00:00I barely sleep at night because every
00:01time I wake up the next morning there's
00:03something new that's launched in the Arc
00:05of innovation a few years can feel like
00:07a long time or no time at all
00:12people ask are we there yet and also
00:14jump to point out that we're not exactly
00:17where people thought we would be
00:19but when Technologies hit that
00:21exponential phase things move really
00:23quickly sometimes so quickly that we
00:26forget that tools like mid-journey or
00:27stable diffusion literally did not exist
00:30one year ago now in a recent episode we
00:33discussed the first wave of technology
00:35used to build digital influencers and we
00:38did that with the creators of Lil
00:39Michaela Michaela was a first mover and
00:42she was built in 2016 but at the same
00:45time she was built with CGI
00:47but today we're actually presented with
00:49an entirely new set of tools these AI
00:51tools that give almost anyone the
00:54opportunity to build a digital
00:55influencer of Their Own
00:57so in today's follow-up episode I
01:00thought it'd be interesting to bring in
01:01two new voices seemingly from different
01:03worlds but also converging on one topic
01:06that topic being how does artificial
01:09intelligence fundamentally change the
01:11way that we all represent ourselves
01:14the first voice is Sinead Buffalo
01:17SHINee describes herself as a futurist
01:19but she also wrote an article titled I
01:21am a model and I know that artificial
01:24intelligence will eventually take my job
01:27but here's the thing she wrote this
01:29three years ago which again depending on
01:32the prism that you have on the world can
01:34both feel like yesterday and forever ago
01:37our second voice is Danny postma Danny
01:40has explored the many ways that AI can
01:43reshape how we express ourselves online
01:44by building the tools to actually enable
01:47so that ranges from the memes that we
01:50post to the tattoos that we put on our
01:52body to the headshots that we all post
01:53on LinkedIn and over the last few months
01:56yes months not years he's created seven
01:59different projects and that ranges from
02:01the things I just mentioned tattoos
02:03memes headshots but also a virtual
02:06modeling agency and honestly if you've
02:09been on AI Twitter recently I can almost
02:11guarantee you've seen at least one of
02:15so we brought in Sinead and we brought
02:17in Danny and together we explore how
02:19these tools might impact the world of
02:20modeling but also creators and just
02:22about any Forum that we as humans show
02:26we'll discuss the deflationary nature of
02:28these tools what they unlock how to
02:30build a moat and so much more and I'll
02:32also say that Sinead unfortunately had
02:34to drop a little bit early but Danny
02:35stuck around with us till the very end
02:37all right let's get started as a
02:40reminder the content here is for
02:42informational purposes only none of the
02:44following is investment business legal
02:46or tax advice and please note that a16z
02:48and its Affiliates May maintain
02:50investments in the companies discussed
02:51in this podcast please see a16c.com
02:54disclosures for more important
02:56information including a link to a list
03:09to have you two on the line today as you
03:12both know we did an interview with the
03:14creators of Lil Michaela just before
03:16this and Loma Kayla is created with CGI
03:19and it was started in 2016. I should say
03:21she was started in 2016 and now it feels
03:25like we're going through a second wave
03:27of sorts it feels like maybe anyone can
03:30create these characters online and that
03:32could be characters that are net new
03:34that are not necessarily representative
03:36of them or it could be representing
03:38their own personality their own likeness
03:40in a new way using you know I feel like
03:43the technology of the Year AI so before
03:47we get into that technology and what
03:50people are building there Sinead you
03:52wrote an article in 2020 with Vogue that
03:55I feel like was really prescient
03:57um it was called I am a model and I know
04:00that artificial intelligence will
04:02eventually take my job
04:04and here we are in 2023 maybe this does
04:06not sound surprising three years later
04:08but back then that was a statement and
04:11so tell me a little bit more about that
04:13article and what inspired you three
04:15years ago to write it yes so um you know
04:18I also work as as a futurist so a lot of
04:21my my time is spent varied in Data
04:23Tracking Trends and I had come across a
04:27company called Data grid using a form of
04:30generative AI called adversarial
04:32networks we would know more formally or
04:34informally as deep fakes if this
04:36technology can so accurately create
04:39artificial identities artificial beings
04:42what industry could that have a lot of
04:44ramifications for and the clear cut for
04:47me was was fashion especially when it
04:50comes to e-commerce a line or segment of
04:54fashion where it's not necessarily about
04:57creative big movements for a fashion
04:59model but instead just small variations
05:01in the posing and that in some ways is
05:04somewhat robotic already we don't have a
05:07lot of variability into the types of
05:08poses we can do or are the things that
05:10we can try when it comes to e-commerce
05:11it has to look good on a website and so
05:13that to me was a very clear trend line
05:16that I was able to to draw and I did
05:18also see as always we just need more and
05:21more fashion companies like zolando a
05:23big e-commerce Giant in in Germany also
05:26researching the technology and then when
05:27you combine it with a little Michaela
05:30was also a big influencer back then this
05:32was like not 2019 when I started to
05:34research all signs pointed to a future
05:36where Society might be okay with with
05:39artificial beings representing clothing
05:42and where it could be a much more cost
05:44effective future for for fashion
05:46companies totally and we're seeing this
05:48be extrapolated not just to fashion and
05:51we'll get to that but I want to hear
05:52when you wrote this article what was the
05:54response and maybe break that down both
05:57by the people in the modeling industry
05:59and then also outside yes so I it was
06:02definitely a shock to a lot of people
06:07for a long time even though technology
06:09has always disrupted fashion whether
06:11that's e-commerce or whatever it may be
06:13that nobody was really talking about the
06:16impact of automation on creative jobs at
06:19that time and especially not fashion
06:21modeling so the Fashion World
06:23a little bit of fear I would say in in
06:26the modeling world but I also tried to I
06:29don't like to just drop kind of
06:30fear-mongering things and leave
06:31everybody to fend to themselves saying
06:33things like it isn't just models right
06:34everybody's gonna have to prepare for
06:37the future of work so providing that
06:39um I think was a little bit reassuring
06:41for people but the amount of responses I
06:44received and not even just in the
06:46modeling world people in advertising
06:48photography all sorts of surrounding
06:51ecosystem players in the world of
06:53fashion that will inevitably be impacted
06:56by a world where we can digitally
06:58generate photo shoots the article was
07:01received as something highly likely but
07:04I think people couldn't conceptualize
07:06how this transformation would actually
07:08come about so it became suddenly true
07:11for people that okay in some way
07:13modeling might be impacted by AI like
07:17other Industries would be but they
07:19couldn't actually understand what that
07:21could actually look like it became hard
07:22to see how that would materialize and
07:25then now in this kind of explosion of
07:27generative AI it's like all of the light
07:29bulbs are going off uh connecting all of
07:31the pieces yeah and as I said it's not
07:33just models I think a lot of people as
07:36you said it's materializing where
07:37they're seeing how these tools today are
07:41maybe not replacing them but are able to
07:43do things that they previously thought
07:45they were safe from the advancements of
07:47these Technologies in fact that was
07:49really how people viewed it they were
07:51like if you're creative you know that's
07:52where the robots are unable to match you
07:55and we're now seeing that maybe that's
07:56not the case and so you know talking
07:58about some of these tools Danny let's
08:01move on to you created many different
08:03projects and we'll get into more of them
08:05but one of them was really you know this
08:07like Photo Studio of the future this
08:09modeling agency where you really could
08:11spin up a model put them in these
08:13different backdrops drop in your product
08:15things like that and so you posted this
08:18a couple weeks ago and it went really
08:20viral I think one tweet in particular
08:22reached like two million people but you
08:24also I think saw maybe similar to
08:27sinead's article a range of responses
08:30and so tell us a little bit about what
08:33that product was but also how people
08:34responded to it I've been dabbling with
08:36different kind of projects and it all
08:38came together and I made this tweets my
08:40product wasn't ready yet it's a deep
08:42agency treatment absolutely viral So
08:44within 24 hours I had to come up with
08:47something to put life because otherwise
08:49there wasn't the website online I think
08:51it went very old got 30 million views as
08:53a modeling agency and it's nowhere close
08:56to that I think there's a lot of other
08:57tools that are way better doing those
08:59kind of things but yeah as you said like
09:01all the technology is coming together
09:02right now all these different kind of
09:04models the ability to generate
09:07fakes deep like non-existing people but
09:10also the ability to put the clothing on
09:12those kind of models it's like all these
09:13separated tracks of development by
09:16researchers over the last past years
09:17that are all coming together in 2023
09:19turning into tools like this since
09:21you've been kind of like in the bowels
09:23of these tool Danny do you think you
09:25know let's say like a year from now or
09:27maybe even five years from now is it
09:29realistic that basically you can
09:31represent what the modeling industry
09:34does today or what influencers do today
09:37with these Technologies and have it be
09:39pretty much like a one for one or a
09:42match in terms of it being as good so
09:44what is able what I'm able to do right
09:46now what others are able to do right now
09:48is the ability to use dreamboof which is
09:51technology to train an AI on top of
09:53someone's face and you can basically
09:54make a clone a digital twin or whatever
09:57there are other models called stalgin
10:00which basically are trained on millions
10:02and millions of photos and they can
10:04generate phases and heads that do not
10:07exist at the moment they they are
10:08non-traceable it's they don't exist
10:10combine those two and you can train a
10:13dream Booth model on a non-existent
10:15person and create a model for example
10:16and then there are other deep learning
10:20models that are able to make a and I
10:23believe the name is like a top-down
10:24photograph of a clothing that's not
10:26being worn by someone and the AI knows
10:28how to put that clothing in kind of the
10:30folding ways to put it on top of
10:32someone's body being able to like dress
10:35the model in that kind of way so the
10:37technology is there at the moment I
10:38believe it's up to a company that
10:41understands the fashion industry to take
10:43it to the another level oh I love that
10:47this sense of taste of the people in the
10:50modeling industry are not just the folks
10:52who look a certain way or able to hold a
10:54camera a certain way but they do have
10:55this background this understanding of
10:57aesthetic and so Sinead tell me a little
11:00bit more about how maybe some of these
11:01companies are using this technology
11:03today are they leaning in and saying oh
11:05actually this is really valuable we're
11:07going to utilize these new tools because
11:10we are in this new era or are they kind
11:13of resistant to this new wave yeah I
11:17think it kind of depends on the type of
11:20fashion company so if you look at
11:22e-commerce and a company although we
11:24might not be as familiar with zoolando
11:26in America they're quite a big faction
11:28Giant in Europe for them to be
11:30researching this technology and showing
11:32very viable proof of concept as early as
11:352019 we know that fashion companies are
11:37exploring this and we can also say a
11:40company like h m they have thousands of
11:42data scientists that use AI to do things
11:45like forecast Trends and understand and
11:47analyze Supply change so they already
11:49have the the somewhat of the
11:52infrastructure to deploy the next
11:54advancements in in what the industry
11:56will be adopting I would say where we
11:59might see a little bit more resistance
12:01is for potentially in in high fashion
12:04when it comes to using AI for things
12:07like design although I do expect as any
12:10tool it will eventually be adopted but I
12:13would argue that there's probably been
12:15certain instances where any of us on
12:17this on this podcast right now have come
12:19across an AI model and just not realized
12:21it uh shopping online so I think the
12:24technology has been here for for years
12:26it's 2023 companies were investing in it
12:29in 2018. it's it's largely here it just
12:31hasn't materialized as broadly at scale
12:34yet but that's coming I think maybe one
12:37pushback people might have is like okay
12:39if this thing is AI generated it's not
12:42quote unquote real but like how real are
12:45the photos that we're seeing today
12:46anyway like ignore the digital virtual
12:49influencer concept the photos that we're
12:51seeing on a billboard like how
12:53photoshopped are those how real is the
12:55end image that we're seeing today I feel
12:58like there's a degree of fabrication
12:59already and so maybe you could just kind
13:01of speak to that concept and maybe again
13:04just like the natural pushback for us to
13:05say well this is not human anymore but
13:09past the original photograph where we
13:12already yeah I mean that it also depends
13:14on how representative the models are of
13:17of society so it could be argued that it
13:19is hard to find examples of diverse
13:23representation in the fashion industry
13:24as it stands today and so therefore
13:26seeing a piece of clothing on somebody
13:28else how helpful is that if it's if you
13:30can't necessarily or your body is a
13:33different shape or you can conform to
13:34the clothes in the same way and in terms
13:36of the actual shoots themselves of
13:37course there is a lot of pinning and
13:40last minute stitching and things to make
13:42things appear to fit a certain way and
13:45then of course the the Photoshop part
13:47towards the end I am not the person on
13:49the computer doing the Photoshop so I
13:51couldn't tell you exactly how much
13:53Photoshop there is at the end of the day
13:55yeah an image definitely does get
13:57altered so it isn't entirely like this
14:00documentary style photography where it
14:03just goes specifically from Shoot
14:05Straight to Market but at the same time
14:07there are interesting social and
14:08cultural friends that are converging the
14:11idea of photoshopping images is becoming
14:14a little bit more taboo people are
14:16speaking up against it that they want to
14:18see maybe some of the stretch marks or
14:20or the cellulite or the or the acne the
14:21things that make people human and so
14:24it's interesting that that's also those
14:26types of Social and cultural Trends are
14:28appearing at a time when we're arguably
14:30becoming less human in some ways or how
14:33fashion and culture is represented is
14:35becoming a little less human yeah I mean
14:36I think maybe one of the most
14:37interesting aspects of AI is that you
14:39can really prompt it to give you
14:41whatever you want and so you can prompt
14:43it literally to say hey I want this
14:45person to have acne I want this person
14:47to be of my race or culture or gender
14:49you can really prompt it to give you
14:51whatever you want but at the same time I
14:53think there is this yearning for what
14:55people again call real and I think that
14:58term is is going to take on a whole new
15:00dimension as we look forward but the
15:02last thing I want to ask just something
15:03as we talk about like fashion and
15:04modeling is maybe a sense of the cost
15:07the cost differential is something that
15:09ultimately plays a role in adoption and
15:12so Danny let's just use an example right
15:14so you have this AI model this person
15:17doesn't exist in fact you did create one
15:18of these uh her name is Alice and people
15:23um at this model does not exist.co I
15:25believe and we'll link that in the show
15:27notes let's just use that as an example
15:29if you wanted to do a photo shoot with
15:32Alice like how much would that cost and
15:33and let's break that down between like
15:35actually training the initial data set
15:37or the initial face and then actually
15:40generating a bunch of images let's just
15:42give the audience a sense of what that
15:45So currently to deep train a model on
15:50someone's face to be able to generate
15:52photos after that it costs 60 cents to
15:55generate this okay yeah so not even a
15:58dollar and then I think per photo your
16:00ads a fraction of a cent right now so
16:03you could generate hundreds of thousands
16:05of photos for less than 100 bucks so
16:08technology is pretty cheap already in
16:11so I believe this is this is from what
16:14I've seen talking to customers people
16:19uh question in my life chat is from Mom
16:21and Pop shops with close clothing brands
16:23who do not have the funds to hire models
16:26to put on clothing for the web shops who
16:28I think probably wouldn't even ever be
16:31able to hire a model to do the photo
16:33shoots for them so I see this as a it's
16:36like democratizing for the lower
16:39spectrum of the small business enabling
16:41them to do the fashion shoots and
16:44yeah I think it's mostly for those kind
16:46of uh consumers that this is interesting
16:48to do yeah and there's other
16:49applications of that like I I did a
16:51presentation recently and I use
16:54mid-journey for all of my images within
16:56that presentation and I probably would
16:59not have had those images I wouldn't
17:01have hired someone at the very least to
17:03go create them and so there is as you
17:05said this democratizing Force this
17:07ability for folks that didn't have
17:09access to a specific technology to use
17:11it in these new ways
17:13um Sinead let's hear the other side of
17:15that like maybe perhaps some of these
17:18who are models may not be hired for the
17:21same roles or maybe they will still be
17:23and in the case where they are what are
17:26the benefits of all this like can you do
17:28a bunch of shoots where you know your
17:31avatar is quote unquote attending the
17:34shoots without you actually physically
17:36being there can you save a bunch of
17:39Fuel and not traveling can you actually
17:42like upgrade yourself in some way I'm
17:44just kind of throwing ideas out there
17:46but how might this actually change the
17:48industry for the better is there a world
17:50in which a model can be in multiple
17:53places at one time and therefore
17:55increase their their income stream and
17:57have more assets of themselves to be out
17:59in the market absolutely that could be
18:02quite profitable quite lucrative if the
18:05industry does evolve that way in a world
18:08where we want to see the the same
18:11influencers and people we look to in
18:14fashion to continue to be the face of
18:16Brands and the clothing but if you look
18:18at the history of technology technology
18:20is often quite deflationary over time
18:23especially when it comes to wages so in
18:26a world where a model can be in multiple
18:28places at one time yes they may be able
18:30to have more revenue streams coming in
18:33but do we see downward pressure on
18:35prices because a lot of it's created
18:38digitally and there's an increase in in
18:41supply of identities that could pass for
18:44being human and so that's where things
18:46get a little bit gray and a little bit
18:48ethically red flags and kind of caution
18:51tape because I will say again it's not
18:52that this is unique to modeling we're
18:54going to see this across the board with
18:56AI it's going to put down the pressure
18:57on a lot of different wages from
18:59modeling to programming but that's where
19:01it becomes a little bit challenging and
19:04then what is uniquely applicable to
19:06modeling and maybe acting is that not
19:09only are people getting automated but
19:13their likeness and the communities they
19:15represent are also getting automated but
19:18someone is still benefiting off of that
19:20very specific identity so that's where
19:23it starts to become a little bit gray
19:25yeah I mean that's okay I mean it's
19:29going to be interesting to see how that
19:30all evolves if you you know like what
19:33regulation gets put in place what IP
19:35people own or do not own that goes into
19:38these models on the topic that this
19:40really does expand past modeling I think
19:43what's fascinating Danny is you've
19:44really went on the Sprint over the last
19:46few months you've created I mean I'm
19:48just gonna name some of the projects you
19:50did tattoos AI which is you know an AI
19:52powered tattoo artist we mentioned Alice
19:54this model that does not exist that you
19:56created you started I think or one of
19:58your early projects was profile picture
20:00AI which is similar to lensa which most
20:02people would recognize we talked about
20:03deep agency meme more people can morph
20:06themselves into their favorite meme and
20:08then most recently you're working on
20:10headshot Pro which is basically people
20:12can create their next corporate headshot
20:15or a company can create a kind of wave
20:18of head shots for their entire staff and
20:21so among those projects first just kind
20:24of tell me what you're seeing as you've
20:26launched these projects what people are
20:29getting excited about what you're
20:31getting a lot of traction with right
20:33away without maybe perhaps even trying
20:35and where actually you know this
20:37technology may be interesting but it
20:39feels more like you know a trek uphill
20:45so first of all like the whole reason
20:47this skills really fast for me and this
20:49gets a lot of attention because it's
20:51indistingual for magic at the moment you
20:52upload a few photos and suddenly you get
20:54hundreds and hundreds of profile
20:55pictures out for it so this is why
20:57people keep sharing it on Instagram on
20:59Tick Tock it goes fire really fast the
21:02second like I think what mostly gets
21:04solved at the moment is I get a lot of
21:06messages from folks who say I don't have
21:08any profile pictures I don't have any
21:09photos and suddenly I can look really
21:12well online this is one of the major
21:14selling points for headshot for example
21:16well like me for example I hate going to
21:18photo shoots because I always look super
21:20awkward I don't smile like so for them
21:23this like solves an actual issue that
21:25they can look good on their uh LinkedIn
21:27same one with profile picture you can
21:28turn yourself into whatever you want
21:30lensa showed how popular that could be I
21:32believe Lancer did 40 million dollars
21:34with that I like this as an art I like
21:36to dabble with it I like to try things
21:38out I get an idea I have to work it out
21:41gets easy and easy because it uses the
21:43same technology and the head Shield has
21:44been the biggest success traffic wise
21:47Financial wise at the moment for remote
21:49teams which is branded that at the
21:51moment like good luck trying to get a
21:53photographer if your team is remotely
21:54and you have people in Asia America
21:56Europe and you want to get the same
21:58photo style like you cannot send a
22:00photographer all around the world to get
22:02the same cell but everyone can upload a
22:05photo train an AI model on them use the
22:08same styles to generate it and basically
22:09everyone gets the same ad shift for that
22:11team if you're creative enough in it you
22:13try enough there are use cases that
22:15people think it's gonna take away the
22:17photographer but I think it's gonna
22:20different kind of things that weren't
22:22possible before with this technology
22:24like having a photographer everywhere in
22:26the world trying out different
22:28hairstyles for example uh things that
22:30weren't able before that are enabled Now
22:33by this technology yeah it's kind of
22:35crazy if you go to headshot Pro I think
22:37you have this live counter
22:40um and last time I checked it said 450
22:43000 plus headshots were created and
22:45that's fine I think it's a million now
22:47because I have to manually update it and
22:49that's two weeks live now yes I was
22:50gonna say I just checked it last night
22:52you kind of alluded to this but there's
22:54ways that people are using this product
22:56like you're you're creating these
22:58projects and you're releasing them into
23:01and you're kind of tinkering and
23:02thinking oh maybe people will use it in
23:04this way but I've seen you tweet about
23:05the fact that you've actually been
23:06surprised that as you release this tool
23:09people are using the tool in ways that
23:11you didn't even expect as the Creator
23:13well that's that's like uh with Provo
23:15picture I learned from the headshots and
23:17d-map shields because we used to do only
23:19these artsy Styles and people kept
23:21requesting hey can I get a LinkedIn
23:23photo for myself so we kept thinking
23:25with that way yeah and just on the note
23:27of what people are being drawn to I
23:29think maybe one way to have a pulse on
23:32that is what people are willing to pay
23:34for and so among your projects have you
23:37noticed any Trends there in terms of
23:38things that people almost seem allergic
23:40to like oh this is cool this is novel
23:42but like I'm not gonna pay you 30 for
23:45that versus maybe the headshot example
23:48of something that people have ingrained
23:50in their mind like oh a headshot is
23:52worth X dollars so with profile picture
23:54we started pretty expensive at the time
23:56and I've been lowering the price lower
23:58lower and I think we're almost around
24:00cost price at the moment five dollars to
24:02generate them so yeah lensa basically
24:04killed the whole market and after they
24:07went live right so yeah I think mostly
24:11um it's the high quality output and I
24:13really wonder where we're going to go in
24:14the next next few months because these
24:16models keep releasing keep releasing
24:20I barely sleep at night because every
24:21time I wake up the next morning there's
24:23something new that's launched that
24:24already makes the whole technology
24:26obsolete it's it's going so far so
24:30I don't know where it's gonna end up
24:31well I know I feel I feel like a lot of
24:34people would agree with that sentiment
24:36that it's moving so quickly it's so hard
24:38to keep up on the note of it evolving
24:40though right these new models are being
24:41released all the time what ways does
24:44this like really change or shape the way
24:46that people can represent themselves
24:47online I think you know if you use a
24:49headshot example that's kind of a
24:51one-to-one parallel to what we did
24:52before this technology existed but maybe
24:54what our new exciting ideas that people
24:57can look forward to I mean one example
25:00of you know this in fashion is like I
25:03can't remember how long ago this was but
25:04like Bella hadid's spray-on dress was
25:06like a new kind of concept within
25:09fashion where people are like wow we
25:11have this new technology that
25:12fundamentally changes like what we can
25:14create and so Sinead maybe I'll start
25:16with you there but any ideas on how this
25:18really again doesn't just replicate what
25:20we've done before but maybe unlocks a
25:23whole new wave of creativity on the one
25:26hand it puts design tools in the hands
25:28of people who wouldn't have been able to
25:30step into the market and so what does it
25:33look like to have an AI system that's
25:35been kind of refined on a certain style
25:37that you like that you're able to kind
25:39of tweak and then digitally apply those
25:41images to your body or your photograph
25:46um maybe you can't really afford the the
25:48high brand luxury fashion but you can
25:51generate your own version of luxury
25:53using these systems and digitally apply
25:56the clothes to you afterwards and I know
25:58that the digital application of clothes
26:00has been something that's been seeming
26:02to be in the pipelines the last few
26:03years but now it really is possible to
26:06retrospectively apply clothes to to
26:08images and if you look at what the
26:11internet and social media did to Fashion
26:13we saw the birth of all of these small
26:15brands that finally had a shot in some
26:18ways it actually became really cool to
26:19be the Undiscovered brand that suddenly
26:21kind of you know was discovered and
26:23popped off by a certain celebrity
26:25figuring them out imagine what happens
26:27when more of us get access to to the
26:30these design tools fashion is really a
26:32is really a communicator of culture and
26:34now we can push the bounds on that
26:36almost infinitely where we conduct photo
26:39shoots who gets to be in them when we go
26:42digital things change a lot of course
26:44again you know there are ethical
26:46challenges that come with that even
26:48things like copyright am I allowed to be
26:50inspired by Celine for example have an
26:53AI system look at the latest collection
26:55and then say put me in clothes that look
26:58like this the latest Celine you know
27:00suit and shoes but it's not Celine where
27:03does the IP go there and you know are
27:05there real designers on the line who may
27:08see knockoffs at a whole other scale
27:11than we already do so we do have to
27:13figure out what are the boundaries of
27:15this technology where does IP start and
27:17stop and a world where we're a lot more
27:19digitized but I think the the what could
27:21be possible and who gets to be a
27:23designer I think could be really
27:25interesting definitely Danny what about
27:27you anything come to mind in terms of
27:28how this technology algae maybe unlocks
27:31again like a new wave of creativity or
27:34New Opportunities Photoshop shaped up
27:37this whole industry where suddenly you
27:38could make photos and you're adjusting
27:40it afterwards so you would make a photo
27:42and Photoshop was the the
27:43post-production I wonder how AI can do
27:47something and I think it's going to go
27:48there how AI can be the pre-production
27:51whereas you generate the photos you can
27:54do the real photo shoots to solve for
27:58missing missing content that you cannot
28:00make with the AI and then have Photoshop
28:03generation and post generation together
28:05for example what I would think for
28:07headshot Pro is like the customers who
28:09get the best output from me did a photo
28:13very basic photo shoot on a white screen
28:15trained the AI on the photos and then I
28:17could generate hundreds and hundreds
28:19photos of them in a park in an office
28:20wherever they want to be so I wonder if
28:23headshot photographers are gonna
28:24transition towards generating really
28:27really high quality training images for
28:29example and then use AI tools like the
28:31next Photoshop to put their customers
28:33without having them to go outside having
28:35to go whatever make the perfect photo
28:44so instead of just looking at the
28:46campaign you there might be an
28:47interactive option where you could drop
28:49in your own virtual twin to see how you
28:52would personally look in that image and
28:54I think that that could be really really
28:56interesting and maybe be even more
28:58accurate in terms of how things would
29:00likely fit on your body if we could get
29:02the physics of clothes right online and
29:05I wouldn't be surprised if if any social
29:07media is going to do this soon because
29:08Instagram already has photos of your
29:10face right they could already train a
29:13on your on your face in that kind of
29:14sense and you could put the clothing on
29:16it that they put in their shopping
29:17Carousel for example so I think social
29:19media platforms are going to integrate
29:21this in a few years or within the year
29:23for example this is this is going to be
29:24reality yeah I mean I think if anything
29:26will at least learn a lot about our own
29:28preferences because I don't know about
29:30you I don't want to see I don't want to
29:33see my face online everywhere but some
29:34people might but also you might learn
29:36about what you do want to see what kind
29:38of person do I want to see modeling in
29:40front of me online right now I don't
29:42have that option to decide but as we do
29:45get more of these options and and I
29:47think what both of you are pointing
29:48towards is like we have with these tools
29:51eventually more say and it'll be
29:53interesting to see how people utilize
29:55that say right how they want to
29:58participate in that ecosystem and I
29:59think we'll learn a lot both
30:00individually but also as a society about
30:02what we want to see because previously
30:05there weren't so many people involved in
30:07making that decision one place that
30:09would be great to end off is I mean both
30:11of you have voiced how quickly this is
30:12moved moving right in your respective
30:14spaces and by the way respective spaces
30:16could mean Sinead you as a model but
30:18also as a futurist Danny that could mean
30:20you as an indie hacker it could mean a
30:22developer as a nomad as this evolves in
30:25many dimensions what are you doing to
30:27quote unquote keep up and you know you
30:29can interpret that as you'd like I spend
30:32a significant portion of my day buried
30:34in academic white papers to see what is
30:37likely coming down the line in the next
30:39few years and kind of plotting out those
30:41those Trend graphs and those trend lines
30:46feel a little bit of a responsibility to
30:48bring other people along with me and to
30:50make sure that these conversations are
30:52accessible and digestible to everyone I
30:55think we all have a right to to learn
30:57about and try to shape our own Futures
30:58and it becomes really challenging it's
31:00only the same few people invited to the
31:02conversation all of the time for me some
31:04completely re-learning how to program
31:06different languages I'm learning python
31:08learning all the things to develop it
31:10also for me I'm dabbling in all the the
31:12white papers I don't have academic
31:13background so I'm using chat GPT a lot
31:15to summarize those things and
31:17understanding it mostly what I enjoy the
31:19most is most of AI and machine learning
31:21it said it's it's in academic papers
31:25totally un understandable for for yeah
31:28the regular people especially me too so
31:30I like to try things out make tools out
31:33of it so it's more comprehensible for
31:35the Regular Joe my following on Twitter
31:38has grown a lot by just showing what is
31:40possible in those kind of sense it
31:42really is what I enjoy doing so very
31:44little sleep and a lot of building and
31:47well I think your Twitter following has
31:49grown because like you said there's just
31:51so much appetite for people to follow
31:53along and try to understand what is
31:55happening and I'm also just curious
31:57because you work quite early to the
32:00stage here especially with profile
32:01picture.ai it's kind of crazy to think
32:04about that was not that long ago a few
32:05months ago six months ago October I
32:08think it's not even it's not even been a
32:10half a year since all this Tech is out
32:12simple yeah and it feels like forever
32:15ago that seems like a perfect example of
32:17where you were early but then to your
32:19point it moved so quickly this world
32:21knew so quickly that ultimately lenza
32:23came in lenza I think ended up you know
32:26if you were to declare a winner not like
32:28it actually works that way then lenza
32:31took a majority of that market share and
32:33I think a lot of people also have the
32:35question as they debate whether to build
32:37with these AI tools is well if it's all
32:40based on the same model
32:41then you know ultimately do I have
32:44emotes do I have anything that actually
32:46will stand the test of time no I got a I
32:49got a really good wake-up call because
32:50every Indie haircut like we like to
32:52develop right we like to build things we
32:54don't do marketing and seeing lensa
32:56swoop in one month after we actually
32:59launched it already and just did our
33:00Revenue in one hour or something like
33:03it's a wake-up call for a developer you
33:04got to do distribution is everything
33:06marketing is everything it's really
33:07important it's always been important
33:09with product launches and especially
33:10with AI where you have less of a modes
33:12hiring Tick Tock influencers letting it
33:15post it and it's going really fast due
33:17to it but I think someone is going to
33:18come out who has a way bigger budget for
33:20that and then in the sense like you
33:22shouldn't build on a hype so what I'm
33:24focusing on mostly now is SEO building
33:26tools getting backlinks
33:29um trying to get it to stand the time
33:31longer than just a hype focus on like
33:34something small some small problems to
33:37solve and do your marketing
33:40to the distribution don't just build you
33:42got to sell it what might people not
33:45realize is surprisingly hard and so
33:48maybe one simple example of that is as
33:50you're building these models everyone
33:52jokes about the hands I think that's
33:53being solved but I think there's
33:55probably infinitely more things that
33:58you've encountered as you're building
33:59these projects that like again people
34:01see the end result and they're like
34:04Alice looks amazing or like deep
34:06agencies working for me or what a great
34:08headshot but what might people not
34:10realize if they haven't built these
34:11tools is actually again really really
34:15so might look easy on the inside like on
34:17the outside okay you upload some photos
34:19in 20 minutes later the photos are done
34:21but actually so profile picture was just
34:23dream move it was there was no mode on
34:25it and what I'm doing with head Shield
34:27is basically it's 15 different AI models
34:29stacked on top of each other in like a
34:32sequence of doing things together like
34:34it's a modern like they're custom models
34:36built right now so you said if you want
34:39to stand that yeah I think it's at the
34:41moment it's 15 different AI models doing
34:44things on top of each other to optimize
34:46the photo quality to optimize the
34:48training upscaling generating yeah if
34:51you want to go more unique yeah I didn't
34:54know python I didn't know machine
34:55learning I had to learn all those kind
34:57of things so I could build my own models
35:00um what surprisingly easy these days is
35:03and what used to be hard is deploying
35:06running your gpus is used to be really
35:09hard but there's all these startups
35:10springing up since I think October also
35:13since the hype they wanted to capitalize
35:14on okay AI is hot what is Hearts is
35:18hosting the model so that part got
35:20released these times so it's easier than
35:23ever to start at the moment and just for
35:25those who maybe are at the very earliest
35:28stages where they're like okay I want to
35:31do this I want to play around where
35:33would you start like how do you learn
35:35how to train your own models how do you
35:37learn to deploy them how do you learn to
35:38interface with something like stable
35:40diffusion so the easiest way would be to
35:43go to replicate.com they host all the
35:46models you can just select what you want
35:47you could select a stable diffusion for
35:49example and they have a few simple input
35:51files Fields where you can just type in
35:53what you want and that outputs the photo
35:54they make it super easy they have an API
35:57that you could hook up to a no code tool
35:59for example if if you don't want to
36:01build it in no code I believe shapir
36:02loans their stable diffusion integration
36:05so you just plug that into an email or
36:09Discord or whatever so those two are
36:12released to get started with by making
36:14some simple simple AI tools
36:17yeah and if you want to go deeper you
36:19probably need some coding knowledge but
36:21I think 99 of the apps you can build
36:23right now with API wrappers with no code
36:25tools that's super cool because I think
36:28a lot of people who don't know how to
36:29code can actually participate in that
36:31ecosystem we kind of veered from the
36:34original conversation but I guess just
36:35closing off because you have built all
36:38of these projects how do you think about
36:40what to focus on now moving forward I
36:42mean this ties into our conversation
36:44around like what really has a moat what
36:46people are really willing to pay for
36:48especially because you are an indie
36:51hacker and you do need sleep eventually
36:53I wouldn't dabble in any image
36:57generation in that sense because Adobe
36:59already does it Comfort does it already
37:01I wouldn't build any chat Bots because
37:03chat GPT they are shipping I don't even
37:05know how they're doing it but they're
37:07launching something new every every week
37:08well hint Danny they have more than one
37:10person like you fair enough but just
37:13shift like a startup which is exciting
37:14to see because they're competing with
37:16Google and all the likes right no but I
37:17would just pick pick something and ID
37:19something that issues you that annoys
37:22you or you think hey maybe if I put some
37:25AI on top of this it could speed it up
37:27we're going into a recession which means
37:29people want to save money so can you
37:32save money by removing some kind of Step
37:33totally and if anyone wants an idea
37:35Danny would love if you want to jump on
37:37this but a very simple thing that I
37:39still have not seen is the ability to
37:41use these text image tools but in a way
37:44where you can basically train it on your
37:46own brand right so you upload your brand
37:48colors if you have you know let's say a
37:51Blog with a bunch of sharing images from
37:53the past you trained on that and I know
37:55people can set up their own models to do
37:57this but if there was a self-serve
37:59version where any brand today can again
38:01upload the constraints and then spit out
38:06let me know if it exists that you've
38:07seen it that to my knowledge does not
38:09exist in that exact form I'm gonna 100
38:11bet you that count for dusties in the
38:13next three months okay you've been
38:15following yeah okay maybe yeah yeah
38:18we'll see if canva does that but um for
38:20me if I if I if I can tune in an idea I
38:22would love to see I don't know if we
38:24have time yes I would love to have so
38:27openly I launched a a model called
38:30whisper which basically lets you
38:33and all these Voice Text to Voice models
38:37are getting really really creepily real
38:39I would love to have someone build a
38:41tool where I say hey I want the top five
38:43posts of hack news while I was sleeping
38:45the latest news and my favorite people
38:47on Twitter and generate me a 100
38:50customized podcast I can at least in
38:53five minutes while I've walked with my
38:54dog in the morning customize it for me I
38:56think that will be really awesome and
38:58not even that hard to build I guess no I
39:01agree and actually I've been seeing some
39:03folks like there's this guy who's been
39:04creating the AI version of the all in
39:08um and he's he just released the fifth
39:10one but you can actually see how with
39:13every subsequent one that he releases it
39:15becomes less popular because at the end
39:17of the day that novelty factor is
39:19wearing off however to your idea you're
39:22actually integrating that with as we've
39:24discussed like a problem that you have
39:25which is trying to digest a specific set
39:28of information which is the stuff
39:30ranking on Hacker News and actually
39:32digesting that so that's like a very
39:33clear your job to be done versus a
39:36novelty of just like let's listen to
39:38these fake people talk about fake topics
39:41yeah don't don't build on a hype because
39:43you will get some attention on social
39:45media and then it fades away for example
39:47yeah for some reason this reminded me of
39:51um I knew this guy who used to listen to
39:53Spanish as he slept and he was convinced
39:55that when he woke up he was better at
39:58speaking Spanish he was trying to learn
40:00the language did it work he thought so I
40:03don't know if it did but anyway on that
40:06note it's super cool to be watching all
40:09the things that you've built I think
40:11many people would agree with that just
40:13like the the rate of shipping is insane
40:15I'm sure a lot of people are inspired to
40:17go create similar things and
40:19um yeah it's exciting to see what's
40:21being built 100 agree thank you so much
40:23for having me on the podcast really an
40:25honor of course well thanks for joining
40:29thanks for listening to the a16z podcast
40:31if you like this episode don't forget to
40:33subscribe here on YouTube to get our
40:35exclusive video content we'll see you