Transcript · May 26, 2025 · 14.7K words
Ep. 40: Veo 3 Makes Yesterday's Best Look Like a Beta Test +Runway & Midjourney Check-In
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.
0:00:00When did this madness begin?
Play from 0:00:00 What's up everybody? Midjourney Fast Hours Podcast episode 4040. Whoa. Yeah. Whoa. Yeah. Wild, right? I don't even remember when we started this. Was it Was it June or July of last year? Uh, man, I want to say it was like right I think it was like that first week of July. All right. So, right after July 4th or It feels like we've been doing this longer. It's crazy. It hasn't even been a year. I mean, considering like all the things that we've gone into. I mean 40 episodes of diving into primarily Midjourney hours of stuff you know but but like really getting in I think to like at a whole like this just like creative process when it comes to AI and the capabilities the use cases
Play from 0:00:45 the strengths the weaknesses the roadmap you know the promise I mean so it's just like it it's been crazy that all that's happened in less than a year and I think we're in this just like crazy compound effect with with AI speed right now and I think it's only going to get crazier in the next few years. So wild times to talk about and we always have stuff to talk about. That's what's crazy is that this we could do this once a week and we probably don't even have enough time to talk about all the developments that happen which is you know also for anyone that's been here since episode one shout out. You guys are hell yeah. Thank you.
Play from 0:01:23 Hell yeah. Come a long way since episode one. I'm scared to go back and and revisit episode one now. Oh man, that's a you know, it wasn't I listened to it maybe a couple months ago. I was like, it wasn't as bad as I thought it was. Yeah, I remember too. It wasn't as bad, but No, but it's still definitely like we we've got more of the structure down, you know, we still have to work on some of the things like uh like our interview skills and like figuring out how to make that a little bit smoother. But hey guys, we're trying, you know. Hey man, it's it's like this is this is our favorite part of the week. Um, you know, Rory and I, it feels like you and I have only gotten more and more busy. So, I think the fact that we get to catch up and and have this sort of like nerd out session is is awesome. But, um, you know, I think today, right, like last time we covered, you know,
Play from 0:02:10 omniref parameter, we went really deep. We did a lot of experimentation. We showed you guys a lot of visuals. If you haven't seen that episode, highly, highly recommend it. It was probably one of the better ones we've done with visually showcasing kind of the different uh observations and capabilities that exist with those things. Uh for this particular episode, we were talking just about video. How could we not, right? Like video at large across AI, right? Of course, Midjourney's got a part of this uh that's coming. We haven't seen anything yet. There is an update there that we'll run through. But at large, I mean, the pace and velocity that we're seeing with update, update, update, and then we I I feel like we got this just in crazy level up milestone moment that we just
0:02:19AI video is finally getting spicy
Play from 0:02:57 saw with V3. Totally. Big time was. And if everyone, you know, you've been following us for a while, we've been a little bit behind on getting episodes out. So that's an apology basically from me because I've been gone for at least a month. So between between both, you know, conferences and other things that are going on, but it's like we're back. We're back now probably for the summer and we're going to be continuing to do this. But the I was away and I was like I got rumblings that Google was going to drop this whole new creative suite and I was like, "Oh god, please don't do it while I'm away." And you know, luckily it happened on Tuesday and Google just again pumping out great stuff. Now that being said, with this new release of
0:03:29Google’s Flow Suite: Veo 3, sound, and coherence
Play from 0:03:43 what they're calling flow, essentially like a creative suite, right, where you have access to all their new tools in one place. So you can use VO3, which is the new video model. You have Gemini 2.5 which is you know available within it's actually with I don't know if it's within flow but uh there's image 4 now their updated image model so their whole goal is to generate you know a creative workspace that you can do anything in and the real kicker and what the big uh big update was is not just the improved quality of the videos but you can now add dialogue and sound which is just again things you don't think you want and need until it's there and then you're like I I want that and I need that but that causes a whole another set
Play from 0:04:30 of issues, right? Yeah. So, yeah. More variables introduced. Yeah. And even like that from a consistency standpoint, like getting a voice consistent from one video to the next, like now that you have to control that, you know, comes a little harder. So, it's just again I still you're starting to see I think we're going to have a consolidation moment sooner or later. you're starting to see like Google was behind, right?
Play from 0:04:56 Like it felt like they were behind last year and now they're just like pushing ahead. It kind of felt like Google, what are you actually doing last year? Like I they were so quiet for a long time to the point where you kind of felt like it Google I know all this money's coming in, but are you what's going on over there? Are you guys dying? What's the deal? You know, because I mean they had all the the SER, you know, the obviously like the SER uh disruption with with LLMs coming into play, right? Like the way people search and get results for things. I mean, yeah, dude. I It's It's wild. Their their pace right now is is really picking up. And like this is where I don't know, and I'm being candid about this, like who can keep up with them
0:05:02Google's confusing product soup: Flow, Gemini, Imagen, Whisk
Play from 0:05:42 maybe outside of like the meta realm or the Tik Tok realm because it's just about data. Who has the most data? Google has, you know, basically every image that's listed on Google is categorized into, you know, it's going to be probably filtered into their image data process. I'm guessing they're doing the same thing with YouTube. So YouTube gets what, like 30, 40 years of content uploaded every day. So again, you know, if you're not familiar and you're on YouTube and you have videos, I think there's a space you can go off to check, you know, like I don't want to be part of video training, but I don't know if anyone's how many people are doing that.
Play from 0:06:16 Where is that at exactly? It's somewhere in there. um somewhere buried. That's you know our videos are probably in there for the talking head videos that you can now create with V3. But I just don't know who else has that data set other than someone like Meta or Well, I think I think that was interesting because I I you alluded to that I think before we really even saw V2 come out. Uh like I remember you talking about this idea, right, of like who who's got more training data than Google, right?
Play from 0:06:44 because you've got all the stuff from Google itself, the platform, search engine, images, things like that. Then you've also got YouTube as a whole. I mean, um, and those are those are just the bigger pieces of the puzzle, not not any sort of like fragmented products or companies that they've acquired or bought or integrated along the way. I mean, so yeah, I thought when you called that out, it was it was interesting and I and I was nodding my head like, "Yeah, that's totally true. I could totally see that." And now it's really come to fruition here cuz you got to think about where Midjourney has to source images from, right? Like it's not native on their platform. Like Google's sort like basically sourcing stuff natively from their platform. Again, ChatGPT probably going to be able to do this just based
Play from 0:07:30 on the scale and size of who they are. I don't know what happened to Microsoft. I guess they just uh like disappeared and like haven't done anything in this game. you know, you're thinking about the text, the big players in the tech space and what, you know, they're able to do. But Google, you know, this feels like perfect for them. Now, I will say I do have a gripe with Google because they seem to have like 1 million products and they're all in different places and they all have different names and something like ChatGPT becomes so much easier because it's like, oh, I can do everything in one window. I know where it is. 100%. Right? Like that's my one gripe with Google is that there's just so much I don't even know where product overload and and the naming and all that. That's that's such a I I notice that all the time as a marketer, too.
Play from 0:08:15 It's funny that you bring that up because you'll even see that in in smaller SAS companies that have more than let's just call it two products, right? It's like after the product that they're known for, right? Like you're gonna name seven other new products. It's almost like are are these just maybe features or can they be a part of this naming convention for the one product essentially, right? Um and so there's this like chicken and egg problem of is this the company or is this the product suite? Should it just be named with the pro like cuz no one else is going to remember every single one of your product names except for the people in your company or maybe the maybe the actual buyer, you know, at the finish line, the decision maker who's purchasing, right? Other than that,
Play from 0:09:01 you've got an uphill battle trying to promote two, three, four, five, six, seven, thousands of different products and people expected to remember those names and recall what they do. Yeah, good luck. That's what it's like. You go into Google and it's like where do you generate images? Well, there's Imagen 4 that you can go utilize. There's there's Gemini 2.5 now and then there's Whisk. And it's like why are all three? It's like the opposite problem of Midjourney where they have so much in one platform and then Google is separating. each one of these. So, it becomes what like I think they could integrate whisk into and then just it makes it a little bit easier. I just don't know why. And if people aren't familiar with whisk, it's basically like like an ingredients sort of builder like you you put in a character or a style or whatever and you
Play from 0:09:48 blend everything together, right? You can also do that in Gemini just by pasting a couple images in there and saying like blend this stuff. But it's uh it's really just is it are they calling this the AI stu like if that is the case it's like just call it that. Yeah, because then there's AI Studio, but then there's also Flow, which is Flow is where all this, you know, the V3 stuff takes place and then and then and it's like, okay, is that part of this, right? Like again, it's very confusing to your point. It's they just they need to like they're so good at product dev, but then like making it easy for everyone. And then I don't know if I don't know if I have it for personal or for business. I like this one. This
Play from 0:10:33 one's a pain in the ass is I I signed up for flow with the wrong account. I wanted it for my business account, but it was like log you know how when you logged in as a certain profile and it was my personal Gmail and I'm like I don't need it for that. I want the extra space. So now I have to go figure out how to switch my account to whatever cuz I paid just so everyone's aware I paid the stupid $124 a month um because I had to see what was going on here. And you know, granted, the other thing we should say about V uh V3 and Flow is that it's expensive as hell and you don't get that much. So, like I think it's 83 generations you'll run out of your monthly credits. So, you got to use it judiciously in terms of like we like I'm
0:10:45Pricing pain: Is Veo 3 worth the $125?
Play from 0:11:18 only using this for when I'm dialed. It's not a playground like Midjourney is where you can go and just like Yeah, you want to you're not starting from zero here. You're you probably have already got a few of these raw materials or ideas conceptualized in some way. Yeah. What do you um have you seen what happens after you run out? Can you top up? Can you like You can What does that cost? There's additional credits. Let me actually Let me see if I can pull it up so I can show everyone the pricing here just so we're all transparent about it.
Play from 0:11:51 Well, I think you're It is They've got a deal right now cuz I was just looking at it yesterday. It's like 125 for the first three months or something like that. Yes. And then it's I think two double. Yeah, I was going to say I think it's double that. Um after it is because then I'll go into my where I'm almost out of credits. Uh what's the it's a big it's a it's a big expense item. I mean, 100%. Because because when you're thinking about this, unless your company or agency is paying for it, if you're a solo person, this is probably one of the more expensive piece of this is probably your most expensive subscription you've got. Yeah. And they did, you know, they were they were nice and they threw in uh they threw in uh YouTube Premium for everyone.
Play from 0:12:33 So, your $6 a month there or $12 a month, whatever it is. But here's the uh here is the in the inner workings of it. So you know the one thing that I found is you can you know you see how many credits you have. I started with like 13,000 and I'm down to just basically the end already. So So what is a what does an average run credit wise take? Do you think I believe it's 150 per generation? And I think it's like 12 125 in terms of like 12,500 credits I think you get with that if I remember correctly. Yeah. So you have the options for the three models here.
0:13:09Veo 2 vs Veo 3: Best value tips and tradeoffs
Play from 0:13:12 My if anyone is going to use this and wants to play with it, my best suggestion is to start with either V2 quality or fat like uh their quality model. Oh, sorry. Their quality model which is it's a slower higher quality model, but it's it's so much cheaper in regards to credit usage. and then like figure out if the aesthetic or prompt works first and then bring it into uh three. Great call, right? Yeah. Just to save some just to save some credits because you can burn through this, but also you can add credits. How um have you made any uh quick observations on the coherence with two versus three? I mean, like I'm guessing it's it's probably not that different then. In that sense, it's it's definitely better
Play from 0:13:59 or it does feel different, I guess, is my question. Yeah, it's a little little less, let's put it this way, because it's there's so many additional features. You you kind of miss on some of the stuff that V2 did, which was like more extreme motion and does it very well. But with that being said, like the quality is better. Like the quality of the output is better. Like this one. You see, you can see my screen on Twitter here.
Play from 0:14:27 Yep. All right. So, like this one, this prompt, you know, low angle selfie footage of a fat man running and laughing simultaneously. Like, nails that his laugh mixed with gas for air. I don't know if I can play this. If anyone can hear it, can you hear this? No, cuz I think you got the the AirPods in, so it's not going to give me the volume. Fair enough. Regardless, it nails it nails this laugh mixed with gasper error. He has red hair, a mustache, a green sweatband. So, we get all that um nestled next to his receding hairline.
Play from 0:14:57 So, we have the receding hairline with his unckempt curly hair. The camera shakes as he runs, you know, with vigilance and his laughter makes the scene like comical and light, right? So, this is a simple example. Nothing mindblowing here, but just it listens, right? Y it listens. I mean I mean even with the image gen stuff that I played with in the past, the coherence was fantastic. Yeah, I mean you like it was really dialed in to the point where it's like that's what I also can appreciate about Midjourney in that sense of like I like that it has coherence, but it also we we've talked about this so many times, but leaves room for that creative magic to take place. Image Gen feels like it's much more just precise coherence, literal co coherence, I think, is is what I'm sort of trying to say. 100%. And then there's some there's some physics stuff like I had this guy
0:15:08Prompt accuracy and physics: Is Google really listening?
Play from 0:15:43 spraying kerosene on a grill and saying I haven't even seen these brother but you know this is this stuff was pretty I was impressed with some of that like that physics is pretty good. Um this one I had a little I don't know if you you're tuned in to the uh the side talk NYC for that go you know it's just like these crazy videos in New York after the Knicks play. It's just like you know the bing bong. I don't know if you ever seen that, but they just like interview on the street. They say people say crazy stuff and the crowd just erupts into chaos. Um, it's I'll share the account with you after if anyone's interested.
Play from 0:16:17 Yeah, there you go. But like this one I was impressed with. You know, like the camera motion, the way that he opens the door. Yeah. Right. Like just the way that it follows him, the way that, you know, typically that would be really hard shot um for this to produce. you know, something like this with all the stuff going on in the foreground, being able to preserve the midground detail, you know, the way that the water's moving from a physics standpoint, the way the truck is bouncing. Really impressive. It's like it's like almost like right with a lot of these video models at first, you you you would get like one or two variables that would kind of like give it a little bit of a realism sparkle. you could see it, but it like it it lacked punch because there were other variables that were either static or not rendering correctly.
Play from 0:17:02 Whereas this is like really stacking on variables, right? Like the the camera shifting, bouncing, the tusks going up and down, the proportion of the legs like it the legs are the walk is clean, right? Which for a long time was such a problem. Then you've got the person on top, right, bouncing around, the feathers bouncing. Like there just so many different things that happen in sync there. Yeah, like the the t the uh the trunk moving the splashes every Yeah, it's impressive, man. Like this one again, too. Same thing, you know, to the prompt coherence. I was really impressed, right? Like just, you know, camera pushes in on a man sitting at a New York City diner table at midnight.
Play from 0:17:42 Midnight might not be appropriate there, but his arms are folded, his head is down, resting on his arms. Suddenly, a hand holding a plate filled with a patty melt and a massive portion of fries, you know, enters the scene and slides the plate on the table in front of him. He raises his head to show an extremely tired look with bags under his bloodshot eyes. Long night of partying, right? Like this. It's just listening to everything I'm doing. I'm saying, right? And it doesn't The thing here that I that I'm liking is that I don't have to go so deep into aesthetic to get this look. You know what I mean? to do like 35 mm soft blur in the background like you know contrasting colors like leading lines with this stuff like I don't have to describe all that it's already I'm saying man I think we I think we went through a real OG era of um analog like AI right with the level
0:17:53Why less prompt effort = better results now
Play from 0:18:32 of context and things that we had to you know really apply there um especially after that first wave. Um, and I'm thinking like with this in terms of Midjourney, remember like you know when you got early on with I don't know V3 or whatever V4 like you probably weren't getting exactly what you wanted precisely anyway. So it was a lot more of like let me throw in these creative tokens and you know buckets to kind of get something unique that has a loose interpretation of what I've got. And then it went to right when it got to like five, you know, it was us really dialing in these, okay, think about this from like maybe even a photographer standpoint of realism. If you're going
Play from 0:19:18 to go for that, right, you need to be thinking about camera angles, maybe even throwing in a camera and lens, an aperture if you want, you know, shutter speed's an option, lighting, shadows, you know, textures, right? Like all that. And now, right, like that language is sort of like coming back to this state of um it gets it. It it gets it and right like there are a couple different ways to leverage that, right? Like you could still throw some of these things into a GPT if you need some help crafting prompts and stuff like that.
0:19:40Veo 3 vs Kling vs Midjourney: Prompting philosophies
Play from 0:19:47 But the point is is I think like the the analog era of kind of like tapping into each thing and really having to think hard and deeply about and revisit notes and things that you it's probably over, right? Yeah. And that's that's what's interesting here, right? Is there's there's so much so much variation in between each platform and what's utilized. Like I have a totally different way of prompting V2 or V3 than I have prompting Cling 2.0, right? Like it's to I know how Cling works. Cling has been fairly consistent from model to model in terms of what works and what doesn't. Um, you know, it feels like, you know, MidJourneys switched with their models in terms of what works, what doesn't. Like each, you know, each new version you kind of need to prompt it a little bit differently. So yeah, you know, for Clling, I'll give them their I'll give them their roses on that. It's been seamless sort of
Play from 0:20:34 transition each one of them and I really like it for that. Now, the other thing to note here, which is cool, because I was able to, this was something I was trying to do for a really long time, was these like longer oneshot takes where, you know, you just have like one single character and you're following them for, you know, 40, 50 seconds. I've been trying to do this for a while without having like the the sort of image or like the the scene breaks where it sort of like stops and then jumps again, you know, without having to do a ton of editing. So something like this I was like really impressed with that I was able to just take that first shot and then extend and then I have him walking through, you know, an entire environment
0:20:52Scene builder: Longer takes and smart extension workflows
Play from 0:21:20 here. He goes from a party to a dressing room, you know, then to backstage and then he'll eventually come out on stage in front of the audience. But being able to do this now, you can do this is the this is again the X factor I think here outside of like the quality and the sound is that you can now do this within um like a they have a scene. They have what's called a scene builder. So when you're in here, let's just say I like this. You can go add to scene and then you get a timeline and then you can from here you can click and go extend. So I can just extend this clip and keep it going, right? Right. You just when you hit extend, it's just like you type your prompt and extend the clip as long as How's the um How's the cons or how's the consistency quality of that? It's it's
Play from 0:22:07 good. As of right now, though, I want to make sure that I'm Are you having to prompt on top of it in addition to extend to get a little bit better results or kind? Okay. No, I've been I've been prompting to sort of get better results, but as of yesterday, and I don't know because I haven't run this again, but you could only extend with V2. So, let me see if I just run this.
Play from 0:22:33 Interesting. This shot continues. I'm just going to Does that affect quality in any way? V3 to V2 right here. So, switching switching you to a compatible model for this feature. So, you can't extend. So, this is like where the downfall is. Like this first video that I did had sound, right? So now when I extend, it turns into V2. So I'll have to use V2 to continue to extend it, right? Um which is either fast or quality and you'll get a drop off in the quality of the video and you won't get any more sound. Okay. So you will So you do notice the quality drop off. Correct.
0:22:34The catch: extending drops quality and loses sound
Play from 0:23:08 Yes. That's it's it's definitely fuzzier. It's less sharp. Um, you know, the colors are a little bit more. I found VO always has like this wants to be blue. Just Yeah, it wants blue. Hey, Rory, what about um lengths of these videos right now for VO3, right? So, okay, you've got you've generated a clip. It's got sound 8 seconds. You know, you you know, you probably at this point, I mean, unless you had to, right?
Play from 0:23:37 don't want to take it out of another take it into another tool to like, you know, manually add sound and manipulate it. It doesn't make a lot of sense right now. How long are these clips? So, they're 8 seconds each, but you can then jump to can also just generate a new scene in V3. So, you could also do that. You don't have to have you don't have to use this first frame, right? We can use high quality here. And you could just generate like a totally different clip, but you'd have to text prompt it to make it look considerable. So that's the other thing to know is that when I go to So no kind of like style reference for color matching or aesthetic in that regard or no. So the other thing here is when you go to uh when you have your drop down you have text to video which
0:24:17New image-to-video support + third-party images
Play from 0:24:23 is basically uh is V3 at the moment. Right now it sounds like you can do apparently as of as of like 10 minutes ago I just saw on on my WhatsApp that you can now do image to video. So you have first frame, last frame. Um, you know, within here, does the does the image does Google allow you to take outside images or does it have to be Google uh generated? Let's see. I don't know if it uh that's what I'm curious about. I don't even know what this image is, but we'll see. All right, cool.
Play from 0:24:53 Let's see. Crop and save. Okay, let's see if it can do it. Yesterday it wasn't Yesterday it wasn't allowing that. So Oh. Huh. So it is okay which is awesome. So you could take third party images in here for that frame to video. You can also add an end frame here. The other thing that's nice is we have camera control. So you can just you know basically you don't have to describe this in the prompt. You have all of your you know basic camera controls here as well. Uh now the interesting piece is this gets a little bit gets a little bit deeper. So you can go ingredients to video. So, similar to Cling's ingredients where I could add, you know,
Play from 0:25:38 that scenes different people or different objects, right? I don't know what the what the filters are in terms of um like I could add this guy, right? So, I can have a consistent character. How how does it how does it know whether to do you have to explicitly call out that you want that guy versus it maybe interpreting that scene as an ingredient? Does that make sense? Like I haven't tried it yet because yesterday it wasn't available with V3 and I'm just like I know how V2 works. Like it's fine. It does it does work. It does place the character into the scene fairly well. Um so the other thing too to know is like when you're in here, right? All right, you have the ingredients. Then you also have can use
0:25:41Ingredients-based generation and persistent characters
Play from 0:26:23 ingredients if a style's out there. Okay, got it. So there's also if you're doing you want to add an ingredient, right? It'll give you these options generate image upload or use something that's previously utilized. You can also generate and then you can go use image4 to get this, you know, like just uh uh character 1980s movie character like 1980s movie character, right? Like you can just generate here and then use this as a frame. So you can use imageen straight in here to generate frames as well, which I found to be useful. Imagen I've four like I'm not overly sold on it. I thought three was really good. I think there's some still has a look to
Play from 0:27:09 it now. Yeah. But you know, like this guy could totally use that and just start from there, right? And anything you want to do with these people, you just basically add this as the frame. Use this image and then there's your frame and you can start from there. Right? So that's one way to use frames. The other way is if you want to go into this video, right? Like let's just remove this from scene. Oops. Let's add the scene. You can be like, "All right, I like I want to start my next video from here or I want to extract like a frame like this." Like maybe I want to extract this frame. You can just hit save frame as asset. Oh, sick. Love that. Yeah. Which is super nice. That's a Man, that's such an underrated thing, right? Cuz how long
0:27:10Frame extraction: finally, a feature we all needed
Play from 0:27:56 have we been dealing with the fact that if you want a certain frame of any video that you basically just have to screen like pause it and screenshot it? Uh-huh. You know, and then probably lose a little bit of quality in the process. Yeah, cuz you can you can also drag like you can drag to sort of like the timeline. Like I said, you can once you start to add things here, you can arrange like I can flip the scene and put something, you know, in front or behind or get rid of it if I want to.
0:28:08Timeline editing, upscaling, and staying inside the tool
Play from 0:28:22 And then the other good thing is you can basically edit within here or uh you can export within here. So once once you have like a if I go to another one where I might have like a full scene built. Let me see if I have one here that's that's relatively um I don't know if I have one in here with multiple that I started but it's all right. Let's see. No.
Play from 0:28:49 Yes. No. Okay. Regardless you can download the entire thing. So you can just like once you add all these scenes in here, right? You can go and just hit download. It'll download your entire clip. So you have that functionality as well, which is great to have, you know, the fully edited thing in here. And again, you can trim it up and basically edit it how you want. The other thing that I shall mention last is that the the one native feature that is great is when you hit download, you have the ability to turn into a GIF original 720 or you can upscale it to 1080. So you have that ability in here as well. Upscale cost you additional credits or no?
Play from 0:29:34 I did it once and I honestly don't know if it cost me extra credits or not. It wasn't like an obvious thing like where it's like it doesn't say it now which I know like runway will right like on the video it's like upscale of 4K it'll tell you how many creds it'll cost. Yeah. So there's just a lot of there's a lot of stuff going on here. You can see how this is sort of like they're trying to take that Sora model. Now they don't have the things like Sora has like remix and blend and you know the timeline features in terms of little bit extra additional things but Sora is way less coherent than that video is. So I'm just like I said I'm just impressed with this as a as a concept for being able to stay in tool here and not have to jump back
0:29:48Sora vs Veo 3 vs Runway: usability and consistency
Play from 0:30:19 and forth a million times. Yeah. So, you know, that's that's where I'm seeing the best uh the best sort of what's the best way to put it like the best sort of like concepting for a tool yet from a video standpoint. Um, that's me though. You know, I think Runway does a great job. I think Clling does a great job, but there are also, you know, we're at the beckon of this. I say this all the time, like the beckin.
Play from 0:30:48 Yeah. because we've been with Midjourney or Runway or Cling for a long time now. They just keep adding more features and then it's like, all right, there's all this stuff and then like a new one comes out and it's so clean like this because they haven't had all that and you kind of appreciate it. Yeah, it's a dude. Totally. I I think it's the ADHD HD and and us and the wider population because it's Yeah, I I've always felt that way. It's like there's something magical about less sometimes. Um, right. Cuz it's easier to figure it out. You can connect the dots.
Play from 0:31:25 It doesn't feel overwhelming. I mean, I love Midjourney. The fact that there's still so many things though that I either forget about or haven't used in so long just purely because I forgot about it. I mean, it's it's mind-blowing, right? Like Yeah. Um, it's hard to use all of it. You know, like Canva used to be so simple and easy and I don't even know thousand things to do in there, bro. I don't even know what Canvas is doing now. I mean, like I I mean, from a business standpoint, it's it's great. From a user standpoint, I agree, right? Like that was the that was the tool we got, you know, um early on for non-designers, right? Or um you know, lowfi stuff or being scrappy small teams, you don't have a designer, you don't have budget, right? Canva. Yeah,
0:31:43Canva, Figma, Framer: Tools are becoming monsters
Play from 0:32:10 dude. You're totally right. Like, and there was something special about that. Um, now I I can't even I can't even fathom everything that you can do in there now. Um, I've spent some time in there recently, like in the last year or two, but not not taking advantage of um or not really paying attention to like all this new AI stuff that they like I don't really I'm not going there for that, right? Yeah. cuz I used to use it as you know when I didn't want to use Photoshop or something complex and I didn't want to like have to before Photoshop allowed you to just just point and click on a layer. You used to have to go select the layer then move it then select the other layer then move it which was annoying. So go to Canva and just like be able to move everything in one canvas quickly for things that didn't need like all the firepower of Photoshop, right? So I I don't know
Play from 0:32:57 where you know they've added so many things and a lot of times people will tell me, "Oh yeah, Canva can do that." I'm like what? miss that up and even if they can it's almost to the point right where we've talked about this it's like the adoption it's like am I do I see myself actually changing my habit to then now start using Canva for A or B or an A B and C now like just because it can probably not you know yeah there's uh it's it's unfortunate a lot of these older tools that sort of like got me through life like work life afterthought Photoshop is really becoming an afterthought now. And like Figma, I don't even know if you saw what Figma dropped recently, but dropping I've been hearing that they're dropping more and more, you know, and I can't keep up. I I
Play from 0:33:43 can't keep up. But they basically integrated like cursor into Figma. So you can you can do like web design and coding and development in there. Like I they have a new website like an AI website builder. Like I can just go build a website simply within Figma instead of having to utilize one of these template builds. Like if if you're not a seasoned web designer, right, or someone who understands coding, web design kind of sucks cuz you're just at the behest of like what the template can do and that's it. And now you can go in there and customize it the way you want.
Play from 0:34:13 It also has, you know, or just like a or just like a developer on your team like maybe there's just one or two people like that's happening where it's just like I'm glad we had that but that feels like a a big bottleneck here for us like that it's not this democratized thing that more and more people can act on because now now you're relying on somebody else to go in and make changes to every new web page you want to add or build or wireframe you want to reconstruct and this and that and the other and that's just not a scalable thing. Let me let me actually I'll just pull it up real quick so everyone can see it. I know we're off a little bit, but I know there's definitely people here that are using Figma. Um like I was just playing around with the other day.
Play from 0:34:53 I was like, "All right, I just pulled in a bunch of different things." And if I wanted, you know, a different footer or something and just go and basically add and then, you know, navigation, they have different heroes or features that you can add like, oh, let's just add this. Bang. You know, that just adds it immediately. So you have the web design here and then it gives you the mobile design right next to it which I like. If you want to add a page, you just go there and you can design. You know, each one of these is interactive and responsive. So I'm like, this is great.
Play from 0:35:21 I can just build landing pages on here all day. I don't have to go to, you know, don't have to go and like use these template builders on website, things like that. Might need a little I might need a little demo on that. Rory Flynn, dude, it's it's I I was really liking it cuz there's you know what's the other one was was it Buzz that I was using? I don't remember if it was Buzz, which is you can just make assets, right? Like you can just make it's like this is their new version of Canva with AI baked in. Ah, yeah. Generate templates and things like that. That feels that feels long overdue for them.
0:35:33Figma’s new AI website builder is wild
Play from 0:35:53 Yeah, super easy. The other thing they added was this make. Let's see if they have if I have it in the make. Do you see do you see um Framer and Figma kind of that's it conjoining here like in terms of like they're coming closer and closer together. I just typed give me a sneaker ad and then it dropped this which is basically like I wouldn't say it's an ad but it's like a card for a website. Yeah. Hell yeah. That's that's great. I mean the UX UI is like Yeah.
Play from 0:36:22 Perfect. Perfect. You know even then gives you the additional format. How about that? Right. So I was like this is cool you know and I was like what's you can just type in here and it's run through claw sonnet so it can code really well it's basically like using an MCP you can also you know drop in uh you can attach a design to sort of give it guidance you can attach an image which is essentially framer just be like you know here's a website that I like put it in like design this for me this can also work on a few different things so I'm you know I'm loving where all this stuff is going right now it's just becoming becoming a playground and I feel bad for people who are just starting because there's so much to to learn and do but you know they keep making these developments and it's one after the next
0:37:09Why training teams on AI is almost impossible
Play from 0:37:09 you need you need internal uh adopters that are just curious by nature in the company uh and you need multiple people at the moment right because most companies are not going to um spend at the level that they would need to to upskill their own employees, which is a whole separate other conversation. But also to your point, these things are changing so fast. So you could hypothetically, right, bring in a trainer or this or that and the other.
Play from 0:37:42 And that's good in that moment, right? There's a lot of value there. There will be things that will be evergreen for a while, but there will also be things that'll really perish quickly, right? Because it's changed. The whole thing's changed, right? like even just a couple weeks ago, right? It's like, you know, Hedra is like kind of like this cool little thing that you could probably, you know, start to adapt and like with V3, you're like, I don't know, you know, I don't even I don't even know anymore. Right. So, I mean, like the Yeah. Like I like for example, I'm just thinking of a client I have right now. They're they're doing a um they're doing a presentation with uh with basically like AI generated people that will serve as like an agentic part of their software.
0:38:07Hedra who? Veo 3 makes fast pivots a must
Play from 0:38:23 They're going to present at their conference. They're putting it up on a screen. They want to show motion. They want to show them talking and you know making movements and that sort of thing. It's like okay like they're already playing around with hedra, right? And then you got the3 drops, right? And you're just like, well, if you guys are probably trying to do like full body motion and you know, like get them in scenes and you know, like there's a pivot, right? So like you're trying to do a project like that really truly while the car is moving, you know, trying to paint the car while the car is moving or whatever that analogy. This is why it's tough, man. Like because I I've people have sort of I've been approached by a number of people for longer term projects like bigger projects and I'm just like if you're working on something for more than a month or two like basically your first month or two of
Play from 0:39:10 work is just just scrapped because you're doing it with tools that are outdated by that point. So you know it just doesn't make sense. Like it's it's it's hard. You have to be able to do things quickly. Like if you can do if you can utilize the tools and get to an end result quickly, then it's fine. But if you're trying to do like a feature film, Yep. Right. You you Yeah. What are you going to do? You're going to scrap halfway through, you know? And so it's just this Yeah.
Play from 0:39:39 Like I don't think we've ever been at this point with technology ever like that. Like you I I can't at least not off the top of my head, I can't think of a situation like that. Usually there's been these long gaps of time. Yeah. Yeah. And we couldn't do like a podcast on Apple, you know, like what are we gonna get every year, you know, the iPhone, you know, get the the launch of the new iPhone colors like and the operating system and like that's it, you know? So, it's it's this stuff is just such little micro developments at all time and then you get, you know, use something like cursor or lovable or framer becomes big in the zeitgeist and then Figma's like, I'm just going to pull it in. You're back on your You're back on your zeitgeist It's very It didn't It didn't feel It didn't feel like the right word until recently where I love it, dude.
0:39:55Midjourney's next move: what video could look like
Play from 0:40:26 All anyone talks about. No, I know. I know. I know. I'm just giving you I got to give you cuz you you'll pull out like a beckin, you know, like something every episode, but then you'll come back with something that like a zeitgeist. So, my brain just firing on early morning caffeine. We're a little earlier today than we normally are. So, that's true. That's true. SAT words are flowing right now. But this is the other thing that was interesting recently was again and I had this conversation with a few people is just runway right like runway is really creating a lot of useful features. Um and I I think you know some of it gets overshadowed by what's going on with VO. uh it's such a cheaper model to utilize the quality. You might sacrifice a little bit in the initial output, but the references feature is excellent and I think agreed really
0:41:11Runway’s underrated features and clever reference hacks
Play from 0:41:12 well. Dude, I'm so glad you brought that up because I I think that's something worth talking about too, right? It's uh that that exact point of what you just mentioned of the use cases now like they've built in different things that maybe aren't so obvious but you're like you're following other people that are using uh a reference and they've unlocked like a way that you probably didn't think of initially for this right and you're like oh well I didn't even think about like that whereas before it was a little bit more cut and dry you know with with kind of you knew what you could use it for um in the application but you're right like there's now sort of like some some of these gray areas where creatively it's now opened up. There are some different uh possibilities and use cases available with that uh with that references uh part of it, with the style transfer part
Play from 0:41:59 of it, with some of those other things that they just rolled out. Yeah, this is great. I mean, there was I saw I saw Crystal Ball their founder put up like a really good image of just someone that put like a basic framework of a structure in there. It was like car here, person here, this here, and it just it listens. And I was like, "That's awesome." I try I tested that. It worked very well. Just, you know, again, giving it loose direction. I'll I'll try to pull this one up. Yeah, it was it was it was kind of like that feel of um you know, kind of what we were getting with uh with Chat GBT, you know, with uh the sketches or like that wireframing sort of vibe to it, right? Yeah. So it was essentially like you know this is like the full sort of representation of it but it was just like okay man front for you standing arms on child front view
Play from 0:42:45 sitting on stool you know drawing the stool checkered four pattern woman side view standing body facing left head facing right so you know it's giving you all of this which is what I want even the checkered four sitting on the stool like the body position when you look at it it's all pretty much accurate and then using these two you know using this as a reference image these two people and the child. So that that to me I was like all right like we can we can do stuff and like even everything from the pattern on the scarf here. I'd say that's that's pretty well transferred.
Play from 0:43:17 Even you know the the jacket the color of the shirt that stuff matches her look. The sunglasses match his look. The sunglasses match. Didn't put the hat on him regardless. I'm not going to complain about that too much. Probably could have just rerun it and got it the same. But you know that to me it's impressive. like this is a this is a really good translation from this, you know, from this to this, right? Like that's that's as close to an AI character consistency as you're going to get. So, I again just really like what they've been doing and giving them their flowers in that sense cuz we just wanted consistent characters for the longest time. That's it. Yeah. Now it's all Now it's all sort of I was just gonna poke in on this a little bit because
Play from 0:44:02 um pulling up Crystal Ball's, you know, uh X account here. So I just saw this one. Can Run runway fold laundry? Seems like it can. Wow. Okay. So So we put shirts, five shirts, pointed out the color and the item. Sort this and remove the basket from the shot. Um let me not get lost in my browser here. And That's a good one. Not bad.
0:44:26Scene sketching and layout prompting: mind blown
Play from 0:44:28 Five shirts, brown pants. Sick. Got it. Sick. Um, let me go back cuz there was a bunch of really good examples here. Here's sort of where we were talking about like the sketching, right? So, blocking scenes. Mhm. Boom. Boom. Show me the man from image uh two frame through the windshield of a car like the reference in one. You know, like this is this is kind of what we were, you know, alluding to, man. Maybe maybe like really toward the beginning of this podcast was like just the way that you prompt for these things. There are just so many ways to prompt, right? Like initially it was just this imagine you've got an empty box. Yeah. Now it's like, you know, you could do stuff like this, you know, and it's like you don't have to this this is almost like a different muscle that you're flexing, right? Entirely when when you're
Play from 0:45:14 generating this this kind of stuff. Um here's another interesting one, right? where it's like, okay, take your composition sketches, have the model turn them into uh real images and video, you know? So, you've got this, you've got this, then you get the guy, then you get this scene, right? I think the pioneer here was was cling with that. Once I saw that, I'm like, "Oh, this is going to this is how everyone's going to want to do it." It's like get this character, this environment, this object, like just mix it all together. I was going to say, was were they the were they the first ones that you saw that with? Was that them? I can remember where it's, you know, outside of Midjourney where with like, you know, style references that way, but yeah. Um, there's there's so many good ones in here. Um,
Play from 0:46:01 it's it's fairly it's fairly c I don't know if they integrated chatgbt into the back end of this. I was going to say that that's that's what I was going to say, right? Because like it felt like that was you're seeing a lot of this chat GBT type of functionality. Um, Our boy. Yeah, our boy. So, right. Yep. Here we go. Keep the car the same as image one. Make a drive in the location for image two. It's not a hard That's not a hard prompt to come up with. All you're doing is taking two things. You were blending them with control. Right. I thought, was it I think Midjourney was the first one to have a blend functionality. Yeah. In general, the idea of a blend. Now, this is like a I don't even know that I would call this a blend. Uh maybe like a smart precise blend, but like that's crazy. That's
Play from 0:46:46 crazy. Like I want this. This is what I've always wanted. Like layers and like ideas. This is this is the one where I was like this is genius. This is so easy. Right. So you're just doing this. Here we go. You've you're basically blocking the composition, giving it a name, giving it the reference of the guy, you know. Awesome. Like just really really fun. Sick. But this is it. I don't want to write paragraphs anymore.
Play from 0:47:15 No. Sorry. No, no. I mean, I'm done with it. That's what I'm saying, man. And oh, we didn't even talk about this part. Um, this tryon extension. Um Oh, yeah. Which good point, which I don't have it on my computer yet. So, let me see if I can find Do you have it on yours? If not, I'll just kind of pull it up here. But I might have to scroll back a little bit on this because there's an actual I think there's an actual video that was pretty good. I'll look for it for a second here. If I can't find, we'll just look at a static one. But um when that first dropped, can you wait?
0:47:25Interior design from mood board to layout to render
Play from 0:47:51 Scroll back up for a second just real quick. And not to go off. Keep going. Keep going. There was one with the green couch and a bunch of stuff. That one. This one. I want to see how that works. Okay. So here, actually, let me back out one second. So using an interior mood board and a flat sketch. Yeah. Okay. So, please generate an interior image using the furniture style, styles, colors, materials shown in one. So, this this may be a little bit harder to see for those of you watching. So, I'll just kind like there are elements of this. Let's just call it a living room. Uh furniture, objects, decor. Then the second image is the layout using shapes in that wire. What are we calling this?
Play from 0:48:33 Structure. It's a wireframe composition. Yeah. I don't know what you want to call that, but whatever. Overhead floor plan. than this one it's right it's just like basic blocks um and and identifying those objects okay use these uh furniture positions shapes and text labels from the floor plan in two and then there's some additional uh instruction around the wall paint color and the flooring wood flooring is the base interior materials camera perspective must must match the arrow direction indicated in two oh so they even threw let me see if there's an actual photo of that here we Oh, okay.
Play from 0:49:09 Boom. So, he even Yeah. Smart 3D. Yeah. You know, cuz you you'll see this also. Um you used to see this a lot with like lighting setups in photography and stuff where they would just, you know, if someone's trying to give you like a tutorial on it or show showcase. It's like light's coming from this side or this side. This reminds me of that. Smart. I like that though for lighting. That might be a cool thing to play with. Lighting too. There you go. Yeah. Yeah. That's that might be real smart cuz sometimes there's there's trouble with I've always had trouble with getting under light like light coming from underneath some in these image generators. Maybe I'm going to try that.
Play from 0:49:42 I don't know if you've I have I've always had issues with that would solve it. I feel pretty confidently not even have tried that. I feel like that would work pretty well given given if you've got this capability that should work. That should work. Yeah. Let's see. Did it come let's see that final picture though. Did it come out final picture? Yeah. Yeah. I mean, yeah, it placed everything which is nuts. The window is in a different So, you've got like window here. It's not perfect, but if you've got some close enough if you've got some you got some liberty. Yeah. I mean, now we're nitpicking for sure. For sure. You know, now we're nitpicking for what impossible. 100%. And it's also like worth reminding people of there's two
0:49:45Lighting direction via floorplans = next-gen hack
Play from 0:50:30 layers to this. There's the external ready for business era and there's the external like we're showcasing what's possible right now but all of this is in motion. Yeah. To then eventually be unlocked. Yeah. Right. So it's like we're also saying hey look this may not be a perfect today but it's going to be soon. Right. Which is which is also the exciting part. But also take this stuff like these are all concepts. you don't have to use them onetoone like they're all like I took that car example and was like all right let me try with a family portrait and I was like okay this works so it's like it doesn't have to be onetoone it only single use case using that like we just did with that light direction or the the you know camera direction like I thought of that immediately like let me try that with light right because that's like what else how else can we take that sort of structure that people are providing and
Play from 0:51:16 like adapt it to our needs so you just got to play around with some of this stuff and runway is so much 100% % you know and also that is a you know one other interesting thing about this this is a 2D thing right if you had this in 3D it would be much more accurate totally 100% right because this is basically saying okay window if you're in a this is a square room and the window is going to be flush straight ahead at you know 0 degrees or whatever it is it's like okay But um if you've got a a different dimensional model, I I would expect that to be much better even. No, but let me show you this real quick before just to just to touch on that because we were
Play from 0:52:02 that's how I sort of was thinking about it in terms of depth. Right. So like with this one, you can see sort of how it's layered and stacked like where I want the person behind the child and the woman behind the child, but also sort of like how the floor provides depth here. Like that's Well, even even the layering there is smart in the 2D because I I don't I don't know that I've seen a lot of layering yet. Well, that was my, you know, that was my intention. Just so that's good. No, that's good, man. I'm calling that out. I'm like, I haven't seen that. But that's why that's why it's like, oh, can it do that? That's always that's kind of always where it comes back to is can it? Yes. Cool. Now, let's push it further, right? Because you could do that room in a similar fashion. You could probably just draw like little circles and be like, "Put this here. Put this here." You just need
Play from 0:52:47 to get the dimension of the room right that you're trying to to start with. So the um the thing that we got lost in because we were nerding out on that was the tryon thing. Yeah. Um I I I can't find a video at the moment just on the fly, but um that's incredible. So it's the idea, right? Like you've got a picture of yourself. There's a Chrome extension. Again, I don't have on my computer yet, but essentially, you can take an item and virtually try it on, which I knew that technology existed because I had a friend that was kind of in that fashion space and told me about this maybe even like two years ago. He's like, "Dude, it's getting crazy out here." Um, so I'm surprised to see a runway, you know, kind of like get
0:52:53Try-on tech and Chrome extensions
Play from 0:53:33 into that. Uh, it's incredible. awesome lead gen tool, you know, like to utilize it that way. And then just people are like, "What's Runway?" I mean, dude, that adds a ton of value right there. You're you're going into a whole different space with something like that. And that was Cling. Like Cling did that a while back, but they didn't go the extension route. So sometimes there's a benefit to being second on some of this stuff. 100%. Different way. Yeah. I mean the the friction that takes place in trying to be the first to come up with the idea, execute it versus oh, you have a little bit of the playbook here.
Play from 0:54:08 Uh I mean this is just goes without saying, right? But just talking about the the references for consistent building. So you've got two things here, right? Um he says all these images were or what does he say? Um, all the images in the start are created just by combining the first two images and asking Gen 4 to create different variations of the subject in the location that we want. Mhm. Right. So, same girl, just different camera angles and different focuses. Didn't do bad on the feet there. Normally feet to me is worse than hands.
0:54:22Style consistency with JSON + ChatGPT
Play from 0:54:40 That's a You know what? That's a great point, man. I can't even tell you the last feet I've generated. Well, without shoes. Without shoes, right? Without shoes. Right. Feet typically to me have been so much more gnarly than hands. Yeah. Yeah. Love the color in this too. It's nailing that. I was waiting to see if that yellow would translate and uh it did across across a few different images. So there's again this runway stuff. It they I think people just get, you know, start to sleep on runway candidly. The quality is not on a V3 level or a clean 2.0 level. But sometimes you don't need that. You know, sometimes you just need something fun like this and Runway can do that. Um, I think Runway's great and same. I mean, yeah, I think you summed it up
Play from 0:55:26 perfectly. It's got it's got it's got faults. It's not perfect, but there I think now kind of going back to how we started this conversation with Runway. Now, it's not so cut and dry in terms of the capabilities. There's a lot more exploratory things and use cases that are now available. Um, and some of which, you know, don't require all the variables that maybe something like a V3 would, even though V3 does that stuff stuff really well. So, I, you know, like they may have, at least for the time being, both really good defined spaces within your workflow and your tools, right? Yeah, it's I don't I didn't see that if if you put that up there. Is that open available for API yet? I would love to sort of play with that a little bit more in certain spaces, new tools
Play from 0:56:13 coming up that you can bring this stuff into and use all of it. But there's I'm finding myself more intrigued by references than I am with let's say chat GBT at this point. Um chat GBT we think we have I think we've started to sort of unlock the faults. I I think that's one of the you know with the image generator now that within Sora and and in chat windows you seen those videos of the people that are like oh give me the same image like again and again and again.
Play from 0:56:41 Yeah I was telling you this remember the rock example. Yeah. Yeah. It's like that's true though. It does happen. It's true. It's true. It's um you know I tried something you know this too. Um, so I'm just calling that out for people listening, but you do get better consistency if you're turning it into JSON code and you're saying, "Hey, look, I want to I want to create more versions of this." And I mean this in the terms of like that could be that could be your subject, that could be the style of the character. Um, and whatever. And then really have it break down like I'll maybe maybe this is worth sharing. I don't want to go too too off with this, but here. Let's jump in
Play from 0:57:28 here and then let's go. Where is this now? I got to find this. That's going to be the hardest part. I think it's under tummy ache relief tips for my for my daughter. Um, okay. So, let me scroll. This was actually part of a longer thing, but there was like this really cool example and then I wanted to um I saw I saw this somewhere and I want to say who posted this uh No, might have been a Oh, I saw him do something similar. It was I think it was Freepic. Okay. I think it was Freepick's account, but like you know they they they these were animations. So this is just like I took a screenshot of this and I was like, "Oh, I like that like
Play from 0:58:13 just that character style, right?" So then it's like, "Okay, using the same I want to generate a new image referencing the attached photo." Okay, I started there. So I wanted to see how that worked. Then it was just like, okay, I want to, you know, add more and more detail, but I want to like create this same stylized look. So I had it create sort of like that JSON code template for me. That way I could just change these things out, right? And so it was like this stylized 3D, the the pose was a bust, right? cuz I'm not doing anything below the neck, you know? Um, but it breaks out all the variables that you can then change. Huh. Right. So then it was like, okay, cool. Like cuz all I had to do here is like change the background color. I wanted a different hairstyle. I wanted a, you know, I wanted the tattoo still. I wanted a female, you know, blah blah blah. But I made all those changes here. And then like kind of just kind of
0:58:23Mass-generating stylized icons and dogs with jobs
Play from 0:58:59 kept going down this path and getting into like animals, right? And then I'm like, okay, cool. got this. Then I want to come up with like fun. This is where like chat the chat GBT stuff is really interesting because it's like you can ideate right here too. And so then it was like okay like let me do let me think of like different animals you know uh showcasing some sort of like you know funny or clever kind of thing, right? And so I was like okay goat with like a crown you know blah blah blah. And then I got into this part where it was like dogs. I'm like, "Okay, you know, like dogs kind of have their own personalities by breed, right?"
Play from 0:59:35 Yeah. So, like, let me like break this down. And this is where I use chatbt to like really think about all the different dog types and maybe like what they would what their profession would be or what their personality would be in like a human, you know, level translation. So, like this this bulldog, you know, almost like looking like a little bit of like a a gangster or whatever. Um, so then I had it like really go into like Rottweiler Club security. um Chihuahua, LA hypebeast, poodle runway icon, you know, and so like it it started to give me some of those variables that then I could plug in, but then I could take these and say, okay, you know, using the JSON code, right, change these variables out. So then I had to uh generate these at scale for me, right? So I didn't have
Play from 1:00:20 to go in and manually change these every time. I could just copy and paste, copy paste the whole list of the the dogs that I wanted to run, right? And so, so then it started to get much much easier to do. So, I'll just kind of like scroll through these so you kind of see. Oh, see like cuz that that's awesome, right? Like these are cool and like these would be great to animate. It's just like I haven't haven't had time. But like the these were super cool and it's like okay the it's very consistent. It's tight branding. This is if you want to go a certain route with this. Yeah. Right.
Play from 1:00:50 And so like that don't sleep on that JSON component, you know, because that I think is the trouble trouble uh troubling part of like the chat is just like you're going to go nuts trying to keep everything the same without it without having some sort of like binary aspect via code. Uh and and that obviously seems like that's one of the the bigger unlocks with with chatbt specifically in this. But um also lowkey token token mining from that JSON code.
Play from 1:01:18 There's some good stuff in there. I saw there's like, you know, matte lux finish. Like there's some there's some interesting things. Yeah. In the Yeah. This like smooth plastic matte finish, ultra high resolution. There was some good stuff from some of the stuff above. I was like, "Huh, all right. This is, you know, you could have a whole like and these were oneshotted, right?" Like, so it's like if you really wanted to refine these or make like I I wasn't going too back and forth with like tweaks or anything like that. I I was more interested to see the success rate and the hit rate, you know, no gain.
Play from 1:01:53 You know, it's pretty dope. Yeah, I like the I like again the concept. I can see that and take it and translate it. Like if you're doing a branded icon set like to do those every once in a while, probably use the same exact thing, right? That's But this was, you know, that was fun, dude. I mean, and so I think that's that was sort of the interesting thing with Chad GBT. like the the JSON part of it I think is is worth exploring if you haven't already. Rory, you and I have talked about that a lot, but uh I think that's that's one of the bigger things that you can do. I don't know. Do you have any other thoughts on that before we get into Midjourney specific because I know we want to provide just like office hour updates and maybe provide an update on the video part of that. Let's just go right into the updates. I think, you know, Majour's been in somewhat of a
1:02:36Midjourney updates: V7.1, personalization, and video
Play from 1:02:40 I don't want to say a lull period since Omniruff, but it's been like two weeks since they charge like three weeks since they dropped it. So, you know, something's coming up soon, but I think it's good to go. Yeah, you're right. Yeah, it's been I think it has been two weeks. And so, uh I'll run through these really quickly again. Stop me where you want. Uh faster version of V7 planned to roll out next week. So, that'll be next week. Check. Cool. Uh cool. Love that. Potential minor update with the seven one, but they're saying that they're likely jumping directly to V8.
Play from 1:03:07 There's an emphasis on finalizing key personalization aesthetic settings. Um, new SREF planned. So, an improved version is nearing release with a significant style transfer enhancement expected. So, does that mean style weight like that means that or maybe that reskinning retexture? Even better, I don't know. Um, that would be a nice I mean like I guess again then we're we're saying we don't want more busyness and then we're also like we that be that'd be cool, right? I think that's what we're saying. I think that's what everybody's saying. Um yeah, right.
Play from 1:03:43 Initial launch won't include SRF random. Boohoo. Okay. Um haven't run one of those in a while. I have. I've I've been running them again. Yeah. Um get back into it. And there's some good ones with uh these. You could just tell they look a little bit more crisp, more enhanced quality. Yeah. Um, let's see what else. Okay, so here's here's what we're getting at, right? This this conversation, this episode's really been about video. No surprise then that video is like now seeming to come up more and more here with with Midjourney again.
Play from 1:04:14 Internal testing by guides and mods beginning end of month. So we're shooting this the 23rd. So there's a week left. I mean, so that's coming. You've got two video generation methods that they're talking about, text to video and image to video being optimized, multiple quality and pricing tiers planned. So, they're I think their goal here is to make sure that everybody's got accessibility to it. They don't want to price people out of the video component. Yeah. Um, video quality described as promising with distinctive aesthetic and excellent motion and a full launch launch is contingent on upcoming server infrastructure upgrades which are expected within roughly 45 days. So, you know, probably at least two months
Play from 1:05:00 away from like a full roll out, but I don't know if that infrastructure upgrade is a blocker to launch though. Yeah, we've I you know, again, I take this with a grain of salt. Like I wouldn't I'm not expecting to see this until later in the year. You know, they say two months, but I'm just guessing based on where everything's at, you know, they'll probably put resources back into and then how fast are you going to iterate on it once it's out? Because then then you're in this game, right?
1:05:01What Midjourney must get right with video
Play from 1:05:28 Then you're in the game of video, runway, VO cling, you know, you're going to have to you can't just sit around at that point. you're going to have to have a pretty good velocity at that point. I think they need a differentiator, right? They they absolutely need a differentiator 100%. And it's not not like the video generator's differentiator. I think it's being able to like utilize Midjourney like Midjourney for the video like being more creative with it. Not so like linear and sort of like rules-based like the camera moves this way and like you know you have to do it this way. like I want to use like click style reference and then like you know something like that or there you go Rory. Yeah, I like that.
Play from 1:06:09 pan left or pan, you know, like not I'm talking about like the the idea of it, not like the actual motion of pan left, like something like that where it's like it's always been they've always had a little bit of a different way of going about it, you know, stylize chaos. Um, you know, weird weird. Yeah. The way that everything sort of like they think and and dream things up like I would love for them to be a little bit different in that sense so it could be fun. Like I like the ingredients feature. Personalization that's still personalization like like that. I don't know how long that'll take to develop, but I'm looking for like one of those like X factor features from them.
Play from 1:06:49 It's all I think I think first first impressions here have to be great. They have to nail that first impression. So there needs to be really strategic play around when this comes out and it being right at the top at that very moment in time when they le they release it. So it has to be that I think yeah that that too. But like my thought there is if you turn people off right they're not going to come back and no they're not going to come back and do that. I think there are a lot of people though, right? There are a lot of people that are using Midjourney images as the static to then take into the motion with these other tools. So, you've got an
1:07:18The one-shot window to impress
Play from 1:07:34 opportunity here. I think they have to nail it with the capabilities and then right like that speed of iteration and continuous updates. And then lastly, to your point, it's the differentiator. And I thought the way that you described is perfect, right? just be just be who you guys have been. Like you're going to have something that no one else has. You're going to do it in some other way that doesn't exist across another platform. That's it.
Play from 1:08:00 Those three things I think. Yeah. But we're not asking for much. No, I mean it's just some it's just like the little extra spice that they've always brought to it. Not saying that needs to be in functionality, just like like style reference. Like we we didn't know we needed that until we started playing around with the tool. and then they gave it to us and it was like to me revolutionary cuz that was just so different than what every other model was able to provide and I'm looking forward I'm I know they will do something like that. I just want, you know, I think they should also take the sort of runway model and if they're going to do this like they've done with V7, which is just like roll text to video first, let everyone learn it, then go image to video, then go the extra features, key frames, then go whatever
Play from 1:08:46 else you want to do. Like I think that should be sort of like the way that they approach it instead of like here's all this crap learn. Yeah. Yeah. Like Sora, like here's all these things you don't know about. That's a really great point. I think uh there's going to have to be a real focus on that, too. Um more than they've had, right? Like they don't typically do really big big explainers.
Play from 1:09:13 They've they've done some little one or two minute clips on a new feature here and there, but I think you're right. There's there's got to be there's got to be that it should be make it easier for everyone. Like again, you said you have you have one shot at this really with people. They're not going to like just with the other tools that are available and like how complex they are now at this point. You know, there's got to be some level of simplicity or magic, right? Like the magic I think I I think that magic really is the differentiator.
1:09:23Bring back the Midjourney magic
Play from 1:09:42 Yeah, really is. 100%. And still can't get anything that's, you know, Midjourney imagination quality on any other tool. So, very true. Until later time, I will not be uh I will not be abandoning them because Yeah, man. just there there is no there is no other tool that has that magic and I even with these more super coherent models I said this like 70 times but I'm just not I just don't want to control everything anymore like I don't I don't want to prompt that much it's too it's too much thought and too much frustration like I just want to be able to to go in there speed matters speed matters quality matters yeah taste matters you know and sometimes I want new ideas like I want Midjourney to spit out 30% of what I
Play from 1:10:28 didn't prom of. Of of course, dude. I think that's that that to me is just a very like I'm I'm surprised more people haven't uncovered that. Yeah. Yeah. Because we don't always like you we don't always know what we want. You could like default same if you have like customers and you're building products like you have a pretty good idea as to some of the things they need but sometimes they don't know what they don't need or what they need. Um, so you know, there's that level of like, oh yeah, right. Like this does make sense.
Play from 1:10:59 I wouldn't have thought of that. Um, like text, like text and images. We didn't know we really wanted it and then Chad GBT comes out with it. We're like, okay, maybe we've wanted it. Yeah, maybe we wanted that. I don't know. But I think it's just Exactly. I'm not going to go too much deeper. So, I mean, well, I mean, we'll see what happens, right? We still got plenty of time to talk about it. We'll see. There um there a couple other notes here we'll just finish up with. They're planning some UI uh UI improvements for pro users, particularly targeting higher resolution monitors, 4K monitors.
1:11:14Wrap-up: chaotic times, coherent thoughts, caffeinated takes
Play from 1:11:29 Evaluating batch processing features for premium tiers. Um yeah, we haven't seen that batch 8 yet. Remember that was talked about. Um that got de deprioritized. Enterprise and API considerations. I've seen that come up a lot lately. So, it looks like they're finally testing that and and I think they're going to do it sounds like they're going to do a survey, too, uh to maybe gauge that interest, but got to got to do it. God damn it. So, just do it. Yeah. I So, near-term priority. Okay, here are the priorities.
Play from 1:12:00 We'll just kind of recap this and then we'll we'll wrap this baby. So, immediate focus releasing faster v7 approved SRF multi-ubject referencing features. Yeah, don't forget about that, right? Multi multi-ubject referencing features. So multi-omni ref right like yeah that's coming um multiple objects. So then you got subsequent pri priorities would be launching video model server upgrade refining user personalization tools long-term dev exploring vo voice modes advanced UI features and broader creative tools which would be your software hardware things secret projects that we don't know about.
Play from 1:12:37 I don't want to hear about a secret project anymore until until I see it. Same same. It's like it's like like your friend keeps coming over and it's like I have a secret to tell you and you're like what is it? They're like I'll tell you later. I'll tell you later for like two years. Yeah. But whatever. I know this secret. Just Yeah. Just don't even tell me. Just let it be a secret. But did it still like really amped about what's coming because I feel good about this dev schedule that it is feels like there's something coming at all times and not just like that v that V6 lull where it's like when are we getting V7, you know? kind of felt like the back the back half of last year just kind of felt like it dragged on a little bit. Dragged now. It feels like they're they're on a little bit of a tighter ship and I'm I'm appreciative of it. Yeah, it's it's good for us. Like
Play from 1:13:22 we we need to be Go Midjourney Go. We we do we do love you guys. I mean the the product's awesome. Obviously we do this podcast primarily based on Midjourney, so we we do feel very strongly about it, but um yeah, it's exciting to see all this stuff kind of to happen and continue to compound. Man, I mean, this is we're living in wild times here. Who knows what's next? I don't know. We'll see. It's all It's all chaos. People, I get that question so much like, "Where do you see this in six months?" I'm like, "I don't know. I just don't know. I don't have the answer. I can't predict it because it's too crazy." And then companies are really smart, man. They're just getting better.
Play from 1:13:59 They're getting better at understanding what to do, how to launch things, what people want. Like, there's more mass opinion out there. there's more data, you know, data to pull from. So, we're moving, baby, my man. Dude, it was good to catch up. I mean, God, felt long overdue. We've been uh we had to take a couple like two week breaks here uh cuz we had a bunch of stuff going on, but um man, this is a good one. It's a good one, guys. So, I mean, like we covered we covered a lot. Um thanks for joining us. We'll have something fresh for you guys uh on the next one. And um Rory, you got anything to say? Wrap up? Uh like and sub or comment. You guys are awesome. Last video, too. Commenting, dropping a bunch of stuff in there. Uh you know, again, tell your tell your
Play from 1:14:44 wife, tell your mom, tell your friends, come hang out. It's a good place for us all to have a therapy session. So, global community. Yes, sir. And thank you. Thank you very much with that. All right, guys. We'll see you in the next one. See you.
