other you know generators. Should we also, you know, throw in should we talk should we talk SREF update? I mean like definitely interested. Definitely rejuvenated. I think like my observation in like running.

Play from 0:54:11 Oh [ __ ] man. I've run so much [ __ ] over the last week. This has probably been the most I've run in a long long time. Um, but those Fast Hours, dude, you get so many. Yeah. What I'm noticing though is those these SRFs do look more um pronounced. I don't know if that's the right word, but they they like whenever I'm running SREF random, they they look the variance in between each ones are much different. Um the styles almost seem stronger in some way. I don't really know how to explain this.

0:54:15Favorite tokens: dribbble.com, Nike ads, stoicism

Play from 0:54:47 Yeah. But um there's something different there. And I got some good ones finally. There was like a run where I had like eight in a row and I was like, "Oh, okay." Never really like eight in a row. But what's the uh I know, right? Timeout. So you go dribble.com I know is one of your go-tos. Yeah. What's the other some of the other testing tokens you like to utilize other than I know you use some emojis too, things like that? Yeah.

Play from 0:55:15 Okay. So I I mean there are a couple that that I go to over and over again and then there are a couple that I mix up. So um let's go through. Let me see if I can find like a fresh batch where I've like run. These are after I've upscaled them. Jesus Christ.

Play from 0:55:42 Okay, so uh dribble dribble.com I love I always run dribble.com and I always love dribble.com character. Um love to see kind of like the character aspect of it and then dribble depending on like the sref will pull in either a character or it'll pull in like just sort of like a cool little art scene or something. I don't whatever it is. Um so I always do those. Let me find like let me find like an actual like dope ass what I did for I've done I've done Bee hands before. Um Beehance always gives me a Porsche random. I don't know what it is. It's always Porsches, you know. So So like this is after dribble.com dribble.com character. I always like an album cover.

Play from 0:56:27 I always do album cover. Album cover. All right. Sometimes I'll do movie posters. Another good one. After that I really like mix it up. Um, and I think like what I've been doing a lot of lately is I'll I'll just put in songs. I'll go through my Spotify and put in song titles. And like like on that last that SREF I was showing with the videos like I did like a whole AC/DC run where I was doing like Highway to Hell, Shoot to Thrill, Thunder Struck, you know. Um, stuff like that. Um, and like song titles can give you some good stuff, you know? So like I was going through and I did like Dreams I'll Never See, Flirting with Disaster.

Play from 0:57:07 Those are those are songs from like '8s. Yeah. Um and then I was doing like some stoicism stuff. I did like Momento my Amorati stoicism. I like doing Nike ad creative from time to time too because Nike usually comes out pretty good and it's always interesting to see how an SRF adapts to that. Yeah. um did Greek mythology here. But you run with your profile on. I do. I do. I run with the profile on. And then I will also one one thing I really was liking is I went back to remember core, wave, and punk. Totally.

0:57:40Using songs, moods, and weirdcore to test SREF

Play from 0:57:48 And I just started running those oneworders like weird core, dreamcore, dystopian core, like whatever the [ __ ] it is. And I just went through my whole list of like core and wave and punk. And I could throw those in. And so then what I'll do is I'll do basically the like here's how I build it out. Just like I'm going to do this. I'm going to do this. I'm going to do album cover. I'm gonna do um morati momento mory um weird core uh dream core rainbow core I mean whatever it is right so then I'll usually run like after I

Play from 0:58:36 find the srfs from the test right that I'll run with the skull in like R40 or whatever it is I'll like hide all the ones I don't like then I'll go back the ones that I do like I'll you know I'll kind of like call that list again I'm like oh which are the ones that really stick out and then I'll run this I'll run like something where I'm running like probably 10 or 12 different prompts so I would go longer than this typically I usually run a R45 and then now what I'm doing now with exp is I'll run this at zero and then I'll also run with 15. Damn. And that's fun. So, so then you get like sort of like that aspect of like asis and then you also get this aspect of uh experiential but it like I noticed if

0:59:15How --exp interacts with --sref (and when to dial it down)

Play from 0:59:21 you go too high on the experiential it starts to deviate right from the sref code like it starts to kind of like not completely ignore the colors but on but it'll kind of like toss those aside. So like 15 or 10 I would recommend if you do it that way and then I'll just usually run the [ __ ] out of those. You know what I mean? And then out of that list then I'll just go through and I'll say okay I'll look at them first and I'll say is this even worth sharing? And like I'm looking at this I'm like ah probably I don't really care about that. So I'm just going to skip that one. Okay. You know then something like this like oh okay there's some like sick [ __ ] here. So I'll go through upscale the ones I like. Then I'll just consolidate those into a group. And then that's usually when I'll, you know, pump out one of those posts where I've got like the [ __ ]

Play from 1:00:07 22 22 [ __ ] styles, right? I'm always like, how do you like 22 of them? So picky. Yeah. I like I always try to go with like the good ones, you know? Like I feel like I never get stuff that looks like that. And it's probably because my pro it's probably because my testing uh like my testing doesn't look that or testing. I just I go I and I also go through a lot of these, dude. I go through so many of these. You mean because you'll you'll look at some and you're like, "Oh, I like that one." But then like when you get down to it and you're trying to decide which ones you're going to run with like those permutations and [ __ ] you're like, "H, it's not really that cool." So that's when you kind of get to the good stuff.

Play from 1:00:51 And so it's it's a whole [ __ ] process. I wish I could say it's really quick, but it takes me a long time to do cuz I first I got to identify a bunch of good ones. And then I got to expect that like half those I probably won't end up running and I won't like. And then you got to run all the permutations and go through and like pick out of that bunch like did I run the right prompts? Do I like the results I got or not? You know, and then compile them. So I mean it's kind of a just putting that PDF together. It's a lot of work. It's It is fun though. Like I I I definitely do kind of enjoy it. It's uh it's kind of cool. But I you know this new SREF system in case you guys don't know this is where I think it's a little bit confusing and it's going to be hard for some people is because the the new SREF system right broke the legacy system.

Play from 1:01:38 Meaning if you were to use this code um and not attach d-sv space 4, you're going to get something totally different cuz because now with this new system the codes are related to a different style. Mh. And there was I like it was such a weird you know cut off. So like the cut off the only cut off I can think of in my mind is the date. It's like, well, it happened on 616, but like what the [ __ ] does that mean? How am I going to know? Is this a legacy code or not? Right? So, I think that's like maybe a confusing thing that, you know, could use a little bit better sorting and um I totally missed that, you know, but but

Play from 1:02:25 that's but that's that's the thing, right? And that's where I was like, I was already pulling those 22. And then I'm like, I see the new SR. I'm like, [ __ ] Because now I gota explain, you know, these are codes, but make sure you plug in the SV4, otherwise you're going to get something [ __ ] totally different. So, I don't know. They they do need to figure something out with that uh specifically so it's not crazy confusing. But do you never run Can we run Can we do Can we do one real quick?

Play from 1:02:53 Let's do one. Let's find Just find a good code. I'm curious if you like Yeah. I just I just ran a R40. So, let's do just find one that we like. I just want to do permutations SV12 334. Let's see if it works. Oh, I'm just curious. Okay, cuz it people forget that you can use SV12 33 4. I do it. I remember every once in a while. And sometimes those S SV3 and SV1s are bangers. Like you see like none of these are like really dope, you know, to me so far. This one's kind of cool. This one's kind of cool. Kind of.

1:03:00SV6 vs SV4 — why it matters now

Play from 1:03:27 No, no, no, no, no, no, no, no. You know, like interesting. Not sure. Right. Like these are these are the ones I would probably like hold on to, but I probably wouldn't run like kind of interesting, but not like one of those things where you're like, "Oh, I got to run this and see." So, that one's that one's interesting. How did you like that one? I like the skull teeth. I'm going to Let's run a fresh 40. We got hours to burn here. Let's go. Let's run a fresh 40.

Play from 1:04:01 Let's do I don't know if it's just me on this either. This feels um maybe I'm in my own head on this, but when I do an R10 and I run that four times versus an R40, I feel like I get better things during the R10 than the 40. I kind of feel the same way. I don't think that's Is that [ __ ] up? No, I think it's real. I think it's also because you can see them in a more digestible. Like when it's 40, I just want to like get to the bottom, bro. This like give me like something like this. Something crazy. This looks I could [ __ ] with that. So So hit that one with the with the crazy neon colors.

Play from 1:04:42 Okay. Use that prompt. Okay. And then I've definitely hit the high motion instead of use prompt like a million times already. Like when I wasn't expecting to. Right. I know. I was just I It took me a second. It's like muscle memory. Um, if you do permutations on d- SV and then do 1 2 3 4. Let's just see how different. Do you do you They haven't said this, but I mean, is there an SV six or whatever the last number is now that we've got this new system? I mean, is there I'm confused. Let's run five.

Play from 1:05:20 All right. So, let's run this. We'll run this first. All right. So, with Why is it saying SV6 can only be used with SV4? Got it. SV6 for version seven. Wait. So, try. Wait, what? I'm confus I'm so confused with this now. Yeah, I am too.

Play from 1:05:47 So, one, two, four, and six, three. So, it was saying four, but I think Huh. Okay. Let's Well, first let's do this. Can we do this? No. Okay. So, what it's saying is I think one, two, three are those they either don't work or they have to be used with V like a V6 model, I think. Okay. So, I don't know. Do you still want to try that? Yeah, try it.

Play from 1:06:17 Okay. So, let's let's just see. Okay. So, let's try four and six. Should we try five? Like, what is there? [ __ ] five. What did it run? It ran six and it ran four. There is no five. Okay. Or at least five doesn't Yeah, at least five doesn't work if there is one. What's the difference between six and let's see.

Play from 1:06:43 Okay. Whoa. All right. So, four and six. Way different. So this is So then this is it, right? This SV6 is for the new codes. Huh. Okay. Interesting. SV6 is for the new codes. I didn't either till till just now. So here's a question. This is Yeah. Can we go Can we do old SREF codes if with SV6? Do you have one?

Play from 1:07:09 Well, you can, but it's not going to be the same style. Huh. You would have to use four. I see. Which is why that that that was why I was like that 22 is like man I've got some good codes here but now I have to like go the extra mile to explain like these aren't going to work unless you do the-s right because if you just because SV6 now is the default right so if you just plug that in and you don't put SV it's going to default to SV6 huh all right I didn't know that but like let's let's take but but let's take this code now and let's run like just for the sake of it. Let's just since we were talking about it, let's run like an experiment here of maybe like a typical thing I would do. Let's do movie post. Maybe let's do

Play from 1:07:55 poster. Movie poster. Let's do cosmic dreams feels like a good one. Um, let's do like a 3D shape or something like that. I feel like I feel like you Yeah, I used to do that. Yeah, 3D shape. Um maybe 3D shape melting. Hedonism is another kind of cool one. Um let's do like some of these core ones.

1:08:00Building smart prompt batches for testing

Play from 1:08:21 Weird core. Let's do like um Dreamcore. Myth Punk. Myth Punk. I remember that one from back. Carpet Punk. Carpet Punk. I remember that one. Um what's the name of the episode? Carpet Punk. Yeah, there you go. Momento my wabishabi. lobby and let's do a Nike ad creative. And then what we'll do is we'll run that AR45 and then let's do the let's do the AXP.

Play from 1:08:54 Let's do a zero. Let's do a 15. And bang on. Boom. Let's And so and so the shortcut of that too like I don't know how other people are doing this but if you are like mining then then I'll copy this right I'll copy this part and so then once I've gone through and identified the ones that I want to run right I just paste that back in and then all I got to do is click the the code and then run it again. So all right um let me make sure there's a science to this. There is I'm just going to run turbo.

Play from 1:09:32 It's weird because you don't you don't really go for like I try to find my my one test term that I like to use a lot is samurai because I know it's going to give you a lot of detail. Yeah, that would have been a good one. Samurai is always like it always gives me detail so I can see if it's going to be minimal or if it's going to be or it's going to be like maxim maximalist or whatever. But all right, we got some we got some interesting stuff here. Oh [ __ ] Yeah. And you'll see like you'll see like these are different too. Like the reason I like dribble too is because it it tends to be more of like the um the illustrative 3Dish kind of style and then these other ones are a little bit more sometimes a little bit more photo realism based depending

Play from 1:10:19 on the SRF. But it's always interesting. You got to kind of work with the SRF what it's giving you cuz this one wants to be close up and it wants human. There you go. See? Yeah. So, so like that's the stuff I'm like the fact that I got two like what I take away from this. I've got two skulls here. Two three Oh, wait. Right here. Okay. So, I've got two skulls here. Um I didn't prompt for a skull there, right? So, that's interesting. Also, to your point, a lot of close-up stuff. Half these are close up, right? So, that's telling me something. Um to your point, now we're seeing all these faces and they all kind of have like this tilt to them. So, you know, like that's baked in to this SRF, right? And then especially here, I mean, this movie poster thing, I can't tell

1:10:20Describe tool to reverse-engineer SREF aesthetics

Play from 1:11:06 you how basic this is compared to most SRFs. Usually, like the movie poster thing, you get some really interesting things. Same with the album cover. This is so like this is actually probably not a great SREF. I mean, the colors are awesome, but what I'm getting back here is not all that interesting. Now, here's a here's a thing. Try see see if you go back to the uh or just try like movie poster, but run it with like 2:1. Just see what happens at like 2:1 aspect ratio instead of four or five. Sometimes I find the SEFs change like I always run them in four or five to test them because I just want like the to be in a nice comfortable grid across. But sometimes like changing the

Play from 1:11:53 aspect ratio totally changes the aesthetic. I don't think it's change it with this one that much, but yeah. No, I I like I like where your head's up. Let's run that. But I think what we should do, too, we should find a We should find a better one because I do want to show like the variability that I typically get back with these. Like this is interesting. I would run this for sure.

Play from 1:12:19 Interesting. That one's going to be a good one. Yeah. You know, you just tell which ones are going to be good. That's what I'm saying. it, you know, and that's that's what the hard part is because like you get kind of get distracted by these other ones and at the end of the day it's like what are the ones that really [ __ ] pop out don't even waste your time with the other ones. Um, so like I think we should run I think we should run this one. I feel like this would be a really good one to run. So I copied, you know, I had that copied paste. Now we're just going to add this.

Play from 1:12:51 Boom. Let's run that while we're waiting. Now, let's go check out what you said with the movie poster aspect ratio and what changed with that. Nothing changed. Yeah. So, I mean, close up. So, I mean, it's a it's kind of a shame, right? Because these colors are sick. I mean, this is acid trip type stuff. Really cool. I'm going to use this one on like a low SS SW like That's a great point. That's okay. So, that's also a great point, guys. Great [ __ ] point, Rory. Because when I'm doing this, I'm not I'm really not changing the weight. Default's 100, but this this maybe is a great example of you like the colors. These colors are so probably strong.

Play from 1:13:32 It be interesting to lower that maybe down to like a 50 or 60 and see what happens then too, you know, like do you do you break out of that? And I think maybe what I would try is let's go to one of the ones where we wouldn't expect to see just faces like so this movie poster again. Let's go back to this. Let's take this and then now let's run SW50 and maybe let's do like a 40 and let's do like a 70 and see what comes back and then we'll and then we'll dive into that other one here.

Play from 1:14:10 I just got a I just got a 10% notice on my uh on my my battery power. So I might I might have to dip for two seconds. Oh, this guy totally for totally forgot that I did not have my computer. That's all right. That's right. So I mean, okay, so long story short, this is showing like even with lower weight, we're getting colors, but we're still getting them close. Yeah. Kind of a bummer. Kind of a bummer. Kind of a bummer. It's a really good one. Maybe this was the other one.

Play from 1:14:40 This was the other one we ran. I think this is like a This is like a typical example, right? like you're going to get some interesting things here. This is where like adding the XP is nice. You can also rerun these. So, you know, sometimes if I really like kind of where it's going, but I don't love any of these images, I'll reroll those. The characters are always interesting. There's usually a good one in there. Um, album cover. I would say these are kind of weak for an album cover. Usually I get some better stuff than that. Movie poster. Like, there's some I mean, there's some good stuff in here. It's not bad. Um, it wants to be it wants to be landscape, right? Like it wants to show landscape. It does a little bit.

Play from 1:15:16 Yeah. Hedenism, 3D shape melting. Now they, you know, getting some kind of cool stuff there. Weird core. Weird core is an interesting wild card. Dreamcore, myth punk, carpet punk. Dude, you you get so many more illustrated styles than I do. Like I feel like I always get like these these random It's some really weird photo realistic stuff.

Play from 1:15:45 Even with this code though, you think? Maybe not with this code. No, no, no. Just like in when I run S refandom in general. Oh, okay. I see. I see. But um yeah, man. So, I mean these describe and see what you get. Like I always try to like call those neon ones. I just pick one. Pick me out like the Yeah, like that. the girl with the purple face like close up like what's it going to say?

Play from 1:16:11 Like what's it going to how is it going to describe that? Cuz sometimes it's an easy way to sort of take some things you like about uh you know your SRF codes and sort of Smart Man. Smart. Yeah. So like hold just just run all real quick. I'm just curious if anything even comes out. Yeah. like close because sometimes you run. See, there's that's that's why I love this [ __ ] tool. See, like I I wouldn't even been thinking about the scribe at this point. Um, super dope idea though cuz sometimes like when I get that I'm like I don't even know what that art style is or how it's like cuz I might like some of it, not all of it. Or the color like what what did it say for the colors, you know? I like that color

Play from 1:16:57 scheme. I can go look up the hex codes, but I don't know this one. Yeah. like one of the like one of them was really really I think it was maybe this one but it was just like electric neon purple. Yeah. Um so they don't really get it. See like they don't even they don't get that close with describe but not without the SRF. Yeah. Certain things to pick up. I always whenever I see one that's like so random that I don't know what it is. I'm just like let's see what Midjourney thinks of it. What does it think it is?

1:17:00Style reference vs subject leakage test

Play from 1:17:29 Cuz I don't know how to describe it. I don't think they call out the fact that she's got a purple face. No, that's or whatever the color of her skin is, right? They they should, but the color is not as good as it is in the SRF code. You know, the SRF code. Damn. I wish I mean like this this code's not bad. This code's not bad. I just wish we would I wish we would have had a banger for for the episode's sake. But I think, you know, live this is what happens. This is this is it. But I think like this is this is really good. I'm I'm interested to see two. We didn't really talk about this. I know your battery's on low, so we'll kind of wrap this baby up. But like how this affects mood boards, too.

Play from 1:18:06 I want to revisit that because that was also supposed to be a part of this, right? Oh, mood boards supposed to have improved. I think they have personally without even them saying without even knowing that. Remember we were saying I think we were saying this again. I feel like I'm a broken record here talking about last episode or the other episode. It was like these they seem to work better as of recent like more consistent, less glitchiness, less softness. Um I'm trying to think of like maybe maybe one I didn't love to but to try but I don't know. Yeah, these are these are worth going back to. I'd love to test these out. I haven't I haven't added any.

Play from 1:18:48 These are old. Really? Yeah. What was it what was the character one with all those little Which one? That one avatars, too. Hey, I found I found like these cool little guys. Uh I think this was more of a test. I don't I don't haven't really used this, but um you know, just to see if I could like generate more of this style of character. Um I just thought these were kind of cool. They kind of felt felt a little like NFTish, you know. Yeah. The other thing have you noticed with style reference? The one thing I did, I was running it this week and knew noticed that it wasn't as they said I think I think I read to be quite candid everyone I totally missed the SRF update because I got so excited about video. Didn't even didn't even know the SRF thing happened until I saw a Drew post about

Play from 1:19:33 it and I was like, "Okay." But there's no subject leaks. That was one things they called out, which we were always we were always saying, right? It's like, "Yeah, the subject leaks in with the SREF. it leaks into the image. Yes. I never heard them ever talk about that. No, but then then they just updated it and said that we're going to fix it. Then this one's out there and I've noticed it cuz like I I was using like I was doing a testing it. I was doing like a Nike took a bright white just plain all white Nike sneaker and then I used a jungle photo as a you know as a SREF to see what would happen and it didn't pull any of the jungle element in which I found interesting. That's not normally how it would work. It would bring something in of the jungle. Yeah.

Play from 1:20:20 You'd have to really dial the weight and play. Yeah. But it just took the colors. So it just really took like the like it had like some you know like the jungle had like some like you know like haze or fog in it and then it had like really sharp green and really dark black. So it brought all of those elements in but not the I'll see if I can find it before we bounce out of here. But it was just an interesting one that I noticed that would normally that would leak some some jungle you know some jungle stuff in.

Play from 1:20:50 Yeah. I was surprised that it didn't. Yeah. I I I haven't yet tried that part from a third-party style reference yet because all I've been doing is with the with the codes. Um you know what people are still confused by that. People to this day are still confused that style reference there's two two versions, right? Yeah. There's Midjourney style codes that you can unlock or use. And then there's just any other image on the internet that has colors that you can bring in and use, you know, to guide the style or the color or the aesthetic of your image. Um, but I haven't tried I I you know I think that's probably something that we could

Play from 1:21:36 try for next episode is is um how that's now working that subject leakage, you know, diving into that. And I'm curious to see how that's how that's working. I'll show you what I'm talking about real quick here. Yeah. So, this was this was interesting to me because typically I would I would really expect this to come through. This was the reference image, right? I'd expect some of that. Yep. To come through, but it just kind of came through with the green, the black, and the smoke. Right.

Play from 1:22:06 So, that was interesting because I was running a bunch of then I ran image. Dude, I like those. I love those colors. Yeah, it was really cool. I was just like I wanted to play with. So, this again just didn't bring any of the structure over. It just took sort of like the just taking like the vibe, right? It's not, you know, these aren't like my favorite shots of all time. I was just trying to test this with this one jungle shot cuz I was like, you know, cuz now you can use But that's a really good test though because you you've got a very like very simple isolated subject here, but your your style reference has all these leaves, right? Like it's not just like this green color in the im. I mean, it's got it's this whole object texture thing, right? And it's basically just taking that color, which is super interesting. So, I think this kind of is

Play from 1:22:53 proving that point of the subject. I mean that's I I really want to play around with that cuz then if you use you know was I think this was Lref maybe if I'm not not mistaken this was no I use white car in a studio I can't remember which I was trying to like do it through like four different ways to see like oh does image prompt do it you know this one oh so this one I used of ref with the sneaker and then this as an image prompt then like background as an image prompt and then it worked brought it in no problem. But I was interested with the with the style reference cuz it was giving me like a lot of variance.

Play from 1:23:30 Like it wasn't it wasn't consistent. I was trying to sort of figure out what the consistent sort of look was, but there wasn't much. So, you know, even here it's like style weight 900. I tried to dial it up to see if it would bring in the leaves. It didn't bring in the leaves. It brought in some green in the grass. One other one other question I want to ask um next office hours is click on any one of those images again. Oh [ __ ] you just Hold on. There we go. Hold on.

1:23:45Personalization wish list: can ranking images help?

Play from 1:23:58 We're going back. Okay, one second. I don't want to crash your computer, but I I'm I I am curious and I would like some some goddamn transparency around this. Okay, click into one. Okay, so help us improve V7. I like this image. I don't like this image. I'm neutral in this image. Can this factor into personalization? Does it? Can it? Please. You know what I mean? Like this feels like a great opportunity.

Play from 1:24:27 I I'm already generating a bunch of images. I don't have to like it, but you're asking for the feedback. So, how you know I understand that it's applying to the model as a whole, but can't you can't that also be applied to me my own personalization? you know, um because you know, we've we've talked about it. People have talked about it, aired some grievances. They they maybe don't like their personalization now with V7 or whatever it is. I'm not alone. [ __ ] all [ __ ] all y'all. Um but you know what I mean? Like they there's something there changed. I just think like that's a great opportunity to to impact personalization, too, because you've got got all those images right there. You're already clicking into them. I mean, it'd be a great way to just like I don't have to go to a whole another tab just to rank images. And I I know those are random images, but can't I apply that to

Play from 1:25:12 my own images that I've created? That'd be nice. Be nice because, you know, whatever. I haven't ranked images nearly as much as I did with I haven't either, but I haven't either. That's it. I mean, dude, we got I have 3% left. Sorry everyone. It's a good It's like a natural cut off for us, bro. This is a natural cut off cuz we we could go we could go for a while. This is Midjourney fast hour. You know, we already push it. We always go a little bit over the hour.

Play from 1:25:39 We're going, you know, now we're an hour and a half. So anyway, I think this is a good place to wrap. We still, I think, have some things that we could roll into to next week, too, to cover to if we wanted to. I think, you know, um I'm just excited. This is the best I've this is the most excited I've been about Midjourney in probably a year. Yeah. Is I mean, the video stuff. And you know what I loved, Rory? It wasn't this long dragged out thing like Yes, they did.

Play from 1:26:05 They have been talking about video for a while, but it wasn't like one of those things like V7 where it was like, "Okay, we're getting close. One, one week away, two weeks away, whatever it is, and then you got a setback." And you know, like this just kind of hit. It was just like, "Oh, it's [ __ ] on." You know, and um I don't know how it went from like V7 falling behind like eight months to then we did video, right?

Play from 1:26:32 Right. Like I find it crazy that m I know it's a a priority for them but I still feel like it's a side project and like their side project just like awesome you know like I use runway for images now and I use mid I'm going to use Midjourney for video like what like what world are we in this is so so odd but you know this is good job good job on the team for getting this out they said end of June they delivered yeah you crushed it. You got I mean if you listen to this, maybe you do, but you probably don't listen to the whole thing if you do. I don't know. But hey, I think really outstanding job. I think the sentiment has been echoed everywhere I've seen. Not many people I I really

Play from 1:27:20 haven't seen any slander about it. I think people have been very realistic about what to expect from a V1 version of this and it looks damn good. I mean, it's it's really promising and it's exciting times over here, man. Very exciting times. What's We got a lot We got a lot going on this one. We're going to come back probably next week, talk about it. Maybe we'll bring some guests on. Maybe we'll do something else. Who knows? We don't know. We don't know until the day we do it. So, appreciate you all for hanging with us. If you haven't already and you if you wait around to this part of the podcast every single time, like and subscribe. God damn it. But, you know, hopefully YouTube doesn't ding me for that. But we'll be back next week doing more stuff. Drew, crush it. Yes, sir. Rory's

Play from 1:28:05 beard will have grown a couple inches by then. I I had that, Dude. I was looking back. You looking clean shaven, my friend. Look at the Look at the last week. I was I was like a homeless person. So, we're we're back now. We're back.