Loading summary
Transcript78 lines
- [00:02]
A
Welcome to the Practical AI Podcast where we break down the real world applications of artificial intelligence and how it's shaping the way we live, work and create. Our goal is to help make AI technology practical, productive and accessible to everyone. Whether you're a developer, business leader, or just curious about the tech behind the buzz, you're in the right place. Be sure to connect with us on LinkedIn X or Bluesky to stay up to date with episode drops, behind the scenes content and a insights. You can learn more at PracticalAI FM. Now onto the show.
- [00:42]
B
Welcome to another episode of the Practical AI Podcast. This is Daniel Whitenack. I am the CEO at Prediction Guard and I'm joined as always by my co host Chris Benson, who is a principal AI and autonomy research engineer. How are you doing, Chris?
- [00:56]
C
Hey, I'm doing great. Can't wait to get into today's conversation. It's going to be fun.
- [01:02]
B
Yes, yes, an audio podcast. We're going to talk about a lot of interesting visual things. Maybe before we get started, just a Little teaser. Practical AI is posting some videos on YouTube now, so if you do consume podcasts that way, you might go check us out on our YouTube page. But speaking of images, videos and more specifically, image generation, really excited to have with us today Dustin Podell. Daniel, who is co founder and researcher at Black Forest Labs. Welcome, Dustin.
- [01:38]
D
Yeah, yeah, thanks for having me, guys. It's really great to be here.
- [01:42]
B
Yeah. And I know that Black Forest Labs does more than just kind of raw image generation. There's a lot of workflow related things, hardware optimization, all sorts of cool stuff you're involved with. But as we get into some of that, I'm wondering if you can just help our audience with a bit of a state of image generation methods and workflows for the industry. We've talked on the show before about diffusion models and we'll link some of those episodes in the show notes.
- [02:13]
D
Maybe.
- [02:14]
B
But a lot has happened, right? There's a lot of people working on a lot of interesting things and I'd love to kind of understand like over the last year, what are some of those main points that might be good for people to orient themselves to where things are at now?
- [02:29]
D
Yeah, yeah, no, it's a good question. I mean, if I'm allowed to take it even a little bit further back, I mean, the state of. Absolutely, yeah. The state of image gen, video gen, generative models as a whole has kind of gone crazy, so to speak, in the last like three or four years. So where are we now? Where we came from about four years ago, where I first kind of entered into the scene, so to speak, is we were at models that were essentially just doing little blobs of color that were kind of related a little bit to where you were with the prompt, you know, okay, oh, a lighthouse on the beach or this and that. And you would get something that, okay, that vaguely looks like it. And you would chose someone and it'd be, yeah, I could see it, I guess, and maybe it would interest a few nerdy people, but. And then now today we're at the point where, you know, if you've probably been seeing plenty of this stuff online, where seeing whole short films made entirely with AI generation, where certain scenes are almost entirely indistinguishable from, from reality, so to speak. So I would say we, we've, we've come quite far. But I will also say the core of the technology hasn't actually changed that much. It's been a, a pretty nice like forward progress. I don't want to like dive immediately into anything technical here, but, but it's, it's certainly, I mean, for anyone who's been paying attention, I'm sure, or anyone who really hasn't been paying attention, this probably came a bit out of, out of nowhere. So. So yeah, yeah, I was gonna say,
- [04:01]
C
I guess, you know, with the broadly attention by the general public is so much on kind of more of the LLM generative world in terms of, you know, that, that stuff. And everyone's finally on apps, regardless of whether they're technical or not. And so I think a lot of people kind of miss the tremendous advancements you guys are making on that side. And so like, you know, could, could you, could you take a quick moment and now that you've kind of done level, maybe step through some of the things and people may remember, like we talked about stable diffusion and things, a couple of those things that have kind of led up to what we're going to dive into today with some more specifics. And as Daniel mentioned, we can offer some links for past conversations if people want to dive into those specifically. But that would really be interesting to kind of hear distinct points on that timeline as we hit to this point.
- [04:55]
D
Sure, sure. First, before I do that, I guess I'll ask Chris, how technical am I allowed to get here?
- [05:00]
B
You're allowed.
- [05:00]
C
We're going to. So we' like, if you dive into something, we may stop you and say that is, what is that acronym? But other than that, you can go as technical as you want and we'll make sure. That everybody can stick with us.
- [05:14]
D
Okay.
- [05:14]
C
Including me.
- [05:15]
D
Yeah, Yeah. I guess if I'm allowed, I'll do kind of at least the best of my knowledge, overview of the last bit of time since. So obviously the field of AI and ML has been around for a while, trying to structure and learn distributions of different data. Mostly it was used for tabular prediction, all sorts of other things for a while. And then in 2017, some researchers at Google figured out this very cool thing called the transformer, which I'm sure people have heard many, many times about right now.
- [05:49]
B
Which
- [05:51]
D
I guess the best way to go down this path is to say we found a way to train much more general models, like learn from much more general data. Trying to think how to, how to essentially take this story.
- [06:03]
C
It's okay, it's a dramatic pause. It works, man, that's good.
- [06:06]
D
Yeah. Essentially the story I want to tell is the story between the kind of, I don't want to call it the battle that's been going on the last few years, but essentially the story between diffusion and autoregression that's been going on the last few years. So let me start by trying to define a little bit what diffusion autoregression is. Autoregression is anyone who's been playing with language models. Oh, you know, all these, you know, GPT, Claude, all the new stuff, everything that's been going on in the last few years, these are
- [06:38]
B
what you would
- [06:38]
D
call an autoregressive language model. And so what this is, is it's predicting data essentially one piece at a time. So it predicts one piece of data and then it looks, okay, what did it make? Then it predicts the next piece of data. It looks at what it make, then it predicts the next piece, and it slowly, slowly, slowly, slowly builds out. You know, at first it was just very basic language, and then it became, you know, now, I mean, we're having agents and a lot of very, very fancy things with them. On the other side of the coin has been something called diffusion modeling. And I mean, I don't want to get into it because now, now we're doing more like flow matching. But let's just go with diffusion as the, as the base term here. And with diffusion, what you're doing is you're not trying to Model 1, say like in language model, it's, you know, you call it a token, but you assume like, okay, one word at a time. With diffusion, what you're trying to do is essentially trying to think how to tell how to, how to how to clean the story here.
- [07:33]
B
Yeah, yeah.
- [07:33]
C
No, it's all good. I love. Yeah. I feel like I'm watching like a thriller where you're about to, like, reveal the next thing. It's good.
- [07:41]
D
Yeah. So what we do with diffusion is. I'm just going to tell practically what we do and then we can kind of break down. You can ask me questions.
- [07:47]
C
That's fine.
- [07:49]
D
So what we do with the fusion is we take a whole continuous medium. And what I mean by continuous medium is the most famous one is images. The world is continuous. There's not, you know, it's not that. There's, you know, one, you know, like, like a letter is discrete. There's a T, there's an A, B, a C, a D and E and an F. A color shape. The structure of the world. The color, you know, the, the light, the. All of this. This is. This. These are continuous, natural, median, continuous. And so what we do is we take one of these continuous natural mediums that we've now encoded into digital space. So we use RGB, so, you know, okay, 256 colors we typically use for red, green, and blue. And then we encode this into an image. And then what we do is instead of trying to predict. So an autoregressive, you might try to predict, like, pixel by pixel by pixel. Instead, what we do is we essentially try to remove information. And the way we remove that information is by adding noise. So you imagine you, like, add a little bit of grain on top of an image. You can still kind of see the image. Maybe it's a little grainy, maybe it's a little blurry. Oh, you can't quite see it. And you add a little bit more. Maybe you see some of the shapes. Maybe. Okay, there's like. Is that a dog? Yeah, you kind of see it. The color starts to fade, and then you keep adding and eventually you just get, okay, this is just noise. I don't see anything in this image at all. And with the diffusion model, what we're trying to do is we're essentially adding a bit of noise. And then we're saying, now try to predict what the clean image is essentially for this. So, you know, okay, so we add a little bit of noise, predict the clean. It just has to remove a little bit of noise. You add a lot of noise. And now what it has to do is there's so little information. It has to actually, like, come up with essentially, like how to infill this properly. And then what you do is then in. In what we call inference, which is when you actually run the model, you ask for, for an image. What you would do is then you start at a fully noisy image and you say, give me a dog, you know, wearing a top hat on the beach. And then essentially what it does is it tries to remove a little bit of that noise to get closer to this idea of a dog on a top hat at a beach. And you might get some structure. It tries to figure it out. It's very coarse, it's very. Just trying to figure out the shape. And then it, it has a little bit of shape to latch onto and then it does the next step where it removes a little bit more of that. Oh, and it starts to come into view. It starts to, you start to get a little bit more in focus and you do this do, do, do, do, do, do do. And what you see is essentially you completely remove the noise of this image and oh my goodness, now you have this, this completely generated image of a dog on the beach with a top hat. For a long time this was, I don't want to say not good. I mean all, all the, all the models have gotten a lot better. But when it first started out, these looked like blobs of color. It was like these little, like you could barely see it. And, but fundamentally what I just described is still the process that's been happening over the last four years from image generator models for anyone that maybe tried like Dall E mini, I mean the old stable diffusion models that we worked on. You know, I don't know, many, many of the different models are out there now to, to, to these modern video models that are producing whole 10 second sequences. Fundamentally it's doing the exact same process of removing this information slowly with this noise and then just slowly going the other direction, essentially creating info to infill this. If this makes sense. I'll, I'll pause there for a second just to kind of like.
- [11:00]
C
Right, so let me tell you because as I'm listening, very, very good explanation, probably the best one that I've heard. So you are right on target if you, if you just carry on what you're doing. So I'm enjoying this.
- [11:14]
B
And, and you mentioned that when this started, right. You sort of ended with these blobs, et cetera. I think there's a lot of people that have seen very impressive things like the, like you were saying, like whole parts of maybe movies or commercials or something being generated in what, what is the current, I guess, state of the art in terms of whether it be your Your models or others. And I know there's more to talk through as we get in. Like there's more than just raw image generation. There's how this fits into workflows and, and doing certain tasks. But what is kind of the state of the art in terms of what we're able to generate? Where are the, where the bounds currently and in terms of quality and that sort of thing?
- [11:57]
D
Sure. So I'll tell you kind of like in the last year I'll split it up into, let's call it three different categories, but they kind of blend a little bit, which is these continuous mediums I talked about. So I gave the example of a continuous medium with image. Video is obviously an extension of that, adding the time dimension. But then there's also audio as well, which now a lot of these video models generate audio. But then there's also things like, I don't know, maybe you've heard the like Suno or these music models. Some of these are also using this exact same technique of removing the information with noise, bringing it back in. It's just a different medium that they're essentially applying to. I'm happy to talk about our models all day. I love talking about our models. But I also want to be fair to just anyone listening and honest to kind of the state of the world on if you want to, you know, go out and try some, some, some nice things out there as well. I, I, I will say that before I kind of like dive a little bit into, into what I think is, you know, maybe the best one to recommend, I, I will make like one statement of, of that like, like best quote unquote. Here is a little bit hard to define sometimes. And this is something we're always like wrestling with is like what makes a best model for someone. Because you know, obviously you, when someone uses these models, they're typically prompting or putting in now you can reference images or videos or other sorts of things and kind of treat it a little bit like, more like a, like a creative companion. And then like what you expect out of this can be very different for different types of people. That said, I would say that if you want to see kind of the quote unquote state of the art right now for what I would say is like text to video, at the moment it would be Seed Dance. So Seed Dance has done some extremely impressive work with The Sea Dance 2 model over the, over the last year. So they're doing, I think it's now 4k generations up to 15 seconds, which is, I mean it's very nice quality. It's focused very much on like cinematics. I will throw a shout out to anyone that, you know, if anyone from the SORA team ever listens to this. I think Sora 2 is a quite a nice model. It didn't rank so high on some of these like leaderboards that we in the AI community there's a. I'm sure you guys talk about this plenty. Like leaderboards rankings where models place. I think in the LLM side of things, this is much better covered where there's very clear metrics of how well does it program, how well does it do math. On the more creative side of things with video models, image models, audio models, we typically end up falling down to this single preference type benchmark, which is just like a general preference. Everyone in the world goes and can can vote on. You know, on these leaderboards. There's a couple different nice companies and sites that do this, but I feel like this kind of removes some of the specificity that the granularity. That's the word I was looking for.
- [14:43]
B
Listen, I've been to incredible amount of AI events, many of which are good, but many of which are not practical. You know, I love practicality. We're on practical AI and some are just hype focused, some are sales focused. That's why I'm always eager to share about an event that I truly think is practical and useful for people. That's what I discovered last year at the Midwest AI Summit. And they're going to have another Middle Midwest AI Summit October 15th in Indianapolis this year. 2026. One of the reasons why I love this event was there was an actual AI engineering lounge where you could sit down and talk through your use cases with actual experienced AI engineers and practitioners to really brainstorm and come back from the event with actual solutions and practicality rather than just a bunch of content and slides. But there were also amazing keynote speakers, speakers from even that had been on the podcast before, like Rajeev Shah was at last year's event. I would really recommend that you go to midwestaisummit.com and for our listeners you can actually get 20% off with the code practicalai20. So go to midwest aisummit.com don't miss your chance to attend this event and get 20% off with the code practicalai20.
- [16:12]
C
I was going to ask you, are the models that you're describing, are they kind of the traditional diffusion models or. You had mentioned in passing a few minutes ago about flow matching. And so I was just in my Head. I'm trying to kind of categorize how do the ones that you're talking about right now, how do they fit in and like. And how. What is. To go back and pull that up, as you were talking about diffusion and you made the reference to flow matching, how do those. What is that? What is that transition? And how do those models that you're talking about now fit into that?
- [16:45]
D
I'll first keep it very simple and say all of these models are doing the same process of what I described earlier, of adding noise and training the model to then remove this noise. The difference kind of, this is where it gets a bit more technical in how we actually approach this. So we used to do more of this process called diffusion, which I don't honestly know if I want to get into how deep and technical this is. But essentially what we've done is we've cleaned this process up to what we call flow matching. And flow matching is essentially this very simple process of still doing the same thing, still training the model to remove this noise. But fundamentally, what it's learning under the surface is anyone out there who has ever seen, like, a velocity map or like a flow map, what this essentially looks like is if you can envision like a. I almost want to mark a whiteboard behind. You're going to have to.
- [17:42]
C
For the audio folks only. They're not going to see. It's probably about to turn to the whiteboard. Yeah, yeah, it was great. Dustin, if you're on video, you saw it. But Dustin was literally about to turn back to the whiteboard behind him, which I love. Like, if I was in the room, I'd be like, go, man, take me there. Unfortunately, a good bit of the audience won't be able to see it, so you're going to have to describe it.
- [18:02]
D
Yeah, yeah, no worries. Let's keep it purely in the audio space. So I'll draw a picture with my words as best as I can. I spent enough time prompting anyway, so hopefully I can do this.
- [18:13]
C
All good.
- [18:14]
D
But, yeah, essentially the way I like to think about it is if you imagine, like, a landscape, like, I don't know, you can imagine your town or the region you're in and imagine you're looking at it from, like, the sky. And you see, like, your house, you know, maybe right in the center of this map. And what you might want to do now. Now imagine, okay, there's wind going all over. All over the area. And the thing that you want to do is you want to be able to throw up like a Paper airplane from anywhere in this, in this landscape, your city. And you want the wind to carry it so it lands on your house. And functionally what we're trying to train here is essentially that in much, much grander hyper dimensional space, where instead of it being your house and instead of it being a wind in a paper airplane, what it is, is your house. In the scenario is what we would call the manifold of real images. It's essentially the place in. I'm trying not to use too many, stop me if I use too many words here. I want to say, no, no, I'll
- [19:17]
C
ask you, but you're doing fine. Keep going. We're final technical. We'll just get explained it as we go.
- [19:22]
D
Yeah, it's the place in latent space that would be where the actual real images are. So another way I'll try to describe in a simple terms. Okay. With your town here, you're in 2D space. You're where you are above an XY and the paper airplane has to fly in this XY coordinate. So you land and then you have the XY in this space. Instead of it just being two coordinates, it's enough to have all of the colors of the entire image. So that's why I say it's hyperdimensional. It's, it's the center is where literally real images are. And so what we're doing is instead of it being wind and the rest of your town, imagine the rest of your town here is every image that isn't real. And what that means is noise. If you think about noise as a real image, you can go generate a bunch of noise and save it to your computer. But that noise would exist somewhere in this space of all possible images. They're not real images, they're, they're noise. But they, you can, you can save them. You know, they're, they're, they're RGB values. You have them. And so what we're doing with this, this essentially this flow matching is we're training the model. Like when we say we're training it to remove noise, what's really happening under the surface is we're training this flow map so that we can land anywhere in this field of noise. And then these flows, these winds in our scenario, when you take a step, will take you closer to your house or to the manifold of real images. And that's what like as you get closer, it starts to look more and more like a real image. It's blurry, it has this. And then when you actually finally land on this, this manifold, boom, you have hopefully a real image, but you'll have some image, you know, and then the better the model is trained, the better you have mapped this flow to actually take you to where you want to end up.
- [21:04]
B
And I'm assuming. Yeah, no, and I'm assuming because you are kind of mapping this from. From where you start to, to where you end up with this real image, maybe in a way that's just to be crude, I guess, less, Less random. I mean, at, at the end of the day, all AI and machine learning, like, it's sort of, that training process is very much trial and error, but we have a lot of optimizations around it. Right. So I'm, I'm assuming that this then allows you maybe to, I guess, advantage wise, shorten the, the training or make that more efficient. Or am I, am I misconstruing that in some way?
- [21:45]
D
Yeah, yeah, I mean, definitely. Like, one of the things that's taken us a lot further over the last three to four years is just figuring out a lot of optimizations, both in the actual training itself and then also in just, you know, architectures of the actual models have improved. But I will say the, like, fundamental underlying process of just removing this information and finding a path back to it is still the same process. It's. It's just like the car. You know, you've been having cars go down the road since the 1950s, and we've made better engines and better safety and all this, but you're still driving down the road like that. That's.
- [22:20]
B
Yeah, yeah, that, that makes sense. And now, definitely, I, I think that's a good setting in the sense that we've got to a point where, you know, your, the, the models coming out of Black Forest Labs, the models coming out from other places, the models that are even like now in my text messaging, right, I get an image, I can immediately remix it with an image model. Right. In some way. So these things are becoming more embedded in our lives. But I don't know if everyone in the audience, some, some might have been along for this ride where I was like, oh, cool, I can generate an image of a astronaut riding a horse on, you know, you know, wherever. And that doesn't seem that practical to folks. So I'm wondering if you could now kind of given that, that foundation that we have and we know sort of where we're oriented in the state of. I love how you put it on your website, Visual intelligence, which I think is. I love that statement because it gets to more like Language, although it, it's being used a lot in terms of agents and intelligence, like you mentioned, it is only. It is very much a subset of the information that we process as, as humans. Right. There's this visual element, there's the audio, et cetera. So now that we have that foundation, could you help the audience understand some of, I guess, the practicalities and the outworkings of the like? Okay, we can do this now. So what, so how does that, how does that help people in the real world in ways other than maybe just pure creativity? They're certainly like the cinema, like you were saying, that side of things. Not everyone's going to be generating movies, maybe, or maybe more people will, I guess. But yeah, I think you're understanding what I'm saying. Like, where is this going to impact? Where is it impacting me now? Where is it practically going in terms of the application?
- [24:15]
D
Yeah, yeah, absolutely. I'm very happy to talk about this. I think we're going through like a very nice transition period right now where we finally get to leverage these for some very useful things. And if I'm allowed to take a little bit of a tangent and kind of please lead up to it again,
- [24:29]
C
wherever you want to go is good. We're all good.
- [24:30]
D
Yeah, yeah. So, I mean, I guess I'll take it back to like a little bit of a timeline of things. So. Okay, so so early on we had these nice models where most people recognize them for. You put in a prompt, you get an image out. You put in a prompt, you get a video, maybe you get a song. It's just this one way, okay? You make something. This appeals to maybe creators or ad agencies or movie, you know, now cinematographers and I mean, we love this stuff. We love the creative side of this and what it allows because it fundamentally to me this is like a nice potential communication tool. But then I feel like something changed. I don't want to say changed, but there was definitely a timeline moment when we started to move into editing. So was it two years ago now? A year ago? I don't know, around a year. And give or take, we released our first in context editing model called Flux Context. And on the surface you would look at this and go, okay, well this is, this is an image editing model. This is something. You could take a photo, you can clean it up, you can add a hat, you can do silly things, you know, you can whatever you want to do with it. It's a general editing model that's supposed to, supposed to do all these interrelational things. And on the surface, that's very cool. It seems like another creative thing. But if you think about, like, what's actually going on under the surface, for the model to be able to do this, it has to understand a significant amount of relationships in the world and what it means for these, like. Like how these relationships actually interact together. So if I say take a picture of a. Like a. I don't know, we have, like, a water glass here on the table, and I take a picture of this water glass, and I say to the model, knock the water glass over and shows, you know, show me what happens. The model has to understand some part of the actual world. Like, it has to actually model the world in some way to know, okay, it spills over, maybe something gets wet, maybe X, Y, Z happens. And essentially what we were trying to do is train a model that could do all of these types of relationships. So the model is learning not just like, this one thing, but just how the world works fundamentally. So you can do this editing now. We can take this a step further, and we can look at all the video models that are coming out right now, and you can do a very similar thing. You can take, you know, in a knit image and say, you know, okay, this person now grabs a fire extinguisher and puts out a fire. And it has to actually understand these relationships to do that. So. Well, maybe this wasn't our original goal. Way back in the day, you know, we were trying to make very cool models and figure out how to. I mean, I think in some sense we wanted to model the world, but maybe weren't thinking this far ahead. Inherently, through this whole process, we've learned to build these models that are developing this. And I'm trying, really trying to avoid the use of the term world model here, because I'm going to go there
- [27:15]
C
if you don't, I'm just telling you.
- [27:17]
D
Yeah, you might notice me trying to, like, skirt this, because I think it's a little overused these days, but we can definitely talk about it. But fundamentally, too many people, which is
- [27:27]
C
where I was going.
- [27:28]
D
So. Yeah, yeah, Fundamentally, this is what these models are doing. When you train them at scale and you really train them to be general and robust, is they need to be able to simulate parts of the world to get an output. And on one end you can take that and go, okay, well, this is nice for creativity because you can make a film scene and it looks nice, but on the other side, this is why we're starting to move in this. In this area. You know, this area where we're calling visual intelligence. And now we're starting to see how we can actually leverage what we're calling, not just us calling it, this is, this is a general field term, but the representation inside of this model of the world to actually go do practical things. Now, this isn't to say we're going to drop the creativity side of things. We're still definitely pushing on the side. This is, you know, something that we're all still very fond of, but looking forward, like, if our models are understanding these relationships and the physics of the world like this, well, this is a great place to put it in, say, robotics. And actually, okay, now, now a robot has this, this understanding, this model of the world inside of it to go act and, and take actions with confidence in the world. So I'll leave it there as kind of a, kind of trying to build up to it.
- [28:34]
C
But let me, let me ask. I'm just going to go there because that's kind of. You're in the area that I spend all my time, which is, you know, embodied intelligence, RO, you know, UXVs and stuff like that. That's my world. And so even within our space, the notion of world models, there's a lot of interpretation. If you get into a meeting with 20 people, there's 20 different definitions. And we have to start off by sorting all that out in terms of how we're communicating. As you add in this notion there of this kind of contextual understanding that it has a representation. I am curious, before we move on, do you. How do you. Do you. And you've mentioned now robotics, do you see it as the same thing or is it kind of yet another variation of a world model, you know, using the word and stuff? Like, how closely would you, would you believe those two to be related? You know, as you're talking about that, just because both are big topics, you know.
- [29:33]
D
Sorry, sorry. Can I, can I just ask for clarification? You asked them, like, how close do I believe, like, our models are, like a world model or a.
- [29:39]
C
Well, like, when you say world model, I'm just kind of trying to clarify the same thing I did when there's 20 people in the room and you're asking what they mean by it. And as you mentioned, robotics and stuff, we all need this representation of the world out there so that we can do better about, about acknowledging the context of what we're working on, whether it's robotics or in visual intelligence, presumably, are they. I'm just curious, in your mind, do you think that they are very closely related or are they kind of distinct ideas of what a world model is? What is your take on that?
- [30:10]
D
I would say, I would say that they're pretty related. I mean, I, I, to, to me it's like fundamentally under the surface of what we're doing is we're, we're trying to teach the model as much as we can about the world that we exist in and then asking for it to utilize that in some way. And up to this point it's basically been mostly through creative mediums. But like that representation that we're talking about, that, you know. And again, I'm, I'm trying to avoid this term because we, I don't know, it's a little overused the term word model. But it is, I mean this is, this is what it is. It's a, it is modeling, it's creating a representation of the world we live in. And then it is using essentially that intelligence. It has to be able to act in it. So the idea is that if it understands enough of these relationships and the actual like physical properties of it, it's a really good foundation to you know, build and train robotics on top of.
- [30:58]
B
Gotcha.
- [30:59]
D
I hope that, hope that answers it.
- [31:01]
C
That did. No, that was good. Thank you for putting up with that. I was just curious.
- [31:06]
D
Not at all.
- [31:07]
C
It's not unusual to, to navigate that. So go ahead. Daniel. Sorry about that.
- [31:11]
B
Yeah. From the non robotics person in the room. That was really interesting. I appreciate you going into that. And I'm wondering there's that kind of outworking of some of this where you are now kind of understanding the context that's in this model, maybe how it's representing the world. There's also things that I've seen just crossing my paths, whether it be kind of an E commerce or like I say, my, my phone, my text messaging, so on, on the web, where these models are being more and more integrated into workflows. Whether that be kind of like try on these clothes or glasses or whatever, or it's creative tools in like visual editing platforms. Are you seeing that with, with your all's models in terms of kind of what's the state of. And I know I want to get into your model families here in a second. And they do different things. Right. And many of them, I know there's a big some that are, that are open weight. So you might not know all the ways that they're being used. Right. But from at least those partners that you're working with in terms of today, what Are some of those creative and maybe most practical uses of these models that you see out there beyond just kind of the fun image generation kind of side of things?
- [32:37]
D
Yeah, yeah, no, absolutely. I would say again, I'll come back to the. The moment we started to get context into the model that wasn't just text made a huge. Like, it was. It was a fundamental change in what these models could do and how people actually worked with them. Now, our first model, Flex Context, only could take one image reference. It was mostly used as an editing model, but you could reference a, you know, a picture of a product and then tell it to generate like a, you know, a nice product photography set. And this was very nice. Then moving on, we introduced like the Flux 2 family and then, and then the Klein speedier smaller variant that now could take many references. And then now we're looking at, I mean, people and this kind of comebacks comes back again to like, what I was just saying of having this like, representation of how things relate and that the better we actually build this like, representation, the more interesting ways you can essentially like tie in all of these different things. So I mean, you know, throw it out. You know, one of the most commonly used, or I don't want to say commonly used, but like kind of obvious cases is okay, like clothing trying like, this is a very, you know, here's a picture of me, here's a picture of some clothes. Could you please, you know, put, you know, show me what this looks like. I've seen people like, I, I actually did some home decoration earlier this year, just trying to see like, what different couches and furniture look like. In my place, one of the most interesting uses, I'll say that kind of stood out to me that I saw at a hackathon was. And I, I don't know how much this could be used for like real planning purposes, but I just thought it was interesting is someone was taking pictures of fire exits in a building and then generating what it would look like if a crowd was trying to leave through this fire exit in an emergency so that they could actually, like.
- [34:21]
C
Brilliant. Yeah.
- [34:22]
D
So they could actually gauge like, what would this look under, like emergency? Like, where is the crowd? Like, I don't. And obviously there, there's, you know, there's a generated component. So you have to take it with a little bit of grain of salt. But you still could get a general idea like, of this is what this would look like under this scenario.
- [34:36]
C
And I thought, yeah, without derailing us, you just sparked a whole bunch of ideas on that in my head. So like I'm just like, oh, this is great stuff. Keep going. Sorry.
- [34:47]
B
If you've been listening to the show over the past few months, you realize just how transformative agentic AI is, whether that's Claude code or Hermes agent or custom built software that you, you're deploying for operational efficiencies or as new products to your customers, regardless of your maturity. Now this is the world that we're headed towards, this agentic AI world and there's a lot of security and governance teams that aren't letting these agents go into production because of risks related to agency and autonomy. And how do you take care of things like prompt injections or insecure tool usage? There's a lot to take care of and that's why I'm personally spending my time outside of the show working with an amazing team of AI engineers to build Prediction Guard. Prediction Guard is an AI control plane that you run in your own infrastructure behind your firewall. Developers can build on top of this control plane using everything that they want to use. OpenAI and anthropic compatible APIs, MCP servers, frameworks like LangChain. But all of this is plugged into a built in governance harness that enforces your organization's AI policies and all of that telemetry goes back to your monitoring and alerting systems. I'd encourage you to check out what we're doing@prictionsguard.com practicalai you can schedule a demo with me and the team and, and I'd love to get your feedback on what we're doing. So Visit us@prictionsguard.com PracticalAI that's predictionguard.com PracticalAI these are all. So you mentioned the Flux family of models. So that's Black Forest family of models or at least some of the models that you've worked on. Could you just kind of give us a concrete, you mentioned a couple by name but like give us a tour of the family of models and then maybe where I know that there's someone hugging face, people can find them, people can, can look at them, but maybe just give us a little bit of a tour of the model family and then anything that, that you want to share about some of the, some of the distinctions between them, different ones of them.
- [37:04]
D
Yeah, yeah. I mean we're, I would say fundamentally like as a core we're still like
- [37:09]
B
a,
- [37:12]
D
I'll say we're still like a research lab that wants to push the boundaries. So with each one of our model releases we want to try to like level up Some capability of the model, not just, you know make a general improvement but like really see, see some, some new versioning with it. So I mean I, I can take back through. There's not that many models so you take back through a little bit of a history. You know it was a little, what was around two years ago now we released the first, our first family models which was the Flux series. And this was the first set of models that we created after we formed the company after a very fun but intense, I think it was around four or five month sprint to really like build, build our chops here and get something out. And with that we came up with this at the time, this, this essentially distinction between the models where we released the Flux one. So we had the fluxone Pro model which was on our API. We had the fluxone Dev model which was this commercially licensable model but the weights were open for people out in the world to use. And then we had the fluxone Schnell model which was a high speed like step distilled model that was totally. I don't know if it was MIT or patchy, but it was open to use for whatever purposes you want. Then from there we worked on our Flux2Flux Tools series of models where we realized like we needed a lot more control with the models. And this is where alongside this work started the Context project to okay, these people want a lot more control with the models. These models are learning a lot of relationships. How can we leverage this? Which led up to the Flux Context release which I talked about. This was our big edit release. Everything I've mentioned up to this point was still kind of our like historical models. Now we're getting into our more up to date models although they're getting a little bit older now but we have a. I don't know something. I'm excited coming. I don't want to give dates soon. Hopefully later this summer that'll be very exciting to talk about when it comes out. But looking forward to it. Yeah, we'll have to get you back on definitely. Yeah, happy to, happy to come back. But then we got to our Flux 2 series of models and Flux 2 was a very big upgrade for us where we really pushed the capabilities not just on T2I but also on editing. Not just on single image but this is where we also introduce like the multi image, the kind of omni edit where you could put many people, many different items, have all sorts of relationships. This is where we started to see you know, not just these kind of more standard advertising Use cases which still were like the most common. And we really want to support this but like some more, you know, interesting how people kind of build you know, like things like this fire exit thing or other types of stuff. And then right after that we, we came to our Klein series of models which was a, essentially like a size distillation where we really wanted to pack as much performance as we could into a small model for this, you know, for both use on our API, but also just for open weight release. We know that there's many people out in the world who use our models locally and we really wanted to make sure they had something powerful they could use. So it was a text image and an editing model. It was a, it was a very fast model and it was, it was pretty small in comparison to, you know, it was even smaller than our Flux 1 series. And then we've done further, a further speed up on that with our Klein KV model which introduced I believe for the first time KV caching which is this. I'll just say it's an optimization technique that's very common with the language model world that we were able to bring into the editing world to get a very, very big speed up on local editing. So people who wanted to actually like use these models locally but also you know, we also serve this as well, you know. And now, now since then we've done a couple blog releases. One fairly fun research, you know, research blog on something called Self Flow which I'm repping partially because I am on there. I'm not not the lead author. Our lead author is incredible. Hila, she's amazing. But. And then now we're, we're kind of all pushing forward to our next big release which is hopefully going to be. I, I'm quite excited about it.
- [41:07]
B
That's awesome. And just a follow up on that. I know some people out there, our listeners are always trying things, you know, on their laptop or wherever they're, they're, they're pulling down things. I know for quite some time when I tried to access some of these models and run them myself, it was either very difficult to find with my limited resources the right kind of configuration to run this, but also it was sometimes incredibly slow. So to just give a sense of some of those more hardware optimized models that you mentioned. I think the, the client and other models. How maybe my question is can I reasonably run one of these models on my laptop now and create a great image like what's required here? Certainly I'm not going to serve my production Web app in an enterprise environment off of my laptop. But just maybe give us a little bit of a sense because that has changed on the LLM side, right where that has continually updated and obviously the smaller models are not up to the same output quality as the larger models. But you can run a small model, you know, even on a CPU now at least for some tasks that is pretty reasonable. So how has that progression happened on the visual intelligence or image generation side?
- [42:32]
D
Yeah, I would say I'll first state that the LLM world definitely has a lot more people working on it, so they definitely have a lot more of this up and down. That said, with our Klein series, I, I, I haven't run this personally, maybe I'm not the best person in the world to speak on this, but I, I'm 99% certain you could run this on say like a modern M series Mac, like a MacBook Pro. As for speed, I don't, I don't have any numbers I can quote because I haven't tested this them myself. But, but, but this is always like a big kind of trade off in this world of, of how, you know, even on the language side, you know, you see this scaling of a, you know, like a lot of the new big open releases that people are excited about are getting into the hundreds of billions of parameters. And this is why we, you know, this is why we did like the Klein series for example, because Flux 2 was a, it was a 32B model. It was, it was quite chunky. You could run it on like a, you know, higher end local computer, but it was slower. To run locally, you needed some more powerful, you know, computing. This is where like the Klein series popped in. But it's, it's always a trade off because we want to push the bound, you know, we want to push performance and, and make something that we're really proud of. And then we want to try to bring that again back into a smaller, faster scale. And this is something we've continually done, you know, with our Fluxone series we had Chanel. With Flux 2 we had Klein. I don't want to promise anything going forward, but we definitely want to like keep bringing this like bigger power that we generate into these, into these smaller models that people can hopefully run locally.
- [44:01]
C
Absolutely super cool. I'm excited to try some of those with like, as we are starting to wind up and you guys have done such some really, really cool work in this space. One of the things we always like to ask as we're closing out these episodes are kind of like, we want to get a glimpse into your head, into your thinking about what's to come and kind of to frame it a little bit as you're thinking about, like, you're out of the day's work and you're enjoying your evening and your mind wanders and you're kind of thinking about, like, down the road, what would you know, what you want, where your passion is taking you and what you'd like to see. Can you share. Can you share a little bit about what the future that you would like to create with visual intelligence in terms of what's next that you guys haven't addressed? I'm not asking for a product release so much because I know those need to stay in, but kind of just like, what is your aspiration? What's your dream in terms of where these kind of capabilities might ultimately lead and love to get your kind of your dream context, if you will, as we finish up.
- [45:16]
D
Sure. If I'm allowed to give two split answers.
- [45:19]
C
Absolutely. Totally fine.
- [45:22]
D
There's kind of two areas that constantly sit on the front of my mind. At some point they'll hopefully become one, but I imagine the next. And this is no glimpse into saying what we're doing, but this is stuff I'm personally excited about.
- [45:33]
C
Fair enough.
- [45:36]
D
I would say the two areas that I'm the most excited about is one is long context. Truly multimodal models. So now, nowadays, obviously, we're seeing lots of people work with agents constantly. And we're, you know, and on. On our front where we're having these, these more like continuous generative models, we're starting to introduce more and more, you know, contextual usage with these, like, references. And, um. And I'm looking for. I'm looking forward to when this kind of bridges a little bit more. And we have models that not just can, like, okay, you have, say, the agent calls the generative model, but when this kind of becomes more or less a model that not just can, you know, do work in the text and language and agent, but actually maybe can think visually as well and generate audio for you and has the full context of, say, all the stuff you've done over the last few weeks. You know, you don't need to say, give it the reference, give it the right prompt. It just like, already has this context and can reference this as needed in this, like, continuous space. This excites me a lot. The other side, I'll say, is the real time stuff that's, you know, we're seeing a little bit of it now out there. I think that this is like very early and I think it's going to be very exciting for real time video, audio, duplex interactions, you know, being able to, I don't know, you see a little bit of these like interactive, you know, stuff with like genie or you know, play the game but also on the side of, you know, you can bridge this back into like robotics where it needs to take in the real world in real time and make decisions.
- [47:04]
C
Yeah, I was gonna say that sounds really familiar in terms of interest on that side of things. So yeah, really cool. Great conversation and your lead in to what to what Black Forest Forest Labs is doing was also really good contextually in terms of kind of explaining. So hope our audience got a lot out of that. Dustin, thank you very, very much for coming on the show. Great conversation and as you've, as you've hinted there are things to come and looking forward to to having our next conversation as things move forward a little bit.
- [47:38]
D
Absolutely. Thank you so much for having me.
- [47:45]
A
Alright, that's our show for this week. If you haven't checked out our website, head to PracticalAI FM and be sure to connect with us on LinkedIn X or BlueSky. You'll see us posting insights related to the latest AI developments and we would love for you to join the conversation. Thanks to our partner Prediction Guard for providing operational support for the show. Check them out@prictionsguard.com also thanks to Breakmaster Cylinder for the Beats and to you for listening. That's all for now, but you'll hear from us again next week.