Subscribe
Share
Share
Embed
If you are seeking new ways to increase your ROI on marketing with your commerce platform, or you may be an entrepreneur who wants to grow your team and be more efficient with your online business.
Talk Commerce with Brent W. Peterson draws stories from merchants, marketers, and entrepreneurs who share their experiences in the trenches to help you learn what works and what may not in your business.
Keep up with the current news on commerce platforms, marketing trends, and what is new in the entrepreneurial world. Episodes drop every Tuesday with the occasional bonus episodes.
You can check out our daily blog post and signup for our newsletter here https://talk-commerce.com
Brent Peterson (00:02.789)
Welcome to this episode of Talk Commerce. Today I have Yair Adato. He is the CEO of Bria AI and they are democratizing visual generative AI technologies. Yair, go ahead, do an introduction for yourself. Tell us your day-to-day role and one of your passions.
Yair (00:20.978)
So thank you very much for inviting me to this. So my name is Yair. I'm a tech guy. I have a PhD in computer vision. And I actually see Gen.AI for the first time in 2014. And I thought that it's going to change the world. And I started Bria in 2020. So we are in this business of a visual generative AI for a while now. I'm the CEO. I'm also the founder.
In Bria, we created a developer platform for a visual generative AI, where the nice thing about our platform is that first it's for builders, for people that wants to create their own application startup company solution products. And the second one, it's a little bit surprising, it's safe to use every data that we used for this platform, we got a full license for it. So there's no...
issues of copyright, privacy, trademarks and other risks which associate with this technology.
Brent Peterson (01:26.875)
That's perfect. So Yair, before we get started, you have volunteered to be part of the free joke project. And so before we start using or talking about generative AI and this joke was not, I have tried using chat GPT or Claude to generate dad jokes. It doesn't work. There's no joke AI yet. And I think maybe on hogging face at some point there'll be a joke AI tool, but I'm going to just tell you a joke. Give me a rating one through five. So here we go.
Yair (01:53.884)
Okay.
Brent Peterson (01:55.417)
What's the difference between Iron Man and Aluminum Man? Iron Man stops the bad guys, but Aluminum Man only foils their plans.
Yair (02:10.726)
It's definitely something that ChatGPT cannot do. I will give it a four. I will give it a four.
Brent Peterson (02:13.401)
All right, thanks, I appreciate it. Good, so Yair, tell us a year, tell us a little bit about your background and why you started doing this and tell us about 2014 and what was machine learning like then?
Yair (02:30.866)
So I have a PhD in computer science as I sense in the computer vision. And I think that in the last 20 years, you see the revolution in computer vision from physically based computer vision. Let's try to mimic the light structure, the eyes, the brain to more a machine learning driven algorithm. And somewhere around 2020, maybe before, sorry, 2010,
We see the rise of deep learning, which was a huge, huge revolution. And then in 2014, we saw the first time for generative AI, the first time that I saw a machine that generate a new type of data, which I thought that it's mind blowing. It's something which when I saw it for the first time, I thought that it's going to change everything we know about data.
And this is where I started to think about how it's going to be used later on. An interesting thing about generative AI is that it's faster growing and getting more mature technology that I ever saw. It's become better and better every day, much faster than anything that was before that, which is interesting.
Brent Peterson (03:54.481)
Yeah, if we follow the path of least chat GPT or even Claude or Gemini, it is literally month after month. We see doubling the innovation, although I think now it's starting to slow down. So one of your things is that you're responsible. It is a responsible, you're using a responsible AI to, are you focusing on just visual AI? Are you doing generative text AI and video? us a little bit about that.
Yair (04:21.69)
Yeah, I don't know if it's slowing, it's worth revisiting this question. But for us, are doing mostly data which has rights. Visual, music, audio, video, 3D, things that are more easy to associate with the owner of the data. Because it's part of how we build our business, basically, what we believe should be a good, sustainable, responsible business to build.
Brent Peterson (04:52.913)
And how I envision it's sort of like the shop, like the Spotify model where you're streaming and you're giving the artists some credit for that. Is that sort of what you're looking at or you envision that every artist would be part of a community that then would get paid as people are starting to use their content?
Yair (05:15.078)
Yeah, absolutely. We develop an attribution engine. In Spotify, you pay exactly to the data point that you used. And the genius in Spotify is that they came with a model of that you can pay after you use it, after you consume the music, and you don't need to buy that in advance. We are doing the same, but instead of having one data points, which is not relevant in our case, we attribute it to multiple data points. It can be multiple
attributed to 100 different images or 50 or 150. It's according to the impact, according to what was generated and how many of the data points has concept that contribute for this specific generation. So it's per concept. So in a way we disassembly the visual one step further.
and we look on the concept that generates the image or the music or the video. And this is how we pay the attribution, the royalty, and everything is in agreement. So once we have the engine, we go to get the images and Shutterstock and Invertor and deposit for Tor and other, and we license the data with this agreement. So we first license the data, then we train our own model. Now we pay the attribution.
Brent Peterson (06:44.187)
And is that, so I guess one thing about creating content is that the content, once it's created, you've paid them, but now that content exists in perpetuity, is the license the same way?
Yair (06:57.532)
So the license is the same way in terms that once you generate a content, it's yours. The other thing that you are not allowed to do, for example, you cannot use this content to train a new model because we want to protect the data owners. We believe that there will be a new economy of data owners and creators that hold these rights and they will get rewarded according to the...
according to the concept that they will contribute. I actually think that it will be the same in text, in the skills of people, in other elements of how you contribute something and now the machine knows to create something new and you get your share out of it. And it can be for a general content like an image of two kids have a play in the forest or a sunset in the ocean that you can have from Getty Images. It can be also for premium content.
Harry Potter, Captain America, what was the joke about? It also can be part of the content that you can put in this type of ecosystem. Someone can generate, predictable, controllable, and then the data owner will say how much it costs and you will pay for him. So now it's give the opportunity to everyone to generate everything they want if they the right commercial agreement.
Brent Peterson (08:24.667)
What are you most excited about the written, the visual, or the video part of it in terms of how the licensing works and where it's going to go in the future?
Yair (08:36.198)
Yeah, I'm a computer vision guy. I'm mostly excited for the visual elements of that video images, illustration. 60 % of the computing the brain is associated with the visual streaming. So this is what I like. Recently, I more and more appreciate the music part of it. It's a cool stuff in the music and voice.
Text is important also, some other people will solve it.
Brent Peterson (09:12.325)
Yeah, think that so, you know, I think in our in our pre the pre interview I allowed or asked you to put in sort of a theme for a jingle and I use Udio, which is a one. I'm not just an online studio that you can create jingles. I'm also excited about the music part and I think it's it's I like what you're saying about democratization of how you are going to allow artists to to share in that. And I think one of the struggles that I've had in creating
music is it's you don't really have a way of pinpointing what you'd like it to sound like. You'd like it to sound like some artist, but right now you can't because there's protections. Talk a little bit about how you're bringing in. Well, let's back up a second. Do you have to do you have to approach Universal Music and Warner and those types of people to try to get them to license the music to you?
Yair (10:07.378)
Yeah, this is exactly the same concept. And there are also already some startups that are doing that. This revolution is here. It will be possible to generate music, visual, according to some premium content. The question is what will be the economy behind it. And there will be, must be economy. Otherwise this entire machine learning concept that it's based on data will collapse into itself because you will not have enough new data.
Someone will solve it. We have our approach with the attribution engine also for music. We will approach the artists. We say, is the agreement that we're going to have. If someone will generate something which influence or that you contribute the concept, you have the rights to say how much you want to charge it. And there will be a new market. The market as a market will become much bigger. There are two elements here.
how to do the commercial agreement behind this new technology, which is always the case. Internet creates new opportunity for new companies that distribute data much faster from Google to Spotify and Netflix to, if you wish, Uber in some extents and e-commerce of course is one very good strong example. Something similar will happen in generative AI and how to do the...
commercial agreement, arrangement, economy behind it, this is one pillar that we need to very carefully look at. The second one is how to do that in a way that it will be predictable and consistent and controllable. If I want to generate Harry Potter, it needs to be Harry Potter. If I want to generate an image according to a brand guideline,
It must be the colors and the font and the style of the brand. You cannot change it. I don't care what your AI is doing. This is my colors. And if I'm an e-commerce platform and I have a product, I have an amazing shirt with light blue and flowers, pink, this is my shirt. You are not going to change it. And the authenticity is going to be very important to this element.
Brent Peterson (12:34.363)
Do you think that, well, AI is breeding, in my opinion, it's breeding a whole bunch of new types of artists and people that can write prompts, that people are creative in writing those prompts. Is it hard to convince some of the other artists who are doing the traditional way of creation to participate with you?
Yair (12:57.202)
So when we started approach to data owners like Getty and Alamy and Vato, or to artists, we have also artists program. It was really hard to do that at the beginning, like three or four years, no one understand what we're talking about and why should we care? And then Dali came and Chechi PT and people become more aware of that from the artist side.
And then at the beginning, the artists say that AI is the devil and they don't want to speak with us at all. But then we said to them, hey, guys, there is an alternative and you want to be next to the table. You want to be in the room when the new alternative is shaping up. It was a process until we convinced them to join. A few months ago, one of the first artists when he saw his model with his illustration,
back in like walking, he said to one of the C-level in the company, the one that is doing all of the data work, he told her, I think that we discover a new continent. This is how we react to that. And you know what is interesting? We all have the access to the same model train on his art.
When I tried to do something, or very trying to do something, it became lousy because we are not have the touch, we don't have the skill, we don't know what is creative. And when he using the same model, he's doing amazing stuff, like amazing stuff. So all of this, we are going to change, are going on, we are all going to be out of work, AI will replace all of us. I don't really believe that. And again, I'm a tech guy.
This is what I'm doing the last 20 years. I think that it just give us a tool to do the thing that technology is doing all the time, automation. Just another way to do automation. But this time is a very basic concept, automation of data generation, which is a big, big thing.
Brent Peterson (15:11.163)
Do you find the struggle harder in convincing executives to get on board? And I think about my experience in a previous life, our client was Universal Music and it was at the time in which streaming had just become, it was killing their business and they had to switch. And I feel like in the music business anyways, they switched over too late or they were slow to switch from physical media to streaming.
Is it easier now because of that experience for people in like Getty images you mentioned earlier to move into this arena because they have to, they know that's coming and now they've maybe gone through some of that past experience?
Yair (15:58.354)
So some organization adjusts some organization.
It doesn't adjust. It's very simple. Look at the marketing agencies. There are like five or six big marketing agencies. You clearly can see the two or the three that adjust. And you clearly see the two or the three that understand that they are struggling. And now they're talking about merging and acquisition and try to be in a survival mode by creating some kind of...
Adel.
Instead of understanding that technology is going to change the business model they have, which is a risk but an opportunity at the same time, they try to continue with the same flow of the thing that they did and try to solve it with, let's merge with the second largest company. It will not work if you saw that with IPG and Omnicom, for example. But this is only my opinion. Hope that the CEO of Omnicom will not call me now.
Brent Peterson (17:08.001)
What do you see as the next big breakthrough? And you talked about some of those struggles to make the models consistent. And I've experienced that in trying to make a consistent image over time, even over prompts. Has that been one of the biggest struggles to try to get some consistency in how your images look on a generation?
Yair (17:28.658)
I think that as any technology, the technology needs to be to get more mature and people will start to solve the problem not as, wow, we have one big, big, huge model that solves everything, it doesn't work like that. There's a lot of misconception in the market about this technology. I think that what we will start to see now is the same.
Same thing that happened with deep learning, by the way. The black magic of how the trade this model will disappear. People understand that it's not a single model that solve everything. It's a platform, it's a combination, it's a technology. You need to combine between the different pieces and each one of them bring you a different advantage. We just released a few months ago, a few months ago, our eraser, remove object.
And once we create this flow, this pipeline to remove object, and not trying to have one single in painting with prompt engineer that solve everything, but we have one for eraser, one for image, for object, add object, want to replace object, immediately the controllability increase like 10 times, like it's working amazing. And we will see more and more.
productization like that in the market in all domains. Video, it will take a little bit more time because people are still convinced that if they will increase the number of parameters and consume more data, it will solve. It will solve only to very limited extent. You need to be smarter, not stronger.
Brent Peterson (19:19.089)
What are you most excited about for AI in terms of applying it to e-commerce? and especially in your realm where you now can guarantee that the licensing is not going to come back and bite a store. Let's just say a really big store owner, an enterprise level. How fast do you think that's now going to come into play where some of those people can do that? And then talking about democratization.
How far down is it going to go to the average user, the mom and pop Shopify store?
Yair (19:53.85)
Yeah, so we see some big, big e-commerce platform that start to build internal tools. So the mom and pop store owner, when they upload an image can modify it very easily and create instead of one image 20. We see that start, we see companies start to build that. We see companies rethink about
production, how they think about production of e-commerce. We see personalization at scale. If you have the ability to modify the entire catalog for Christmas in one click, why don't you do that? It's really cool actually. And we start to see people think about it. We see retailers, they're talking about e-commerce and instead of just putting several product
in a basket automatically also ask AI what are the product, what you can create and create me a dinner out of this product. And then you can switch between dinners and the same with the same basket. So we see a lot of innovation coming on that direction. I think it will take probably another 12 months until we will see a large scale deployment.
But it's coming, we see that.
Brent Peterson (21:23.761)
Yeah, and so you mentioned 12 months and I think in the past we would have said 12 years, right? I think that the rate of innovation is staggering. Where is Bria going now in the next, what is your goals for this next 12 months?
Yair (21:38.406)
Yeah, so our platform for visual is quite mature. We have really nice, big clients and small clients that use that. We have the ability to generate, to edit images in controllable and predicted way to tailor the generation, e-commerce, et cetera.
One thing that is interesting for maybe a little bit bigger e-commerce, bigger retailers, since we are a developer platform, we also give that as a source code available, which is nice. And again, since we are a developer platform, we also enable to use it any place you want. So if you generate something, if you create a dedicated engine, then you can use it at Bria, at Adobe, at Figma.
Canva to give it to WPP, to give it to your agency, to create a campaign. It's yours. You can decide what you want to do with that, which is a different approach, think. So this is exciting. Video is coming, which is really exciting for us. We just released the first video feature, which is a simple one. It's just remove background, but you need that in order to start to build everything on top of it. And music.
is also coming. It's going to be a very exciting year, I think.
Brent Peterson (23:06.587)
Yeah, lots of big plans. Just from a resource standpoint for the average user who don't understand the sort of resources just running a model on your own takes, tell us about how you've been able to scale up your resources. when I say resources, you're cloud computing the amount of scale that maybe the average user doesn't understand from a technical standpoint to make all this work.
Yair (23:34.608)
Yeah, I think that this is what we do. We are the expert of how to train the model and to make it very, very efficient and controllable. So it will be an easy problem for the builders to build. So we have builders which just use automatic training as a web tool and know how to connect it to Figma or to our platform, et cetera.
And then we have from the other side AI teams, which we help them with not only engineering, but also with algorithms. So we see the entire spectrum. To use the API is quite simple. It's quite straightforward how to use the API. It's just another API. The fact that behind the scene there's a lot of tech, it's almost irrelevant.
Brent Peterson (24:27.057)
That's great. Yeah, as I close out the podcast, I give everybody a chance to do a shameless plug about anything they want. You can promote anything you'd like. What would you like to plug today?
Yair (24:34.514)
You
Yair (24:38.226)
I think that the e-commerce use cases is super interesting. If you think about it, the fact that a lot of the e-commerce on the website is just a product with a white, not boring background is a shame, is a problem of technology and scalability. It's not how it should look like. And I think that what you can do with a platform like Bria, you can change it.
and you create something which is hyper-personalized, optimized, it's depend what you want to solve. And we are offering everything from a startup plan here, just start to build something, to self-service, to enterprise solution. Everything is safe, everything is compliant with future regulation. Speak with us, it will be cool, cool projects.
Brent Peterson (25:37.179)
And do you have, I think I met somebody at the shop talk in Chicago. Do you have some conferences or shows you'll be at this spring?
Yair (25:46.032)
Yeah, so we will be hosted on AWS booth in the next NRF. Actually, the thing that we are going to show there is how to build the e-commerce solution, which is hyper-personalized. Next week, we are going to be in GTC, the Nvidia conference. We are going to be in...
probably in almost every conference, you can find that somewhere. The nice thing is that we have a very good relationship with AWS, Microsoft and Nvidia. And usually you can find us on this conference or the other conference under their booth. So there's a lot of opportunity.
Brent Peterson (26:35.793)
That's great. I'll make sure I get your contact details and the show notes. is there, yeah, it's been a great conversation and thank you so much for being here.
Yair (26:48.114)
Thank you very much, it was fun. Thank you very much.