Generative AI: Practical Use Cases for Fashion, Beauty and Luxury Industries | PhotoVogue Festival 2023: What Makes Us Human? Image in the Age of A.I.

Hi everybody.

Hello, thank you.

Thank you for coming.

So, we re gonna spend the next 20, 30 minutes

taking you through a number of the use cases

that we ve created using generative AI

and how we ve used them in the area of fashion,

beauty, luxury, and much broader.

Clicker s not working. Let s see.

The clicker s not working.

Hello?

The clicker s not working.

Still not working.

We have it.

Thank you.

All right, thank you.

So, Maison Meta is probably the world s leading AI studio.

We started about a year and a half ago,

experimented a lot with various AI tools,

and over the last year,

we ve worked with a number of clients

from Moncler to Zara, L Oreal, H&M, Victoria s Secret.

We also advise companies on how they can use Gen AI better.

Our clients there are the likes of Google, UBS,

Norma Kamali, and Mango, and as they mentioned,

we re also founders of AI Fashion Week.

There is a preview of AI Fashion Week number two,

season two, around the corner.

It ll be continued in New York end of the month,

and then there s the voting, and the winners

will be announced before the end of the year.

So, some other little things, we re recognized,

actually, Cyril was recognized

as Vogue Business Top 100 Innovators.

We re one of only two people working AI.

At a number of conferences, we speak at McKinsey,

Business of Fashion, UBS, and participate in our reports

and other articles on the subject.

Let s get into the meat of it.

So, none of what you re gonna see from here exists.

Everything has been created using generative AI.

So, our first project was with Moncler.

Maybe, do you wanna talk about this one, Cyril?

Yeah.

So, for this one, we started to work with Moncler Genius,

where they ask us to basically recreate their collection

to present into the press.

They did the presentation on Vogue Runway,

and as well as the runway in London in February.

They ask us to augment and use AI tools

to represent their collection,

but push them to more bubbly, to be more fun,

to play with the avatars.

And it got a great success, very viral,

and we were very happy with the result.

These are some of them.

This one is with Pharrell.

This one is with Palm Angels.

This one is with Adidas Originals.

That was their big collaboration collection.

And every time we do these sorts of projects,

there s a lot of creativity

that goes into the creation of the imagery,

but we also work with our clients

to make sure that they get some business benefits.

So, a campaign like this, from beginning to end,

took only three weeks,

where traditionally it would have taken

maybe three months, four months,

and at a fraction of the cost of the client

doing it in this way.

So, we re always looking at ways

in which the brands that we work with

benefit from the work that we do.

So, this project was us working directly

with Stephen Mizell,

and it was around his collaboration with Zara.

So, in this occasion,

Stephen wanted to create something

that is very exacting.

So, exact replica of the clothes

that he co-designed with Zara,

and also, facially,

it was supposed to be somebody who represented him

when he was much younger.

Yeah.

Yeah, it was very interesting to do this campaign

because he wanted us to push the AI

to some level where the upscaling was very hard,

like, for example, the grain,

the details of the leather,

and also, he had to be obviously shot in his style

with his type of lighting, coloring,

and then after this,

we went into retouching and make it perfect.

But when you zoom in on the original files,

you can see every single hair.

We really pushed the AI

and worked in patchwork techniques

to make it to this quality.

So, I m actually curious,

how many people here have used

a generative AI platform or a tool?

Like, MidJourney, Stable Diffusion,

anyone?

DALL-E? Yeah, a little bit.

So, one of the things that, I guess,

makes what we do unique

is the ability to actually recreate clothing or objects

exactly as they are in the real world,

but using Gen AI.

So, if you go to Mid Journey

and you try to create a red dress,

it s very hard for that red dress to be created

in exactly the way you want it.

And certainly, brands that we work with

want their product represented in the images that we create.

So, we ve, I guess, created a workflow

that allows us to actually recreate the clothes

or the objects from our clients brands

and place them on people or in different situations.

So, this jacket, for example,

the button on the sleeve, the materiality.

The seams. The seams,

everything is probably within 95%

of what you will see online or in the store,

but it s all been created using Gen AI.

And that s what makes the work that we do

a lot more complex than otherwise.

So, these were some other examples of that.

So, Revolve, we ve been working with for a long time.

We have a number of different projects

and partnerships with them.

So, this is an example of a scalable campaign.

So, another benefit that using AI brings to projects.

So, this project was to celebrate 20 years of Revolve

and as part of their Coachella activation and celebration.

So, it all started with the images on top left,

which was created for social media.

And then they decided that they wanted to change the use

and use them for out-of-home advertising posters.

And then eventually they asked

us to create these billboards.

So, these were the first billboards globally

that were created using Generative AI.

So, they re very difficult to do

because you have to upscale them.

They re about 20 meters by about 15 meters

or 12 meters high.

And what Revolve was able to do with the time

and the budget that they saved in creating these,

they were able to actually reinvest those

in buying dozens of billboards all around California

to actually then promote the campaign

or promote the celebrations.

Do you wanna talk to this one a little bit?

So, on the other side of the campaign,

they really liked the clothes in the jeans.

So, they asked us to upscale them

and then they created some version of them to sell them.

So, they were also like the first with us

to kind of like take some AI imagery

and then bring them back to the real world

and create a full collection from them.

So, they were doing it as a test on a small run

where they wanted to see how from imagery, purely AI,

coming back to the real world and creating the collection.

And it worked pretty well and it got sold out.

They did like a small run.

And then, yeah, we had some great press on it

with business of fashion, of our business as well.

I mean, the scalability here,

imagine if you were doing this

with a traditional means of photography

or working with any traditional type of studio.

Every time you wanna have different usage,

you have to go back, get different rights,

maybe reshoot things.

With this process, we started from beginning to end

and we were able to take it from online to out of home

to billboards all the way to production

all in one fell swoop.

So, in terms of brands being able to use generative AI

to really make work more efficient,

this is a very, very good use case.

So, H&M Home.

So, H&M is going through a lot of changes.

They re doing a global rebrand.

And what they wanted to do was to bring a lot of creativity

back to their headquarters in Sweden.

With that, they also wanted to make sure

that they reduce their carbon footprint.

So, rather than them flying teams

to different parts of the world to do shoots in location,

we worked with them to do these tests

as a way for them to think about

how they would do shoots in the future.

So, again, everything that you see here

was generated using Gen AI.

The candles are exact replica

of what you would find in a store,

which is a very difficult process to do.

And then, you can see the lighting,

the translucency, shadows,

everything that you would get from traditional photography,

you can get in this way too.

[Cyril] I mean, for this one was like the pillows

and yeah, I mean.

The materiality, the fluffiness,

all of this can be produced as part of the process.

And then, what they can do is actually

create these objects using AI

and then place them in any environment

that s relevant to the audience.

So, maybe this would be more for the British audience

and then the Japanese audience would have these objects

in a different setting.

So, that allows them a lot of fluidity

in terms of creating content.

So, this is actually an example of localization

or personalization with Coach.

So, the tabby bag here,

I think it s probably 70% of their revenue

from non-fashion or non-apparel sales.

So, what we did here was create an AI image, if you like,

like an AI catalog that created the bags

in different materials and so on.

And then, we re able to place them

on different personalities in different environments.

So, for example, they could use the content on the left

to appeal to a working lady living in New York

all the way to the right,

will be a different audience for them

that they can do that with.

And all of this, again, is done much more quickly,

much more efficiently than a traditional photo shoot

would have you do it.

Maybe you wanna talk to the materialist?

[Cyril] I mean, here we were showing them a little bit

like the texture and the materiality

we can get from clothes.

So, they can put this for their research as well

and show how it looks from a macro side,

like those two leathers.

And this is like a weaving.

We can recreate almost any texture, any color,

any materiality very, very quickly.

So, this was interesting project we did with Unilever.

So, we didn t know this, but they told us that

when they have beauty products, creams and so on,

they want to show their customers or their consumers

how the product feels when you put it on,

how does it feel in a sort of watered environment.

And so, traditionally what they would do is

they would have a bath or a plate

and they would try to recreate bubbles

in exactly the shape, exactly the translucency,

exactly the size, and it would take days

for them to do that.

You can see here other examples.

So, they came to us with this sort of challenge of

could you use AI to create textures and materiality

of creams and other beauty products

that we can use to communicate

how our product feels on a skin.

And so, this was all done in a matter of a couple of weeks

with them being able to choose the exact shape of bubbles,

the exact translucency, the exact size,

something that, again, would have taken them

many, many days in a studio waiting

for that exact shape of a bubble to appear.

Do you want to talk to this one, Cyril?

Yeah, sure.

So, for Mugler, it s about artificial creativity

where they ask us, they gave us like a brief,

like the same as like a photographer,

and saying what if we will relaunch

and rebrand Mugler angel perfume.

So, they wanted to have some example of the actual bottles,

keeping the essence of the angel bottles,

but also giving some new shapes and new styles.

And they also ask us for more like transgender fluid

type of models and then futuristic environment.

So, they wanted to have this pretty much as like a preview

before creating the shoot.

So, it s not necessarily going to be the campaign,

but with this, they can see quickly how the final image

or how we can look,

and then they can pick their environment,

creating their set afterwards and picking their models

and envision the entire campaign.

We don t see AI replacing traditional photography

or traditional shoots or traditional campaigns,

but in this occasion, AI can really help those teams

to be better prepared for the day when the shoot comes.

So, everybody s on the same page,

understands what the final outcome needs to be,

and that makes things happen a little bit quicker

and a bit more cost efficiently than would be otherwise.

So, this was another example.

As you can see, the bottles in each situation are different

because they wanted to test with different shapes

of bottles as well.

This was a campaign we did with Pangaia.

With any sort of movement in new technology,

we saw that with Web3 and sort of the gamified look,

there s a certain aesthetic that comes forward.

And so, we leverage that aesthetic of this oversized,

exaggerated, almost like impossible environments

using generative AI to create their summer campaign.

So, these are all the things

that maybe are not really possible to do in the real world

or very difficult to do.

And with GenAI, you have this sort of a very different

approach and ability to really push the boundaries

of what s possible, but still

make them feel hyper-realistic.

Yeah, and for this also, they really wanted to have

their exact product placed in, obviously,

like the bug, the t-shirt, the hoodies, everything,

like the coloring, everything had to be exact,

but we could play with AI and kind of like exaggerate

and put like big flowers or very fancy.

So, for this one, it was for the jewelry brand Pandora.

They wanted us to create some environment

for their jewelry.

So, this one, we mixed two technique.

We worked with the 3D assets for each of the jewelry,

and then we created, recreated the background.

So, this one was like paper craft style.

This one was for the graduation, love and ribbon.

So, they wanted to like have some very fun backgrounds

and then we mix it with their 3D pieces

to do the incorporation and have this as a final project.

And they really like it and they haven t launched it yet.

So, for now, it s not published yet,

but probably very soon now.

Yeah, so for this one, we work with Victoria s Secret.

That was a test where they wanted to see,

they wanted to do a Barbie campaign.

So, they wanted to see like how can we make those.

So, first we recreated the logo, the VS logo,

the same as Barbie and all this.

It looks like 3D, but it s all made with AI.

We were able to use like those Control Net tools

and all this to get like the Barbie look style.

And then for the model, so the models,

they wanted to have like a specific type of models,

a specific type of body type,

and the closing obviously will have to be perfect as well.

So, we had to create mini AI models called Loras

to create those and place them afterwards.

And like here, they wanted to have bubble,

but they also didn t want it to look like too real.

They wanted to keep still like a feel of like CGI

and a little bit of 3D.

So, we have some clients that ask us like,

we don t wanna see that it s 3D

and some they want to see that it s 3D.

So, for this one and the one after,

we kind of like have the skin a little bit more flat

and not as much texture.

And we can feel that it s not like real right away.

And this really goes to underline or show the flexibility

that these tools bring to the table,

to go from here to here.

So, the amount of creativity that you can bring

to any project is really amplified

by the way that you can use the tools.

And then this was a test that we ve done with them.

We continue to work with them so that they can do,

if you like the shoots that they would normally do

for their fragrance and beauty products.

Again, using AI to really sort of push the boundaries

of what s possible with the tool.

I don t know if there s anything else.

[Cyril] Yeah, no, no, that s that, yeah.

So, this is one of our favorite, favorite projects,

which was actually delivered only about two weeks ago.

So, I don t know if many of you in Europe

are familiar with Norma Kamali.

She s one of the most iconic designers, American designers,

who s been designing for 40, 50 plus years.

She s 78 right now.

And she came to us and said,

I want you guys to create a AI tool

that can design with the aesthetic of my brand,

with the DNA of my brand,

not only to help me and my team in the design process,

but also once I m gone, I want the next designer

to continue to design my collections

with sort of my aesthetic, with my DNA built into it.

So, with this, maybe...

So, for this project, we started basically

to work with her data set.

So, she gave us like about 15,000 imagery

for her swimwear collection.

We started with the swimwear,

where we then trained a model based on her keywords,

on her specific type of cuts, fabric,

all this had to be like tagged per image.

So, it s quite a bit of work.

And then after that, we train a whole model

on all her imagery, all her keywords.

And then after we just play with prompt

and then the model is extremely flexible.

So, we can go from this type of drawing.

And this after is generated in a matter of

20 seconds per image.

Now, we just put like, I want a tech spec design

of this type of swimwear.

And then they just come up like right away.

And then after that, we can also push it

to see like more editorial or see it like in real.

And so, the model really learn her style and her keyword

and can come up with any type of style,

any type of imagery based on her DNA, on her style.

And she s extremely happy right now.

She s playing with it and she s planning to use it

to help her design her next swimwear line.

So, after that, so that s like another type,

like more like American comic drawings.

We had, yeah.

So, we had a lot of fun doing this.

It was like a first time in two months full work

to create that.

And then she s extremely happy with the result

and what it can bring.

And again, imagine going from designing a collection

with just keywords, with some prompting,

with some references, all the way to this sort of thing,

literally in a matter of minutes.

Well, and by the way,

these are not retouched Photoshop at all.

They re just coming raw from the AI.

So, it takes 20 seconds to do,

to come out with just the proper prompt.

The hands, sometimes like this one, we can see,

it s not like the details,

but these are like very easy fix with AI.

You can in-paint it or even with Photoshop.

But here, we mostly concentrated on the closing

and on the style of each of her swimwear line.

And the result, we really like blown away.

Like, yeah, or any type of drawing.

So, this is American, we can go a manga,

we can go technical drawings, everything.

And yeah.

So, everything that you ve seen,

I think we mentioned that it s very difficult to do.

There s a custom workflow that we ve developed

over the last year and a half

to allow us to bring so much flexibility to the Gen AI

and also be able to recreate exact products,

which is very difficult to do

if you just go to a MidJourney, for example.

So, now we re working to develop a platform to,

almost like a B2B SaaS platform, a software platform,

to allow companies, brands,

to do the work that we do manually,

but actually take that in-house

and be able to do that for themselves.

Yeah, so a lot of brands come to us.

Can you come back?

A lot of brands come to us and ask us to do it,

and we use a lot of AI tools,

and it s not really easy to install on your computers.

First of all, on Mac, most of the creative works on Mac,

and on this, you need a PC, you need powerful PC,

you need to code a little bit, the installation.

It s still a little bit hard.

So, for this one, we break it in four,

where we wanted them, one, to be able to create

with custom models,

enhance, where they can then rework the detail, all that,

and then in-painting and painting,

so they can fix the hands, the hair,

change details on each imagery, and train.

And train is an important piece,

and we can see it on the next one,

where basically, you ll be able,

with a very small dataset, like five to 15 images,

to put it and then add some keywords to it,

and then press train.

Either it s a person, an object, or a style.

Then you press train, and then it will train for you,

and then you will have your keywords,

and you will then be able to generate new AI imagery

with just your keywords.

So, we ve been working on that.

There s nothing yet that is available on the market,

like that, and hopefully should come out very soon, yeah.

Yeah.

Maybe you should talk to this one, too.

Yeah, so, this one, same.

Because we ve been working with a lot of fashion imagery,

we created a lot of fashion datasets for the last year,

and then we worked with photographers,

and we created a very big dataset of imagery,

and we wanted to be able to create AI fashion models

that can do any type of imagery,

from runway, to editorial, to all this.

And we were able to fine-tune over,

same, similar than normal,

about 20,000 images.

And from that, we now have a model

that is extremely flexible,

where we can ask any type of knitwear, or cuts, or details.

These were knitwear prints, knitwear prompts.

But then, we played a little bit on the runway.

We asked detail on the inside of a volcano.

All those details, like the tweed skirt,

and all these were added to the prompt.

So, really, the words and everything we put inside

is just coming out.

Yeah, like this one, with a huge wave on the background,

like this one, with an oversized moon.

And all this comes from the prompt,

and have very good flexibility,

compared to tools like MidJourney,

where it s still a little bit hard

to get anything that you want.

And part of our goal is also

to make this available to the public.

We re speaking with Parsons and FIT in the US,

to make these sort of AI models available to the students,

so they can actually train and be able to use AI

as part of the work, as part of the study they do.

We re gonna keep on adding to it, fine-tuning it,

and really to have the reference of AI and fashion model

that people can use really easily,

and create imagery.

So, the last thing we re gonna talk about

is AI Fashion Week, which is sort of what brought us here.

This is season two.

We did the first one in April of this year,

back in New York.

There s a few objectives behind AI Fashion Week,

and I hope you ll have a chance to look at the preview

of season two around the corner after these next few talks.

One of the things we wanted to do

was create a global platform where creators

or creatives using AI can showcase their work.

There really doesn t exist that sort of environment

right now, at least not for the fashion space.

So, for the first season, we had over 400 submissions

from literally around the world,

in every continent.

I think it was over 50 countries

that had submitted their collections.

So, we showcased them online.

We showcased them at an exhibition in New York over two days

and then made the qualifying collections available

for people to vote on.

The same happens here.

So, in two weeks, we had over 14,000 people

register on our app and vote for their favorite collections.

So, this was the top 10.

The top 10 winners are here.

We can see here we have...

Where is Patif?

This one.

I think, actually, they re here.

So, if you guys want to talk to them,

they re also here.

We have Patif, Jose Sobral from Portugal.

We have Mathilde Marino from Portugal as well, number two,

and then Ope from New York.

And then all those three win from the jury.

And then we recreated their collection with Revolve

to actually launch live last week on revolve.com.

So, for them to enter this fashion incubator,

being able to launch their collection,

and then after have like a brand for two seasons,

and then after they re open to go and then keep on going,

keep on selling, keep on doing collections and all this.

So, really helping them from AI fashion designers

to real fashion designers.

You know, here what we wanted to show was the full cycle

that you can actually use generative AI

to create a collection that s produced,

that can actually be sold and worn by people.

And, you know, we want to support these people

to be able to launch their brand.

You know, one of the very difficult things,

at least in the fashion industry,

because it s so antiquated,

is that, you know, you can be a great designer,

but you don t really know how to work with supply chain

or source materials or deal with building a website

and doing marketing and customer service and returns.

So, that s why we partnered with Revolve,

which is a $1.4 billion e-commerce and fashion tech company,

mostly in the US,

but they obviously operate here in Europe as well.

And as part of the incubator,

the designers worked with Revolve and their team

to be able to do all the things that they need to do

to actually launch a brand.

Again, sourcing materials, producing samples,

doing the production, all the shipping,

all the customer service.

So, this way they re actually able to focus on their craft

rather than worry about all the administrative

or logistical things that happen.

Yeah, here we can see that was like his AI imagery

and this one was like his like custom-made top and skirt.

And most of the images and the collection

looks similar to the AI

because some of them were like really hard to reproduce.

So, it s been like redesigned a little bit.

And we re showing a preview,

there s like three pieces in the main room as well.

And then we ll do the whole presentation

in New York end of the month.

Yeah, and then this time it will be the same,

that all the submissions to AI Fashion Week

will go through the same process, public voting.

If anybody wants to go online,

you can go to app.fashionweek.ai.

There s some barcodes or QR codes outside as well.

You can view over 150 qualifying collections.

Yeah, we have I think 168 for now, yeah.

168 qualifying collections.

It actually all opened today.

So, today s the first time that anybody can see

all the qualifying collections.

People can vote on that.

And then we have an unbelievable jury this year

who will narrow down the top 20 to the winners

who will again go through the same process with Revolve

and be able to produce their collections

and sell them online.

This year for the jury, we have Pat McGrath

who s a world famous makeup artist.

We have Matthew Drinkwater

who s the head of the innovation agency

at London College of Fashion.

We have a fellow Italian, Ferdinando Verderi

who s a creative director

and was the creative director at Italian Vogue and Prada.

There is Tiffany Godoy from Vogue Japan.

Matthew Drinkwater.

I mean, we have like really amazing jury.

We actually showing them as well

on the panel introduction outside.

But from the competition,

we had quite a bit of Italian as well.

We had about, I would say the first year,

we had maybe 20 and this year about like 40 designers

about like from Italy.

I mean, they re not all from fashion,

but they re just playing with all those AI tools

and they create their whole collection and yeah.

So yeah, these are real imagery of the actual collection

that was shot two weeks ago in LA from O.P.,

one of the finalists, one of the winner.

Well, thank you.

And with that, we ll finish that.

Thank you so much for your time.

[crowd applauds]

Are there questions?