May 12, 2025 • 19 mins
Interested in being a guest? Email us at admin@evankirstel.com
Imagine turning your biggest frustration into a business that serves millions. That's exactly what happened when Joaquin and his team got tired of spending hours hunting for quality images while designing websites back in 2010.
"The main idea was we're scratching our own itch," explains Joaquin, CEO and founder of Freepik. What began as a simple search engine for free images has evolved into a comprehensive AI-powered design platform that's revolutionizing how creative professionals work. Their user-centric approach has driven continuous innovation, from adding icons and photos to embracing the transformative potential of generative AI two years ago.
Today, Freepik offers a suite of sophisticated tools leveraging approximately twenty different AI models under the hood. Their video generation capabilities provide end-to-end production with integration to leading models like Glyn, VO2, and Runway. Their image generation suite features their proprietary Mystic model alongside Google's Image N3 and various Flux options, all enhanced by carefully curated visual styles. This technical sophistication is wrapped in an accessible interface designed to simplify complex workflows.
Most remarkable is their unwavering commitment to accessibility. Despite the computing costs associated with AI generation, Freepik maintains a strong freemium model, positioning themselves as the most affordable option in the market. "When you make something 10 times cheaper, you get a different use case," Joaquin notes, explaining how dramatic price reductions enable entirely new applications and ultimately expand the market.
As they navigate the complex ethical landscape of AI-generated content, Freepik remains committed to respecting creator rights, using only properly licensed images or those without explicit restrictions. Their enterprise solutions even allow companies to disable specific models based on their comfort levels with different technologies.
Looking ahead, Freepik continues innovating with new interface paradigms, team collaboration features, and intelligent assistants that orchestrate complex workflows automatically. For Joaquin, a technical founder who witnessed the early internet, today's AI revolution represents an even more profound technological shift with unlimited potential.
Want to experience how AI is transforming creative workflows? Visit Freepik today and discover tools that make sophisticated design accessible to everyone.
More at https://linktr.ee/EvanKirstel
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:01):
Hi everyone.
Fascinating discussion today,talking about the intersection
of Gen AI with creativity anddesign, and so much more, with
the CEO, founder of FreePic,joaquin how are you?
Speaker 2 (00:16):
I'm doing great.
Speaker 1 (00:18):
Thank you, Evan.
I'm really intrigued by yourjourney and the mission at
Freepik.
Let's go back to the beginning.
What was the big idea behindFreepik?
And you've been at this for 15years.
That's quite a journey.
Speaker 2 (00:34):
What was the original vision.
Yeah, it's making me old.
We started in 2010, beginningof 2011.
And the main idea was we'rescratching our own itch, so to
say.
So we were making websites andat the time, the bottleneck for
us to make a great website wasto get great images.
(00:54):
It was a slow process.
We were all the time on theinternet chasing for great
images and we said, hey, whydon't we make it faster so that
we can just make better websites?
And we did a little searchengine for free images.
We call it FreePic, verynaturally, so you can think of
(01:14):
it as originally, as kind of aGoogle for free images.
It was in some vertical and wemoved from there.
Like we have been iterated sincewe launched it, like looking at
what will make our productbetter for our users, like
understanding that our coremission was to help people make
(01:35):
great designs faster.
Okay, so that was our mottolet's help them make something
great faster.
We look at the main pain points.
We added icons, we added photosand we created our own content
and something we should down thesearch engine.
And, of course, two years ago,we got Gen AI and we saw that we
(01:58):
can completely change approach.
We can, we could like reachsomething that was way more
generic than the product that wehad until then.
We could make the images thatthe users needed, very specific
and very bespoke images.
So we started on the journey ofGenEye Images two years ago,
and it's been a wild ride sincethen.
Speaker 1 (02:20):
And you have quite an amazing, you know, value
proposition here and the sitelooks incredible.
You know, maybe describe AI,the role of AI.
Now it's pretty central to manyof your products.
How do you see AI redefining,reinventing the creative process
(02:41):
?
Speaker 2 (02:42):
Sure, listen at its core, like the first models that
we got were something a littlemachine where you input some
text and you get a little imagein output.
Fair enough.
So we said, okay, this is alittle bit simple, and people
out there, especiallyprofessionals, they need more
reverse workflows.
(03:02):
They need sometimes they get animage but it's not quite there.
That happens quite often.
Very often the image that youget is 70% there, but it has
glitches, it has things that youneed to fix.
What people did at the time wasto download the image, to go to
another software like Photoshop,change it, tune it, maybe go to
(03:24):
Lightroom, fix the colors andthen ship it.
And we said, ok, can we justmake this last mile edition kind
of integrated in the generation?
Can we make something thathelps you generate that image
and then redash it, upscale it?
That was a huge problem with nogood solution.
(03:45):
Maybe changing the aspect ratio, all the things that the
professional user needs.
Can we put it all in a singleproduct so at the end of the day
, we don't have any more likeone single model that we use.
We have under the hood like 20different models.
I like to think of it like acomputer.
Inside a computer, you havemany different microchips, you
(04:08):
have a CPU and you have memoryand you have the hard drive and
you have the GPU many differentthings and you need to put them
all together to make a product,to make something that people
can use and like it or not, andthey buy the best computer.
They don't buy the best model,they buy the best workflow, they
buy the best product and that'swhat we are doing.
Speaker 1 (04:30):
Fantastic and you have a kind of freemium model
that serves both entry-levelusers, enthusiasts, as well as
professional designers.
How do you cater to such adiverse audience?
Speaker 2 (04:43):
Listen, freemium it's on our ethos, Like we started
something that was created alsoto make the great design more
affordable to people.
So it's kind of on our DNA.
So since the very beginning, wehave always had a very strong
freemium position.
It's not something that is asmall thing for people, it's
(05:06):
something really that can standon its own.
But of course, the bigdifference between AI models, AI
product and a stock product isthat there is a marginal cost to
create a new image.
There's a very solid cost to doAI.
So the premium product on theAI suite, I have to confess, is
(05:28):
like lower.
It's less generous than on thestock side because it's just
impossible to make it with anacceptable margin, but it's
still very approachable.
It's something that we want tomake very affordable.
If you compare FreeBig even thepaid product you compare it to
our peers we are usually by farthe cheapest in the market and
(05:51):
it's not because we want to becheap, but because it's on our
DNA to make it accessible.
We really want to make itaccessible to everybody and we
also believe philosophicallythat when you make something 10
times cheaper, you get adifferent use case.
The difference is notquantitative, it becomes
qualitative, it enables new usecases that were not available
(06:12):
before, and very often those usecases, they are, like another
magnitude bigger.
So even though you have a lowermargin, if you price it very
low, you usually end up makingmore money at the end of the day
.
Speaker 1 (06:27):
That's a great commercial approach.
So you have on your websitetools like Sketch to Image, ai,
video generation, backgroundremover, on and on.
What are some of the featuresthat have gotten the most buzz
and feedback from customers?
Speaker 2 (06:44):
As you mentioned, video is a huge one.
Our upscaler is a huge one.
On video, you can go now, likeall the way to the final product
, like end-to-end.
We even have an online videoeditor where you can stitch
together multiple clips, addtext, add static images.
So this one is phenomenal Underthe hood.
(07:06):
It has access to all the latestmodels.
So you have access to Glyn, toVO2, runway, luma Labs all of
them.
So that requires some expertiseon the part of the user, because
very often it's not trivial toknow what is the best model for
what they want to do, and that'ssomething that we need to
improve.
We want to make it even easierfor users so that, if you don't
(07:30):
know exactly what model to pickup, we make a decent default
choice for whatever you want tomake.
So video is going very, veryhigh.
The audio one is picking upsteam lately.
So we have a product that helpsyou make voiceover, helps you
make sound effects, so youintegrate video with sound
(07:51):
effects, lip sync, and you haveeverything that you need to make
great commercials.
So that's a vertical use casethat we are doing.
That is working very well and,of course, the image generation
site is huge.
That's our traditional coreexpertise and it's the biggest
one that we have.
So, by image generation, we haveagain under the hood, like
multiple models, to get an imagegeneration.
We have again under the hood,like multiple models, to get an
(08:13):
image.
We have our own model, which isMystic, that delivers top-notch
quality images.
Then you have Image N3 fromGoogle.
You have Flux Flux 1.0, 1.1,flux Dev.
You have many different visualstyles that have been created by
our team, so each one of themhas been hand curated and we
(08:37):
generated what is called a LoRa,which is kind of a
specialization of a model.
So this delivers exactly thevisual style that you are
looking at.
So, in general, imagegeneration will be first, video
generation will be second.
Speaker 1 (08:53):
Fascinating and you have obviously to navigate very
difficult ethical considerations, copyright issues that must
take a lot of your time andeffort to work through.
Speaker 2 (09:07):
Absolutely.
We took a lot of time to getcomfortable with the legality of
the models, so we run multipleconsultations to check the law
in Europe and in the US.
So I'm going to talk about theone that I'm more familiar with,
which is the law in Europe, andit is very often the most
restrictive set of laws.
(09:28):
Very often, if you are good inEurope, you are good worldwide.
So how it works in Europe isthat you can use an asset to
train a model if the author ofthe asset has not withdrawn the
rights to use it to train models.
(09:49):
So it's an opt-out model.
So by default in Europe you canuse any image where the author
has not opposed to its use Tomake it work.
In Europe it's mandatory andthat's very recent.
When you create a model, it'smandatory to list what are the
(10:10):
copyright holders of the imagesthat you are using so that they
have an opportunity to oppose.
But in general, all the modelsthat we use, they have withdrawn
all the images from all thestock sites and there was a
project that was created so thatartists can oppose to the usage
of their images to train AImodels and I think the number of
(10:34):
images that it has where peoplehave opposed is like one
billion.
So none of those images havebeen used in any of the models
that we host.
Now there are some models thattake it one step below, one step
beyond that.
So we have some models that arecompliant with that law, for
(10:55):
example, flux.
They are not using any imagethat has a post, so it's clean.
Then there are some models thattake it one step beyond.
I'm talking about Google ImageEntry.
They use a slightly morerestrictive policy, which is
they only use images that theyhave licensed themselves, and
that's also true for the modelthat we have developed.
(11:16):
We only use images that we havelicensed to use.
So the modeling by Google, whatthey did is they paid the
author of the images.
In this case, there aremultiple data sets that you can
license to train those models,and ourselves we have many
images that we have licensed andthat we can use to train our
(11:36):
models.
So that has been our philosophy.
Speaker 1 (11:41):
Fantastic, and this space remains so dynamic and
competitive.
Every day there's a new model,a new announcement open source
versus the big tech giants.
How do you stay ahead of thiswave?
You know, how do you manage toleverage best-of-breed
technology but stay ahead of thecompetitors.
(12:04):
What's your philosophy there?
Speaker 2 (12:06):
Well, for the most part, we try to be in a position
where we benefit from thosetailwinds, so to say.
Okay, so we are.
As I mentioned, we are buildingthe bridge between models.
We are building all thescaffolding, all the
infrastructure that people needto use the models.
When you generate images, youneed to be able to search past
(12:27):
images easily.
You need a place where you arestoring them.
You need to be able to sharethem with your colleagues.
There is a lot involved intoworking with images, not just
the pure generation, and that'swhere expertise is Now.
Concerning the models, we havebest in class state of the art,
(12:47):
and when there's a new model, weusually integrate it in ours.
So we are very, very fast tointegrate the latest and best,
and we're in a position wherenow, when there's a new model
popping up, it actually reducesthe number of bugs that people
have with our product.
When you look at complaintsfrom people, the number one is
(13:09):
hey, I'm trying to do this, I'mexplaining it clearly, but it's
not picking it up.
That happens quite often.
So every time there's a newmodel, the percentage of errors
just goes down.
So we have happier customerswhen there are new models
popping up.
Speaker 1 (13:28):
Yeah, really well said.
What are your customers askingfor?
Next, you have so manydesigners enthusiasts.
What are they asking for interms of new features, new
functionality, new services atthe moment, Listen, the answer
(13:49):
here is that it depends on thecustomer.
Speaker 2 (13:51):
We came from having a very core customer that was the
graphic designer, the marketer.
Now, with AI, the truth is thatthis has expanded to multiple
expertise.
We have people working onfilmmaking, we have
photographers, we have differentkinds of professionals now, and
each one of them has adifferent pet peeve.
(14:13):
One thing that is common tomany of them is you mentioned it
before like legality, we wantsomething that we can purchase
at our company.
So they want insurance and theywant, like, the legal certainty
that they can use the modelsthat they have.
So something that we haveintroduced.
We are launching an enterprisesubscription where admins can
(14:35):
shoot down models that they arenot comfortable with okay, even
though, again, like we havechecked all of them, according
to our lawyers like all of themare not comfortable with okay,
even though, again, like we havechecked all of them, according
to our lawyers, like all of themare legal to use.
But different companies mayhave different opinions and we
respect that.
So that's one concern, likebeing more friendly to
corporates.
Another one is working in teams.
So we launch projects, welaunch sharing of projects, and
(14:59):
then it's like there have beenlots of different ways and an
incredible amount ofexperimentation in the user
interface on how you createimages.
So we started with, I would say, a relatively trivial it was
not trivial but kind of trivialimage generator, and now we are
(15:21):
moving into experimenting.
Ok, how about having anassistant?
Because now we have so manytools?
Can we have a central way toask the machine for what they
want and see how it uses thetools that it needs to pull it
up together?
Can we enable now, automaticworkflows?
Can we help the users?
And now you have a place whereyou say, hey, I want to make
(15:45):
this kind of image, maybe tomake the particular image that
you want.
It's like combining sixdifferent steps and makes it by
itself encapsulated in a singleworkflow.
Now we are experimenting.
Also, we have a visualrepresentation of your workflow.
That is great to work withgroup of images.
(16:06):
Sometimes you want to clusterimages and you want to say, okay
, I want to use an image in thestyle of this cluster here and I
want to use the object that isin this other cluster here and I
want to take this cluster andwant to upscale it.
There are many things like thatwhere, with the traditional UI,
(16:30):
it will require many dynamicon-the-fly clustering and there
is usually no record of what wasthe selection that you made.
So you pick like 10 differentimages, you do your operation
and the selection is gone andsometimes you need to do like
multiple operations with thoseselections.
(16:50):
So we are experimenting withnew user interfaces to make it
easier to work with groups ofimages.
So there is I mean it's a longanswer, but you know short
answer is it depends on the user.
Different users have differentneeds and we are trying to see
what are the common themes onthose things.
Speaker 1 (17:13):
Wonderful.
Well, it's amazing to hear,looking at your biography,
you're a technical co-founder,ceo, your first company was
acquired by Google many, manyyears ago and you're a developer
, I guess as well.
What's it like staying on topof the tech stack?
You're building Hardware,software, gpus, cloud.
(17:36):
I imagine you have a lot goingon there as well on the back end
.
Speaker 2 (17:40):
It's super exciting.
I often say that this is thebest time I got in my career and
I remember clearly when theinternet was starting.
I remember at university likethe first email, the very
beginning before the WWW, whenwe had GoFair.
You know the good old days whenthe internet was like sprawling
(18:01):
was starting and it was superexciting.
And this is I think this iseven bigger than that.
It's like the feeling that Iget.
If you get back to the days ofthe internet early internet it
was almost like almosteverything that you tried that
was new kind of clicked in theuser because it was solving a
(18:24):
huge problem that wasdistribution.
It was difficult to distributethings okay, so you had this new
shiny hammer that was useful,uh, and that you know.
There were so many use casesthat were applicable to that and
, yeah, it's kind of the same.
It's funny how many differentthings we try and the hit rate
(18:44):
is super high.
The hit rate of the number ofthings that we try versus the
number of things that click tothe user is huge.
So I think that we are reallyat the very beginning of
something that we'll redefine inthe next 20, 30 years.
Speaker 1 (19:00):
Well, I can't wait to embark on that journey and it's
exciting times, I agree, themost exciting time in my adult
lifetime and you're doingamazing work onwards and upwards
.
Congratulations on all thesuccess.
Speaker 2 (19:14):
Thank you.
Thank you.
It's a work of a dream, and Iwant also to say thank you to my
team.
It's around 500 people thathave been doing an amazing job,
and I want to highlight that wehave been through an amazing
transformation.
It was not trivial to move froma company that was working on
(19:34):
an adjacent domain, like stockimages, and was able to move
solidly into Gen AI.
Speaker 1 (19:42):
And much more movement and transformation
ahead.
So exciting times.
Congratulations.
Take care, wakil.
Thank you and thanks everyonefor listening and watching.
Take care.
Hi everyone.
Fascinating discussion today,talking about the intersection
of Gen AI with creativity anddesign, and so much more, with
the CEO, founder of FreePic,joaquin how are you?
Speaker 2 (00:16):
I'm doing great.
Speaker 1 (00:18):
Thank you, Evan.
I'm really intrigued by yourjourney and the mission at
Freepik.
Let's go back to the beginning.
What was the big idea behindFreepik?
And you've been at this for 15years.
That's quite a journey.
Speaker 2 (00:34):
What was the original vision.
Yeah, it's making me old.
We started in 2010, beginningof 2011.
And the main idea was we'rescratching our own itch, so to
say.
So we were making websites andat the time, the bottleneck for
us to make a great website wasto get great images.
(00:54):
It was a slow process.
We were all the time on theinternet chasing for great
images and we said, hey, whydon't we make it faster so that
we can just make better websites?
And we did a little searchengine for free images.
We call it FreePic, verynaturally, so you can think of
(01:14):
it as originally, as kind of aGoogle for free images.
It was in some vertical and wemoved from there.
Like we have been iterated sincewe launched it, like looking at
what will make our productbetter for our users, like
understanding that our coremission was to help people make
(01:35):
great designs faster.
Okay, so that was our mottolet's help them make something
great faster.
We look at the main pain points.
We added icons, we added photosand we created our own content
and something we should down thesearch engine.
And, of course, two years ago,we got Gen AI and we saw that we
(01:58):
can completely change approach.
We can, we could like reachsomething that was way more
generic than the product that wehad until then.
We could make the images thatthe users needed, very specific
and very bespoke images.
So we started on the journey ofGenEye Images two years ago,
and it's been a wild ride sincethen.
Speaker 1 (02:20):
And you have quite an amazing, you know, value
proposition here and the sitelooks incredible.
You know, maybe describe AI,the role of AI.
Now it's pretty central to manyof your products.
How do you see AI redefining,reinventing the creative process
(02:41):
?
Speaker 2 (02:42):
Sure, listen at its core, like the first models that
we got were something a littlemachine where you input some
text and you get a little imagein output.
Fair enough.
So we said, okay, this is alittle bit simple, and people
out there, especiallyprofessionals, they need more
reverse workflows.
(03:02):
They need sometimes they get animage but it's not quite there.
That happens quite often.
Very often the image that youget is 70% there, but it has
glitches, it has things that youneed to fix.
What people did at the time wasto download the image, to go to
another software like Photoshop,change it, tune it, maybe go to
(03:24):
Lightroom, fix the colors andthen ship it.
And we said, ok, can we justmake this last mile edition kind
of integrated in the generation?
Can we make something thathelps you generate that image
and then redash it, upscale it?
That was a huge problem with nogood solution.
(03:45):
Maybe changing the aspect ratio, all the things that the
professional user needs.
Can we put it all in a singleproduct so at the end of the day
, we don't have any more likeone single model that we use.
We have under the hood like 20different models.
I like to think of it like acomputer.
Inside a computer, you havemany different microchips, you
(04:08):
have a CPU and you have memoryand you have the hard drive and
you have the GPU many differentthings and you need to put them
all together to make a product,to make something that people
can use and like it or not, andthey buy the best computer.
They don't buy the best model,they buy the best workflow, they
buy the best product and that'swhat we are doing.
Speaker 1 (04:30):
Fantastic and you have a kind of freemium model
that serves both entry-levelusers, enthusiasts, as well as
professional designers.
How do you cater to such adiverse audience?
Speaker 2 (04:43):
Listen, freemium it's on our ethos, Like we started
something that was created alsoto make the great design more
affordable to people.
So it's kind of on our DNA.
So since the very beginning, wehave always had a very strong
freemium position.
It's not something that is asmall thing for people, it's
(05:06):
something really that can standon its own.
But of course, the bigdifference between AI models, AI
product and a stock product isthat there is a marginal cost to
create a new image.
There's a very solid cost to doAI.
So the premium product on theAI suite, I have to confess, is
(05:28):
like lower.
It's less generous than on thestock side because it's just
impossible to make it with anacceptable margin, but it's
still very approachable.
It's something that we want tomake very affordable.
If you compare FreeBig even thepaid product you compare it to
our peers we are usually by farthe cheapest in the market and
(05:51):
it's not because we want to becheap, but because it's on our
DNA to make it accessible.
We really want to make itaccessible to everybody and we
also believe philosophicallythat when you make something 10
times cheaper, you get adifferent use case.
The difference is notquantitative, it becomes
qualitative, it enables new usecases that were not available
(06:12):
before, and very often those usecases, they are, like another
magnitude bigger.
So even though you have a lowermargin, if you price it very
low, you usually end up makingmore money at the end of the day
.
Speaker 1 (06:27):
That's a great commercial approach.
So you have on your websitetools like Sketch to Image, ai,
video generation, backgroundremover, on and on.
What are some of the featuresthat have gotten the most buzz
and feedback from customers?
Speaker 2 (06:44):
As you mentioned, video is a huge one.
Our upscaler is a huge one.
On video, you can go now, likeall the way to the final product
, like end-to-end.
We even have an online videoeditor where you can stitch
together multiple clips, addtext, add static images.
So this one is phenomenal Underthe hood.
(07:06):
It has access to all the latestmodels.
So you have access to Glyn, toVO2, runway, luma Labs all of
them.
So that requires some expertiseon the part of the user, because
very often it's not trivial toknow what is the best model for
what they want to do, and that'ssomething that we need to
improve.
We want to make it even easierfor users so that, if you don't
(07:30):
know exactly what model to pickup, we make a decent default
choice for whatever you want tomake.
So video is going very, veryhigh.
The audio one is picking upsteam lately.
So we have a product that helpsyou make voiceover, helps you
make sound effects, so youintegrate video with sound
(07:51):
effects, lip sync, and you haveeverything that you need to make
great commercials.
So that's a vertical use casethat we are doing.
That is working very well and,of course, the image generation
site is huge.
That's our traditional coreexpertise and it's the biggest
one that we have.
So, by image generation, we haveagain under the hood, like
multiple models, to get an imagegeneration.
We have again under the hood,like multiple models, to get an
(08:13):
image.
We have our own model, which isMystic, that delivers top-notch
quality images.
Then you have Image N3 fromGoogle.
You have Flux Flux 1.0, 1.1,flux Dev.
You have many different visualstyles that have been created by
our team, so each one of themhas been hand curated and we
(08:37):
generated what is called a LoRa,which is kind of a
specialization of a model.
So this delivers exactly thevisual style that you are
looking at.
So, in general, imagegeneration will be first, video
generation will be second.
Speaker 1 (08:53):
Fascinating and you have obviously to navigate very
difficult ethical considerations, copyright issues that must
take a lot of your time andeffort to work through.
Speaker 2 (09:07):
Absolutely.
We took a lot of time to getcomfortable with the legality of
the models, so we run multipleconsultations to check the law
in Europe and in the US.
So I'm going to talk about theone that I'm more familiar with,
which is the law in Europe, andit is very often the most
restrictive set of laws.
(09:28):
Very often, if you are good inEurope, you are good worldwide.
So how it works in Europe isthat you can use an asset to
train a model if the author ofthe asset has not withdrawn the
rights to use it to train models.
(09:49):
So it's an opt-out model.
So by default in Europe you canuse any image where the author
has not opposed to its use Tomake it work.
In Europe it's mandatory andthat's very recent.
When you create a model, it'smandatory to list what are the
(10:10):
copyright holders of the imagesthat you are using so that they
have an opportunity to oppose.
But in general, all the modelsthat we use, they have withdrawn
all the images from all thestock sites and there was a
project that was created so thatartists can oppose to the usage
of their images to train AImodels and I think the number of
(10:34):
images that it has where peoplehave opposed is like one
billion.
So none of those images havebeen used in any of the models
that we host.
Now there are some models thattake it one step below, one step
beyond that.
So we have some models that arecompliant with that law, for
(10:55):
example, flux.
They are not using any imagethat has a post, so it's clean.
Then there are some models thattake it one step beyond.
I'm talking about Google ImageEntry.
They use a slightly morerestrictive policy, which is
they only use images that theyhave licensed themselves, and
that's also true for the modelthat we have developed.
(11:16):
We only use images that we havelicensed to use.
So the modeling by Google, whatthey did is they paid the
author of the images.
In this case, there aremultiple data sets that you can
license to train those models,and ourselves we have many
images that we have licensed andthat we can use to train our
(11:36):
models.
So that has been our philosophy.
Speaker 1 (11:41):
Fantastic, and this space remains so dynamic and
competitive.
Every day there's a new model,a new announcement open source
versus the big tech giants.
How do you stay ahead of thiswave?
You know, how do you manage toleverage best-of-breed
technology but stay ahead of thecompetitors.
(12:04):
What's your philosophy there?
Speaker 2 (12:06):
Well, for the most part, we try to be in a position
where we benefit from thosetailwinds, so to say.
Okay, so we are.
As I mentioned, we are buildingthe bridge between models.
We are building all thescaffolding, all the
infrastructure that people needto use the models.
When you generate images, youneed to be able to search past
(12:27):
images easily.
You need a place where you arestoring them.
You need to be able to sharethem with your colleagues.
There is a lot involved intoworking with images, not just
the pure generation, and that'swhere expertise is Now.
Concerning the models, we havebest in class state of the art,
(12:47):
and when there's a new model, weusually integrate it in ours.
So we are very, very fast tointegrate the latest and best,
and we're in a position wherenow, when there's a new model
popping up, it actually reducesthe number of bugs that people
have with our product.
When you look at complaintsfrom people, the number one is
(13:09):
hey, I'm trying to do this, I'mexplaining it clearly, but it's
not picking it up.
That happens quite often.
So every time there's a newmodel, the percentage of errors
just goes down.
So we have happier customerswhen there are new models
popping up.
Speaker 1 (13:28):
Yeah, really well said.
What are your customers askingfor?
Next, you have so manydesigners enthusiasts.
What are they asking for interms of new features, new
functionality, new services atthe moment, Listen, the answer
(13:49):
here is that it depends on thecustomer.
Speaker 2 (13:51):
We came from having a very core customer that was the
graphic designer, the marketer.
Now, with AI, the truth is thatthis has expanded to multiple
expertise.
We have people working onfilmmaking, we have
photographers, we have differentkinds of professionals now, and
each one of them has adifferent pet peeve.
(14:13):
One thing that is common tomany of them is you mentioned it
before like legality, we wantsomething that we can purchase
at our company.
So they want insurance and theywant, like, the legal certainty
that they can use the modelsthat they have.
So something that we haveintroduced.
We are launching an enterprisesubscription where admins can
(14:35):
shoot down models that they arenot comfortable with okay, even
though, again, like we havechecked all of them, according
to our lawyers like all of themare not comfortable with okay,
even though, again, like we havechecked all of them, according
to our lawyers, like all of themare legal to use.
But different companies mayhave different opinions and we
respect that.
So that's one concern, likebeing more friendly to
corporates.
Another one is working in teams.
So we launch projects, welaunch sharing of projects, and
(14:59):
then it's like there have beenlots of different ways and an
incredible amount ofexperimentation in the user
interface on how you createimages.
So we started with, I would say, a relatively trivial it was
not trivial but kind of trivialimage generator, and now we are
(15:21):
moving into experimenting.
Ok, how about having anassistant?
Because now we have so manytools?
Can we have a central way toask the machine for what they
want and see how it uses thetools that it needs to pull it
up together?
Can we enable now, automaticworkflows?
Can we help the users?
And now you have a place whereyou say, hey, I want to make
(15:45):
this kind of image, maybe tomake the particular image that
you want.
It's like combining sixdifferent steps and makes it by
itself encapsulated in a singleworkflow.
Now we are experimenting.
Also, we have a visualrepresentation of your workflow.
That is great to work withgroup of images.
(16:06):
Sometimes you want to clusterimages and you want to say, okay
, I want to use an image in thestyle of this cluster here and I
want to use the object that isin this other cluster here and I
want to take this cluster andwant to upscale it.
There are many things like thatwhere, with the traditional UI,
(16:30):
it will require many dynamicon-the-fly clustering and there
is usually no record of what wasthe selection that you made.
So you pick like 10 differentimages, you do your operation
and the selection is gone andsometimes you need to do like
multiple operations with thoseselections.
(16:50):
So we are experimenting withnew user interfaces to make it
easier to work with groups ofimages.
So there is I mean it's a longanswer, but you know short
answer is it depends on the user.
Different users have differentneeds and we are trying to see
what are the common themes onthose things.
Speaker 1 (17:13):
Wonderful.
Well, it's amazing to hear,looking at your biography,
you're a technical co-founder,ceo, your first company was
acquired by Google many, manyyears ago and you're a developer
, I guess as well.
What's it like staying on topof the tech stack?
You're building Hardware,software, gpus, cloud.
(17:36):
I imagine you have a lot goingon there as well on the back end
.
Speaker 2 (17:40):
It's super exciting.
I often say that this is thebest time I got in my career and
I remember clearly when theinternet was starting.
I remember at university likethe first email, the very
beginning before the WWW, whenwe had GoFair.
You know the good old days whenthe internet was like sprawling
(18:01):
was starting and it was superexciting.
And this is I think this iseven bigger than that.
It's like the feeling that Iget.
If you get back to the days ofthe internet early internet it
was almost like almosteverything that you tried that
was new kind of clicked in theuser because it was solving a
(18:24):
huge problem that wasdistribution.
It was difficult to distributethings okay, so you had this new
shiny hammer that was useful,uh, and that you know.
There were so many use casesthat were applicable to that and
, yeah, it's kind of the same.
It's funny how many differentthings we try and the hit rate
(18:44):
is super high.
The hit rate of the number ofthings that we try versus the
number of things that click tothe user is huge.
So I think that we are reallyat the very beginning of
something that we'll redefine inthe next 20, 30 years.
Speaker 1 (19:00):
Well, I can't wait to embark on that journey and it's
exciting times, I agree, themost exciting time in my adult
lifetime and you're doingamazing work onwards and upwards
.
Congratulations on all thesuccess.
Speaker 2 (19:14):
Thank you.
Thank you.
It's a work of a dream, and Iwant also to say thank you to my
team.
It's around 500 people thathave been doing an amazing job,
and I want to highlight that wehave been through an amazing
transformation.
It was not trivial to move froma company that was working on
(19:34):
an adjacent domain, like stockimages, and was able to move
solidly into Gen AI.
Speaker 1 (19:42):
And much more movement and transformation
ahead.
So exciting times.
Congratulations.
Take care, wakil.
Thank you and thanks everyonefor listening and watching.
Take care.
Popular Podcasts
True Crime Tonight
If you eat, sleep, and breathe true crime, TRUE CRIME TONIGHT is serving up your nightly fix. Five nights a week, KT STUDIOS & iHEART RADIO invite listeners to pull up a seat for an unfiltered look at the biggest cases making headlines, celebrity scandals, and the trials everyone is watching. With a mix of expert analysis, hot takes, and listener call-ins, TRUE CRIME TONIGHT goes beyond the headlines to uncover the twists, turns, and unanswered questions that keep us all obsessed—because, at TRUE CRIME TONIGHT, there’s a seat for everyone. Whether breaking down crime scene forensics, scrutinizing serial killers, or debating the most binge-worthy true crime docs, True Crime Tonight is the fresh, fast-paced, and slightly addictive home for true crime lovers.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.