How MVRDV is using AI to design their buildings

okay so look at this project done by mvrb this is an image of the base 3D model and these are all variations that were generated using AI each one slightly different than the other some influence materiality others change spatial design ideas and others just certain parts of the design now mvrdv is probably one of the most interesting architecture offices in the star spectrum and this is because since the 90s they've been not only designing projects that are different and to be honest kind of crazy but because they're also constantly investigating and researching on the side

from projects centered on roofs to City Planning and even Furniture they're committed to a research based design practice mdv is a firm that does design through research not by testing a million things or by making a nice shape it's always by uh creating a hypothesis testing it figuring out if we agree with it do we like it do we not and we through design research then gets to a final conclusion this is KAS esbach an architect and project leader at mgtb KAS has been involved in numerous projects and is constantly emphasizing how the office's success

is centered around one main ingredient embracing change in Innovation 's tradition is to embrace uh this growth and embrace this ever change so it's the our tradition is to change and to accept these things and to go full force into them but with AI entering the architectural realm how will this tool reshape the way mbrv approaches design will it enhance their creative process or make it dull and boring this is a story of how one of the most avanguard architecture firms in the world steps into the future with AI as their companion in design okay

so now architecture is always changing and evolving and the tools which create our future buildings also change and evolve so when a new technology like AI comes along and Promises to be a super tool and enhance the architecture design it's a no-brainer to accept it and use it to push architecture even further but one of the first questions that arises with a new tool is almost always why should we choose a new tool when our current one works perfectly if you can't do it in the traditional way you shouldn't do it through AI so so

everything can be done in the traditional way but AI simply increases the speed of it and makes it more efficient and faster this is Freddy Fork he is an architect with a masters in Building Technology from TU Del and AI specialist inside of mvrdv so more or less he has been at the center of this transformation of architecture and AI so I I really enjoyed that part where they push the boundaries and this is what what I'm doing here trying to push the boundaries of applying AI to architecture here at a VDB AI is transforming

the way we think about architecture design it's not just about the automation of task it's about enhancing our creative process AI allows us to experiment at a pace and scale that was previously unimaginable the AI is is not a magic wand but it's a power tool it just simply helps us uh get results much faster and each new tool throughout the years has brought something unique to the table enhancing our ability to visualize and realize our designs but with AI we're not just looking at another logical step in this Evolution we're witnessing a leap a

long time ago back in AutoCAD I don't know how long it took you to make a photo realistic render and how much money then slowly when fate came in okay you can do a really nice render in about 3 days now with ncape you can do a nice render in one day and now with M you can do a nice render in maybe 30 minutes now clearly for mvrdv integrating AI into their design process was a step into Uncharted Territory in order to fit a new tool into an extremely intricate architecture process they would have

to start the same way they start any of their Designs By Design through research I think about July of last year I mean the first Whispers of M AI or mid Journey version 3 was kind of going through the office and people like what what is that what's going on and then there was a lunch meeting by what by a few of the Nerds in the office where we all sat together we're like okay what are you doing with your M what have you done with it and we're showing each we're sharing these images and

then what happens is since MTV is a studi based office so we have nine different Studios within one MTV title uh we started sharing uh this knowledge so slowly the from the Nerds it kind of trickled down into the whole office until finally many people were uh challenged to work with MIT nbb's initial dive into AI wasn't a structured plan but a series of organic curiosity driven Explorations it was about shaping Knowledge Learning from each other and collectively envisioning how AI could fit into their work this group of people we brainstormed a few Studio sessions

in which we then uh grabbed smaller teams of maybe 20 people and then presented them how mid Journey works how stable diffusion works and then uh give them a design challenge so that's maybe an hour in total 30 minutes we give a little lecture 15 minutes we give them to play around with themselves and then they have to present their thing so I remember in version three of mid Journey we were asking people to generate floor plans sections elevations and renders of one fake project which we made up and then ask them to present it

to each other and then finally we had the vot on which had the most convincing design was really great s to see so as mvrv delved deeper into the world of AI they began to uncover its potential and its limitations this phase of exploration was crucial it set the foundation for how AI would be integrated into their future projects but integrating a technology as advanced and as nuanced as AI into the refined and intricate process of architecture design was not a straightforward path it involved a steep learning curve to learn new software technical hurdles and

the task of aligning ai's power tool capabilities with the firm's Creative Vision what was extremely interesting for me was actually typing mvdz what does it mean if you ask somebody with different type of bias AI has integrated into it what comes out well colorful pixels was the answer and then I thought to himself that's not what we're doing that's not what we're about at all of course another big problem was the rapid Pace at which AI technology was evolving I mean just look mid Journey launched in July of 2022 stable diffusion in August and Chad

gbt in November of that same year and some results have gone from looking like this to now looking like this AI is growing so fast like you're out for for a month you're working on something else then you come back and you're you're like oh wait what did I miss I used to be able to type this in and it used to work this struggle was not just about technical adaptation it was about a philosophical adaptation as well I mean how does a firm known for its meticulous research driven approach Embrace a techn technology that

is inherently experimental and unpredictable then you also see the limits of ai ai couldn't just knock out uh mvv design uh because it's missing all of that other stuff that we're talking about for the the research the cultural context the locals cost of programmatic requirements and it was through this exploration of trial and error that the office started using mid journey and stable diffusion and testing to see how the initial stages of a project could be enhanced with AI tools if the normal process would look something like this where you can talk to the client

sketch some ideas out pass it onto a cat software and start refining it from there AI would start enhancing this process and playing a specific role in different parts many times we use it in the first stages we sit down we talk to the clients we go back home we're like okay what what did they want and we just throw a few things in there see what comes out and maybe something that triggers us but then as we go through the process these things get refined and Def find more and we have to deal with

things like the widths of doors and stuff you know which Min Journey doesn't take into account for stable Fusion so we start to develop those and then we reach a point in the design process where we have let's say a blue foam model which then we try to Reiner into this mid Journey or stable defusion where we overlay different types of facades and kind of without too much effort create a atmospheric image which relays well to the design which adheres to all of these architectural rules the floors the doors the walls the roof so through

this constant testing and research they found that mid journey in stable diffusion could not only be useful in the initial stages of the design like brainstorming and coming up with crazy ideas but also in the more technical processes and then what is really fun is later in the process it starts to play a role again the moment that everything is defined but we have to test uh you know 20 different types of window opening sizes and you just like rock it in see what happens sometimes horrible things come out as you're probably see now on

screen or sometimes beautiful things come out which excite us and we're like okay now let's develop that further this new look into using AI to design was also having an effect on the teams that worked in each project so if before AI a team of 10 people use their time and effort to create repetitive tasks like test window openings or facade materials now they could dedicate their time to more creative parts of the design process that are more akin to what an architect is supposed to do and not what a robot or AI can do

instead of having to lose people super talented people in the productional workflow we're able to use their brain power in a more uh beneficial way for the process so actually efficiency just keeps increasing as we start to implement these tools but of course with every new technology especially one as impactful as AI comes a set of responsibilities and ethical considerations as many of you know many current AI models use information from all over with the internet to generate images so regardless of copyright ownership of credit anyone can create an original image and the ethical boundaries

of creativity are ignored or blurred lots of these models now have been trained using web scraping throughout the whole internet and it takes absolutely all images whether they're copyright or not so I do think that the proper way of advancing the future is knowing where the data and the data sets are coming from and and being able to use models that that use these uh these data sets at mvrdv we are using only images that are property of mvrdv to train the lower models that generate the design process now the advantage of creating their own

models is that it becomes a sort of personalized AI tool that can understand the office's perspective Mission and principles which is something we will see much more of in the future I think where each architecture office has a group of really creative employees but also has a set of distinguished AI language models that will separate them from the others so the the one thing I'm really worried about is that it's going to be kind of a cheap trick that people apply it doesn't take too much effort to just type in great design beautiful atmospheric light

vegetation and then applied on to just some random shape so if people start misusing and abusing then we're we're going to start getting posts on Instagram where it's like render versus reality and you just see these really beautiful over-the-top jungle Interiors or jungle buildings floating with one little column in the middle and then the reality is completely the opposite but then when I'm hearing myself say this I think to myself you know what this probably won't happen I only see a positive influence for the world so as mvrdv's Journey with AI progressed they reached a

pivotal point of integration this was the moment when AI ceased to be an external novel tool and became an intrinsic part of their design process we have three main branches that we started to use rank sty customization research and AI renderings talking about M Journey everyone's been using it and using more for this typ of brainstorming at the initial stages where you get interesting images then there is the prom styles in stable diffusion and and and there we really start typing our own specific PRS that we start to use more often and give us more

results that are closer to what we want also tools like stable diffusion allow you to grab your existing handmade models and start testing facade options or volumetric changes we were able to use these bluef models we were with a very small team apply a really quick facade on them and throw them into mid Journey with just a few key words we able to knock out beautiful images which really excited the client to a point that we were able to continue with the project The Next Step would be embedding it's also trying to to teach the

model what type of graphical uh Styles or or concepts that uh that we would like to get from the model and and finally the most interesting one is uh Lowa trading and the most complex one that we actually teach it our no concept okay so think of Lura trading like customizing your favorite video game character you have a character that's already design that's the pre-trained AI model but you want to make some changes to make it more you like adjusting the its hairstyle or outfit but not completely redesigning it from scratch Laura training does something

similar with AI models it takes an existing AI model and makes small specific changes to it this is way more efficient than creating a new character every time you want a different look it's like fine-tuning the model to do exactly what you need like teaching it to recognize certain architecture Styles or to create music in a particular genre without having to rebuild the whole model what we tried just recently and it worked out pretty well is teaching stable diffusion how to generate the valley so the the project the valley that we we worked we feed

some images tell the St diffusion model this is what the body looks like and when we use the low it will give us results as of the body and then the third branch in which mvrb is using AI is for architectural rendering where they use stable diffusion along with an additional model called control net to precisely modify the renderings add materials and change style so control net what it does is it freees the geometry and once you have KRON set up it will follow as as possible to the The Prompt that you give it but

there will always be surprises it's not a static result where in in in landscape or in Lan you really specify this is this material this you you have to be kind of open to a certain degree of surprise but that also means that you can get inspired by it so we worked in a collage based manner having very quick and dirty collages where the project leader would share with me the these collages and then we would with the very low D noising strength D noising strength is is that the other creativity that the model has

we reduce the D noising strength but then it makes the collash keep the general view of what we want to see but then really improve graphically or change the style to something more like for example Studio we or something more cartoony or so that the the client would would like the the images rather than doing it the traditional way so what was at the beginning a tool for exploration that was merely explored and talked about is slowly becoming a structural part of the design process in mvdb and the best part is that the Creative Design

through research process that they are known for is not being interrupted but augmented and improved and as a matter of fact they're already trying to see what a more integrated AI future would look like for an architectural office but I am really excited about the next steps what's going to happen with AI for example imagine an AI integrated Bim software where you can just type in to Rev hey all the the walls in the bathrooms of the more expensive Apartments uh give them this layer and it just does it so once again it's all about

removing the tedious part of the work and uh really making it giving people more time to design and research now in this new normal AI is redefining the limits of what's possible in architecture it's not just about generating quick renders or creating visually stunning designs it's about using AI to explore new ideas to push the boundaries of cre creativity and to see projects from angles previously unimagined I see that because we start to have this conversational chat type design approach AI could actually help us rehumanize the design process we start to see the images or

see our architecture from the user's perspective and because of that we could not we not only have to focus on the technical criteria but we can also see it from the user point of view because we have the ability now to to see how would look like almost instantly it it doesn't mean you can generate your building in one second and that we lose our jobs it actually just means that we can do more things and more fun things in a shorter time span leading to more exciting designs so I'm 100% Pro AI integration into

architecture as we look into the future it's clear clear that AI will continue to play a pivotal role in shaping the world of architecture driving Innovation efficiency and above all inspiring new ways to imagine and create the spaces around us our office's design ethic and Beal and thought process is so based in this research approach to design AI merely is just like a tool that you use to enhance your your process so I want to thank C and Freddy for helping me create this video I love seeing how offices like mvrdv are pushing the boundaries

and redefining what it means to be an architect and what it means to be Innovative if you would like to know more about mvv there will be a link to their website down below and also if you want us to talk more about Ai and how to use it as Architects let me know in the comments thank you guys for watching this video I'll see you in the next [Music] one