A Deep Dive into Freepik's AI Innovation with CEO Joaquin Cuenca

Embark on a captivating journey with Joaquin Cuenca, the visionary CEO at the helm of Freepik, as he unveils the secrets behind transforming a simple image search tool into a behemoth of creativity visited by millions each month. Cuenca's narrative whisks us through a landscape where AI-driven innovation meets the demands of digital artists and designers. The discussion peels back the layers of Freepik's user-centric philosophy, highlighting the ingenious use of artificial intelligence that enables a staggering 1.5 billion image downloads yearly.

Navigating the stock image galaxy can be fraught with black holes of generic content, but Cuenca sheds light on how Freepik's AI technology sculpts the nebulous search for that elusive "lady in a red dress" into reality. The synergistic dance between AI and human creativity sparks a new era of personalized image generation, mitigating biases and elevating processing speeds to empower users in their quest for originality.

As we cast our gaze toward the horizon, the future of AI in image editing emerges with Picasso, Freepik's latest suite of tools. Cuenca delves into the intricate balance struck between the necessity of free services and the economic realities of premium offerings. We stand witness to the transformative ways artists leverage features like Reimagine and sketch-to-image conversion to craft visuals that were once inconceivable, cementing the role of AI as a potent ally in the creative process. The complexities of copyright and evolving business models are dissected, ensuring that Freepik's commitment to innovation marches forward, hand in hand with the artists' ever-expanding canvas of possibilities.

Energize your sales with Shareme.chat, the proven texting platform. 

ShareMe.Chat 
ShareMe.Chat platform uses chat-to-text on your website to keep your customers connected and buying!

Mediaclip
Mediaclip strives to continuously enhance the user experience while dramatically increasing revenue.

Buzzsprout - Let's get your podcast launched!
Start for FREE

Disclaimer: This post contains affiliate links. If you make a purchase, I may receive a commission at no extra cost to you.

Chapters

• 0:02 Freepik's Evolution in Image Creation
• 9:19 Stock Image Generation Challenges and Solutions
• 20:41 Ownership and Business Models in AI
• 27:40 The Future of AI Image Editing

Transcript

Erin Manning: 0:02

Welcome to the Dead Pixels Society podcast, the photo imaging industry's leading news source. Here's your host, Gary Pageau. The Dead Pixels Society podcast is brought to you by Mediaclip, Advertek Printing and Independent Photo Imagers.

Gary Pageau: 0:18

Hello again and welcome to the Dead Pixels Society podcast. I'm your host, Gary Pageau. Today we're joined by Joaquin Cuenca, the CEO and co-founder of Freepik. Now Joaquin has been on the podcast in the past December of 2021. So you're encouraged to go back in time if you haven't already listened to that episode, and listen to that so you can hear the background of this interesting company. But they've got so much more news about how they've kind of pivoted and added tools and added functionality to their website that I said it's time to have them back on. So, hi, Joaquin, how are you today?

Joaquin Cuenca: 0:52

Hi, Gary, I'm doing very well and thank you for having me back.

Gary Pageau: 0:55

So, for those who aren't going to take the time or may have forgotten the last episode, tell us a little bit about FreePik, kind of you know the elevator pitch, if you will, of kind of what, basically what the company does. Then we'll go into where you're going.

Joaquin Cuenca: 1:11

Sure, so we started in 2010, at the end of 2010, beginning of 2011, as a stock image website. Very initially, we were a search engine. We help people find great images that were available for free on the internet. We pivoted we started creating images and allowing artists from the world to upload images to Freepik. We created a marketplace and we came aside for stock illustrations, photography, icons.

Joaquin Cuenca: 1:43

We had plenty of content stock illustrations, photography, icons. We had plenty of content and we became very, very popular because a big part of our collection was available for free to everyone in the world. So I think we are the most popular stock site in the world in terms of downloads. In terms of visits we get over 100 million visits per month. In terms of downloads every year we get like 1.5 billion downloads on our page and lately, in the last year, year and a half, we have been creating new tools on our platform to help people create and edit images and, of course, it's been with using the latest developments.

Gary Pageau: 2:28

Uh on ai yeah, so let's talk a little bit about that, because you you've pivoted a few times over the years, right, I mean, you started out as one thing, then you added another thing, then you've done anything. And adding tools is a big jump, if I think, from uh, because instead of being a, a resource for people to find things and to you, you know, find what they're looking for, then now they can create their own vision on your platform.

Joaquin Cuenca: 2:52

First of all, let me say that we actually are in a luxury position because we we pivot from being a success to, hopefully, a bigger success.

Gary Pageau: 3:03

Right, yeah, yeah, yeah, yeah, exactly.

Joaquin Cuenca: 3:06

It's not like we were failing. Before, like, as a search engine, we were very popular and we were creating our own content. We were able to deliver more value to the users, and that's what we're looking at. Okay, so we are listening to the users and trying to understand what we can do to give them more value. So one of the feedback that we get the users and trying to understand what we can do to give them more value. So one of the one of the feedback that we get is that people don't always find what they want on the website.

Joaquin Cuenca: 3:33

Again, right, and there are more ways. There will be multiple ways to address that problem. You can go and source more content because there are diminution returns on that. Right, you get content that just becomes too niche, becomes too expensive, and we are very, very cost-conscious and we pass those savings to the user. So our offer is very, very affordable. It's part of what we are. So there is a point where we need to cut our losses Again. We cannot chase the long tail forever. We need to be very conscious of what we're getting into the, and the plan B was to, uh, give the user enough flexibility so they could create the assets that they want without, without having to commission them to write a platform, and that's why we started getting into AI again to build a platform, and that's why we started getting into AI, and then it's been one step after the other.

Joaquin Cuenca: 4:29

Like a consequence of doing that is that we created the first text to image that we put on Freepik. But the problem with that that users have is that very often it gives you bad results, let's be honest, because you don't get the right image right.

Gary Pageau: 4:45

right because I mean to be honest. Most people don't know how to describe an image right, absolutely so sometimes uh you know, an image is actually way better than words.

Joaquin Cuenca: 4:58

it's very difficult, exactly, and computers are not extraordinary yet at understanding long sentences. So you give them a great description and you're not guaranteed that they will generate what you want, and you have no guarantee that it will be flawless.

Gary Pageau: 5:15

Or even like, if you change the sequence of those images, right, If you put the same number of words in one way, you'll get a certain set. But if you change them around so maybe certain descriptions are in the front of the sentence it can come out completely different.

Joaquin Cuenca: 5:33

Yeah, let's say that what we call understanding for computers is still a bit fuzzy, it's complicated.

Joaquin Cuenca: 5:41

We could not address all the issues, that we have been changing our technology to address what we can address, right.

Joaquin Cuenca: 5:50

So something that we did is that we found ways to speed up the generation and in a way, we decided that that was good because we will be able to fail faster.

Joaquin Cuenca: 6:01

Right, we are going to give you wrong images and these users will be able to discard them very quickly and find, uh, those that actually work and not much faster, uh. Another thing that we did is we realized that sometimes people have a very precise idea on their heads what they wanted and it was very difficult to describe it and even when they did it, we were not understanding it. So we added a mode where you can sketch an image type, a description and a sketch actually in the composition. So you get an image that gets generated based on the composition and the description. Right, and the cool thing, the wow factor, is that you get those images generated on the fly as you draw. Right, and the cool thing, the wow factor is that you get those images generated on the fly as you draw Right, because we were able to speed so much the generation that we were getting images in under 100 milliseconds. So we were able to generate them on every stroke.

Erin Manning: 6:59

Right.

Joaquin Cuenca: 6:59

Okay and again and we keep iterating Like another one is that we see that images very often have. They are okay, but they have some glitches Right A part of the image is actually the opposite of how we humans work. Like we humans, we create an image, you think about drawing something, we usually get all the details right, but then you start building up the composition and if you are not a good designer you lose track and you know it ends up completely messed up. You have the wrong size. You know the sizes are all wrong. A computer is actually completely the opposite. They nail the composition but then when they go into the detail, they completely, they completely get.

Gary Pageau: 7:46

Everyone's seen those AI-generated images with people with six fingers and an elbow going the wrong way and you just know there's something off, right.

Joaquin Cuenca: 7:55

Yeah, and actually that changes how people consume those images With a human-generated image from our first sight. If it works, it's good With an AI generated image. You actually need to find out all the mistakes.

Erin Manning: 8:10

Right.

Joaquin Cuenca: 8:11

It's like a game where you can get in trouble if you don't spot the mistakes Right. So we are building a tool that is a retouch, where you can go over small sections of the image. Go over them and it regenerates that part. It tries again. We're trying to patch the limitation of the current systems.

Gary Pageau: 8:35

You talked about something earlier that I took a note on, which I think is interesting, because you can't chase the long tail forever. I think that was very interesting because there is that sort of idea back in the old stock photo days or the stock image or stock graphics ages, that we just load in all this stuff. You know, eventually people will buy that one icon or that one thing because it will just be everything. And I think what's happening now, like you said, it's sort of that's, that's just, it's just wasteful. It's like there are things people aren't going to ever use, but they are going to want something customized that suits their own interests or needs. At that point, yeah.

Joaquin Cuenca: 9:19

Listen, there are there are cases where, by definition, almost users cannot use something that exists already, like if you are looking for a logo for your company obviously you cannot use a stock logo. And there are bespoke use cases, like if you want a drawing of your father, your mother, your kids, whatever from a picture, it doesn't work. You need to go to an illustrator, or now you can do it with AI. Well, there are many use cases where stock images were not cutting it, and what happens in practice on the stock sites is that, first of all, you need to get an approximation of what the user searches. Right, because you don't have the right thing. Okay, if you want something very precise a lady with a red dress that is looking backwards and just start getting precise, at some point you need to compromise. So, when you have a collection of images that you can show and you don't know which one is the most relevant one or the user search, you need to start ranking them and the signals that we all use. They're usually based on past performance.

Gary Pageau: 10:33

Right.

Joaquin Cuenca: 10:34

Also, if many people prefer that image in the past, it's likely that it's very good and that will be better than some other image, right? So basically, what you get is a game where a few players start winning consistently because they won in the past and you get some images that are left in dust because they don't get their interaction and they never surface. So there is always a subset of the images on a stock collection that have a decent number of downloads and revenue for the creators of the images, and a huge number of images that never see the light and sun, so to say, a waste, I mean. At some point. I think we need to understand that it will be difficult to monetize images that are just too similar to so many other images that we have, because there's so many, like the lady with the red dress you just mentioned.

Gary Pageau: 11:37

There could be thousands of variations of that right.

Joaquin Cuenca: 11:43

And listen if you come up with something original. The challenge here is that people usually you need people searching for this original thing, right, original? Right which is exactly kind of contradictory right you need to nail. It's a very difficult game because you need to nail something kind of new, but it that is trendy.

Gary Pageau: 12:04

Right.

Joaquin Cuenca: 12:04

Yeah, exactly.

Gary Pageau: 12:07

It's almost like a hashtag, right, in the sense that it has to be unique enough that the content will stand out, but it has to be something people actually search for. Yeah, so if I search for a volcano, but instead of lava, it's chocolate pudding coming out of it. That's very unique and specific and may not exist, but who knows, maybe it's chocolate pudding coming out of it. That's very unique and specific and may not exist, but who knows, maybe it does. I don't know, but it's kind of interesting how, now that people can have these tools, they can actually create that. Let's say, for example, I was a company that made chocolate pudding. That might be something that I want and I can't find that in stock. So maybe I go to your company and I create that.

Joaquin Cuenca: 12:50

Yeah, and for the record, we try to play synergies with the stock content that we have, like in your example, if you find an image of a volcano, that is kind of close to what you want, but it's not spitting chocolate or something.

Gary Pageau: 13:04

Yeah, exactly.

Joaquin Cuenca: 13:06

You can. We want to make it very seamless to go from this image and say, okay, I want this, but spitting chocolate. Yeah, this image and say, okay, I want this, but spitting chocolate, yeah, and yet the volcano is spitting chocolate just like the other one, and actually counting that as a second yeah for the person so it's an iteration right yeah, so we try to use the images that we have as seeds of new creation, and when the user does that, we attribute it it to that.

Gary Pageau: 13:35

When you were developing your AI platform, were you able to use your own data sets to help train the AI?

Joaquin Cuenca: 13:45

So far, we have used not so much of our data set. We have used like a few million images, but it's more like correlating the searches that we have on our platform. Yeah, particular images that behave well into existing models Okay, we have been working. We find them for different reasons. One of the reasons is to get results that are closer to what we have Right On the platform. Another reason is to try to remove bias from the models. For example, taking a stock model, if you search for women just that word women in 30% of the cases it comes out naked that were women in 30% of the cases it came out naked Right, and that's not what you should expect, right, right, I hope so.

Joaquin Cuenca: 14:42

You know that statistic doesn't come from what users expect. It comes from the statistic of the images on the internet.

Gary Pageau: 14:49

Right, exactly.

Joaquin Cuenca: 14:50

Yeah, yeah, yeah.

Gary Pageau: 14:51

And you need to rebalance that statistic to get closer to your expectations Right, because if somebody on your platform, if they were looking for that, they would probably specify that.

Joaquin Cuenca: 15:00

Exactly. I mean, it's very different.

Gary Pageau: 15:03

Yeah, I mean your customers and clients and whatnot, are not really a reflection of the image catalog on the Internet as a whole. They're looking for something different. So that's for a business use or promotion or commercial or an ad or personal expression or something like that.

Joaquin Cuenca: 15:24

This is it, you know, like sometimes nudity is perfectly valid yeah.

Gary Pageau: 15:30

But they will probably specify that in the search thing.

Joaquin Cuenca: 15:32

Exactly, but they need to be precise. You don't want that by surprise or by accident. So that's another reason to fine-tune models, and I would say another big one is to make them faster. Actually, the method that we have to make them faster is through fine-tuning, so it gets a bit technical, but it's one of those fine-tuning, so in one of those fine-tunes, so they are in one of those fine-tunes. It's what we use data that we have from images that we have from our platform.

Gary Pageau: 16:00

I'm kind of fascinated a little bit about some of the tools you have, because it's not just and again, because it's becoming more and more commonplace. You know the AI image generator where people are basically, you know, licensed, you know getting, you know data sets or whatnot from. You know some of the various pools that are out there. It seems like anybody can put this stuff in there.

Gary Pageau: 16:23

But you've kind of taken the approach of it's almost an iterative process in the sense where you know it's pretty easy to refine the image as you go without having to resubmit the entire query to begin with, because and sometimes it's frustrating as a person who's trying to generate an image let's say, for example, you get something, you just want to tweak it a little bit. But if you, in a lot of cases, if you resubmit it, you say, okay, I want the dress to be blue or whatever, you're not going to get the same look, you're going to get something completely new but with the dress being blue, whereas I think with your system you just can kind of do like tweaks and kind of keep that same look as it goes.

Joaquin Cuenca: 17:03

Yeah, let's say that in our mind, there is a big gap between having a nice technology and having a product. The early technology here, which is pretty amazing, is being able to convert a text to an image or, more properly, an embedding to an image. The product needs to take into account how people will actual people will use that. Having to type the full description of an image is actually pretty difficult, because you need to come up with ideas that describe the lining, the perspective, the graphic style of the image. If you don't do that, you will get a very narrow set of results.

Gary Pageau: 17:46

Right.

Joaquin Cuenca: 17:47

And most people don't have top of their mind, like different names from lining conditions, perspectives, styles, right, and that's something that we surface in the UI to make it easy top of their mind, like different names from landing conditions, perspectives, styles, right, and that's something that we surface in the ui to make it easy to play with. That again, we win by being fast so users can iterate very quickly, and by giving many, many choices, right. Okay, something that we do internally is that we also play with what we call the prompt. We don't take verbatim the problem that you submit and we try to with what we call the prompt. We don't take verbatim the prompt that you submit and we try to inject some alternatives to the prompt so that you can see more variety without having to specify it in detail.

Erin Manning: 18:23

Right.

Joaquin Cuenca: 18:25

In a way, how I see it. I often use an example about Chat GPT. Everybody knows about Chat GPT. The core technology here was the GPT. Obviously you know, everybody knows about Chat GPT. The word technology here was the GPT, and the GPT was a model that was able to complete text right. So you start typing something and it's just.

Gary Pageau: 18:45

You know it was an autofill type of technology exactly like it is the technology.

Joaquin Cuenca: 18:49

The technology is something that presumes sequence of text. This is it, and OpenAI had the brilliant idea of saying okay, the sequence of texts that you have to complete is this is a conversation between an assistant and a user. The user says colon, and there they put what the user says. System says colon, please complete. And then comes a conversational engine. It's a way more useful product. Sometimes you need this slight twist to take a technology and then to build a product that is used by everyone.

Gary Pageau: 19:27

So what do you think the impact is going to be with the AI-generated product line to your existing business?

Joaquin Cuenca: 19:37

Well, very difficult to forecast. There is a line of images that will be saved from AI generation, right, for example, all the editorial images. Sure, then we want photos from the real world. They are images from places that, in my opinion, are quite safe, because, if you want to know how Malaga, our hometown, is, you don't really know Malaga, not the dream of a system.

Joaquin Cuenca: 20:06

Likewise if you want to know how people from different places in the world really are. There are many use cases that are not okay with an approximation to that. They want the real thing, right, sure. So I will say that there is a set of images stock images in general, editorial images that are safe from AI-generated images, and there are a few ones that I expect in the future to be kind of obsolete.

Gary Pageau: 20:36

Sure.

Joaquin Cuenca: 20:38

It will be generated with more precision.

Gary Pageau: 20:41

Let's talk a little bit about kind of the ownership case, because you know, when you're talking about a stock image or a stock graphic or something, there's an ownership there. The person's created it, they own the copyright. Well, at least in the US they've determined that in AI images no one owns it.

Joaquin Cuenca: 21:22

Yeah, it's a hot topic Again. Multiple people have different opinions on this one.

Joaquin Cuenca: 21:29

So in Japan, for example, they rule that whatever comes out of a model is completely disconnected from the dataset that was used to train this model, and, as far as I understand, the European Union and the US have not yet issued a clear guidance on this matter, so it may be that it will maybe not be considered fair use if that happens. So in our case, we will need to retain a model from scratch using data that we have licensed. I will say that the vast majority of, I think, all AI generators will disappear in the short term, and I think the whole GNI industry will take a hit of probably a couple of years, and the speed at which we are making progress, it will probably not take longer than that. There are already a few generators that are using only licensed data. Sure, so with the Google one, it's using licensed data. I don't know the status of the last tally. Firefly from Adobe, famously, is using licensed, with the caveat that they use some images.

Gary Pageau: 22:45

They use images from mid-journey sure it'll be interesting to see what happens. I guess I'm. What I was thinking of was. Let's say, for example, I'm a person who, up until now, was creating a lot of stock pictures, right, I was going to the eiffel tower and I was taking a lot of pictures, and I did sunsets, I did sunrises, I did crowded days. I built a portfolio of those images that I wanted to monetize. Now, for certain uses, maybe some of those images will be because, like you said, they're unique or they're editorial or something like that but for the most part, I think those kind of people are going to have a challenge because, let's say, they start creating images within a generative platform, they still don't, depending on how the law shakes out, they're not going to own that image, right? So it'll just be interesting to see what happens to your creator community.

Joaquin Cuenca: 23:38

I mean stock image contributors. They were not having an easy time before. Already, it was challenging On this one, we don't know what will happen In our case. We are also licensing data to be used to train new models, and when we do that, we share revenues with contributors of 50%.

Gary Pageau: 24:04

I'm not saying there's not an opportunity, it's just going to be a very different one.

Joaquin Cuenca: 24:06

Yeah, I mean just speaking from our experience on like stock photos. Still early days but we see potential this year to get more revenues from training AI models than from the traditional model of license, sure Image.

Joaquin Cuenca: 24:28

Okay, but it's absolutely true. I mean, there is potential for some of these use cases to vanish, to disappear. So, in your example of the Eiffel Tower, I will take any time of the day, the real Eiffel Tower with all the details. I will not take a dream of that. But you are right, there are many use cases that probably many of them that had low value or that didn't exist before. The way I think about it is that it's actually also opening new doors. There were things that were impossible before that now become possible, like if you wanted an illustration for a block, for example it was not viable to uh to hire an illustrator to do that right for a single.

Joaquin Cuenca: 25:15

Okay, you don't get enough revenue for a single block, right, and now you can do that, right. Okay, you wanted to illustrate a book for your kids. You could hire an illustrator. But we all know what happened is that you will not win.

Gary Pageau: 25:27

Right.

Joaquin Cuenca: 25:28

Now it's something that can be done and those models at least those who are paying a license for the images that they used to train they extract some revenues and part of those revenues get to the original people that contributed those images.

Gary Pageau: 25:46

Awesome. So you said something earlier also that I made a note of but I thought was interesting, which was that your system allows you to fail faster, which I thought was interesting. Can you expand on your thought process behind that? I just thought that was kind of an interesting phrase.

Joaquin Cuenca: 26:06

Yeah, it's been very realistic. There is a small fraction of the generated images today that actually work. So to get a good image with current state-of-the-art AI generators, it's a process that still takes minutes. It can take longer than that. In a way, by failing faster, you can get rid of all the images that don't work and you can succeed faster also. That's part of the thought process, right?

Gary Pageau: 26:40

I just made a note of that phrase. I'm going to use that somewhere. I'm not wasting time, I'm just failing faster phrase. I'm going to use that somewhere. You know I'm. You know I'm not wasting time, I'm just failing faster. So talk a little bit about how the business model it is for pre, for free pick on this. You're using a kind of a name called Picasso, kind of to bundle these things under on your website. It's free pickcom, slash Picasso, p, I, ks-o, and then you've got your selections under there. What was your idea behind that, that sort of segmentation of it?

Joaquin Cuenca: 27:16

Yeah, I mean. You know, the only thing that is harder than getting a brand well-known is to repurpose a brand. Right, that's even more difficult. A brand?

Joaquin Cuenca: 27:25

well known is to repurpose a brand Right. That's even more difficult, okay, so the thing is that Freebik is very well known as a place where you can get free images, and we already made a pivot by allowing paid images on the platform. Yeah, yeah, that was. Picasso is something completely different for us. It's giving people a set of tools to create and to edit images, and we thought that we needed to market it as a separate thing, so we created Picasso and they bundled all the tools that we have been developing and they all have an angle where we try to help creatives, artists, create images faster. Yeah, you know, something funny is that we actually have not seen any cannibalization so far in our collection. We still have the same number of downloads, still like growing little by little on the content side, but we get new people joining on the AI side that are creating new things. Sure, and both are using the program.

Gary Pageau: 28:31

And then there's a revenue model there right when people have to pay at some point to generate this, because, I mean, I know it's a shocker to most people who are on the internet and think everything is free, but there's a lot of cost overhead to what you're doing actually, the gross model is completely different with stock.

Joaquin Cuenca: 28:52

Uh, you have an initial cost to get an image, but then it's so virtually free to distribute that image. Yeah yeah, with ai generators you actually pay a small fee every time you generate an image. Yeah, so actually the cost to run this part of the business is pretty substantial, yeah.

Joaquin Cuenca: 29:11

So it's really difficult to do that for free. You know we need to charge people, yeah, so we have a free tier. But then if you join, if you get the free subscription, you get a much, much higher limit on the number of images that you can generate.

Gary Pageau: 29:30

And is there a limit on, or do you have various, say, listen, you get a higher resolution download or that kind of thing, or is it just on the number of images you generate?

Joaquin Cuenca: 29:40

No, there is. Also. There are limits on number of images. There are limits on number of images that you can enhance. There are limits on that. You can redact. You cannot enhance up to 4K if you are a free user. So there are certain limits. Enhance is a very, very expensive process, for example, until you only get one. I think that's the last limit that we have and if you become premium, we try to rise. In general in the platform we try to be unlimited within reason Right. We try to prevent people from crawling our site.

Gary Pageau: 30:18

Yeah, yeah, that would be a big challenge, I would imagine is to not have your costs being exploded by people just generating, generating, generating Because you are being charged. It's a small amount but it's still going to add up over time. If you're doing millions of images a month that you know there's a cost there, whereas in the past, just doing a catalog search for a volcano, going back to the volcano, there's very little cost for that. So it is a very different model. How did you decide or are you still playing with? What is that offering? Is that going to vary based on the products, or is it kind of seeing what the uptick is? Because I imagine you probably have a goal of we have X number of free users and we want Y number of premium users.

Joaquin Cuenca: 31:02

Well, in general, we want to. First of all, we want to build products that add value to the user. So the first metric is retention of the users before we get into revenue. And that's important for us, because even free users give us a ton of value, because by observing how they react to the data, to the images that we, we create, we can know when we fade and we can know, you know, how to improve the product. But then we try to think of okay, how much can we give or freeze that we increase that retention and when it's to ask for, for uh, the user to become premium, to become a subscriber.

Joaquin Cuenca: 31:43

It's not an exact science. Exact science. We try to be generous on the free side, but at some point we just set the limit. But in general the rule is try to make the premium limit as high as realistically possible.

Gary Pageau: 31:59

What is the most surprising thing you've seen for use on your platform, because I think the sketched images is crazy and I can't wait to play more with that. But I mean, have you seen anything come through some of your users? That is just an amazing application.

Joaquin Cuenca: 32:18

Reimagine had a ton of incredible applications.

Joaquin Cuenca: 32:23

So Reimagine the one that we build, where you submit an image right uh, it actually thinks about variations of this image, right, because it understands the image, uh, and then it dreams of different images that look similar to that one. That has been super interesting, like plenty of artists are using it to can get a little bit of certain DVDs like, surprise me, I want things like this, but it's not accessible. Okay, so give me new ideas. Uh, that has been very useful. So, in general, the the reaction to reimagine has been spectacular. The reaction to the sketch to image uh, it was mind-blowing, the sketch to image. I I guess that they surprise everybody because for many was the same the first time that they saw something like this. I remember showing it live to a person here and they were speechless. They were like just like this, saying what is this? You know, without it, I mean, what is this? What is this?

Gary Pageau: 33:33

magic. You're showing what is this.

Joaquin Cuenca: 33:35

Yeah, yeah, but they were serious, you know, yeah, yeah, yeah, uh, people were very impressed.

Gary Pageau: 33:42

Yeah, because I yeah, I think that is sort of like. There's a many lot of people you know I've had this discussion with a lot of people about sort of this. Stuff is that, you know, not everyone is a storyteller, Not everyone's an artist, but everyone wants to be right. And for the ability for someone to just, you know, sketch that raw idea they have and then see it on a screen and then be able to iterate that to kind of get it to the way they want it, I think is going to be a really interesting tool to see how that gets used you know, we have an artist in house, melissa.

Joaquin Cuenca: 34:14

She's a photographer, she so she started photography. She has a very strong background in photography and now she's an excellent ai artist and her feedback was that before I had all those images on my head, you know I knew what they wanted, but I didn't have a budget to set up the scene, set up the lights, you know it's very expensive, yeah, yeah exactly that's one to be to capture like image that is on your head, right.

Joaquin Cuenca: 34:44

And well, she says, is that now, with AI, I can actually instruct for the right image and I can get the images out of my head. So for her it was kind of liberating, yeah, and she was doing absolutely stunning pieces of art.

Gary Pageau: 35:00

Yeah, that's awesome, that's great. So where can people go to get more information about FreePIC? Where do you want them to go?

Joaquin Cuenca: 35:09

Well, I would suggest to just go to our landing page, freepiccom. Browse from there Again. We have more traditional stock content images, and we have a section where we explain all the AI tools that we have. All we ask is just you play around. Okay, play around a little bit. Look at our tutorials that we have on YouTube. I hope that you will find you know new ways to work with email.

Gary Pageau: 35:37

Yeah, it's a great tool set. I really like it. Like I said, I've been playing with it. I have an account there. I've been playing with it. It'll be fun. You guys will probably get a premium out of me at some point once my free ones dry up, but thank you Well. Thank you so much, sir. It's always good to see you. Thank you, hopefully we'll see you in person this fall at some industry events, and it's always great to have you. Joaquin, thank you so much.

Joaquin Cuenca: 36:00

Thank you, gary, it was my pleasure. Thank you.

Erin Manning: 36:03

Thank you for listening to the Dead Pixels Society podcast. It was my pleasure, thank you. Thank you for listening to the Dead Pixel Society podcast. Read more great stories and sign up for the newsletter at wwwthedeadpixelssocietycom.