AI-Driven Design Practice – Stable Diffusion in Architecture (Can Esbach / MVRDV)

so uh welcome everyone thank you so much for joining uh it's really brilliant to see how many people have joined right now and this Zoom event uh that we have cast here with us it's almost I see the number around 400 450 and also around 200 people are watching live on YouTube and yeah we're we're super excited to have you guys in here and this is uh very exciting to see you guys again in one of our PA's accounts I see people from India Iran us UK Europe all over the world in the chat I was following and thank you so much for joining we'll have a small chat with Cass today uh Cass is joining from mvrdv and we're going to explore different uh methodologies and also different working uh workflows that he applies in his projects in mvrdv and different techniques that uh they're using Ai and uh and they're implementing in their projects so thank you so much again I welcome Cass in here hi Cass hi Amit nice nice to see you and I'm surprised to see so many people that want to come listen to what I saying from all over the world so hello to everybody as well this is a crazy and very cool thank you so much for joining uh actually this number is you see is almost like 25% % of the people who have joined almost more than 2,000 people have signed up for this event and uh we have the capacity only for 500 people in this Zoom meeting Zoom webinar let's say and uh the rest of the people will should have to watch it on on YouTube uh it's live right now and uh people are watching right now on YouTube so if you're in if any way you couldn't get into the zoom webinar please have the link uh people uh our team will send the link in the zoom chat and also the main thing that we're going to uh mention about here is the Discord channel that right now we have and around 700 people in PA Discord uh online will be following this this event as well so if any communication is needed or required it would be awesome if you can just uh start chatting and ask asking your questions right in the Discord Channel it would be awesome so because we want to keep all these communications uh recorded and available to watch even afterwards as well even if you can use YouTube or uh the Discord channel the chat section of today's uh event so thank you so much everyone I'll just have a small chat with cast then we will have a short introduction about PA event and future uh workshops then we will deliver the stage to cast to start so thanks everyone for joining welcome Cass so in in short and in summary what are we expecting to learn from you got you you today uh in this event um so um in very short I'm going to prepare six really useful pragmatic workflows and we're going to run through them I'll I'll show them off once and then I'll ask you guys guys everybody in the chat and the the Discord to join us and try it themselves give them 10 15 minutes and then reflect on it together at the end well that's great that's great so what type of tools are we focused on today like a mid journey and uh yeah so stable diffusion mid journey and a bit of Photoshop and Rhino to support that that's great and how long have you been in mvrdv and uh what are some of the key projects that you were focused on recently uh I've worked now at MV for five and a half years almost and for about three and a half years I've been project leading there and I think my most famous one one you guys have also posted on your site is the Shen and terra's project which is yes spectacular and uh have had the luxury of working on the projects like the valley and tripis and uh yeah just some really spectacular projects mhh that's that's great and uh and one last question that I have how have you been embracing the AI tool in the office and maybe you can just briefly uh in a couple of sentences uh tell us uh what was what kind of revolutions have it made into your workflows and design processes and also thinking uh outside of the box where human couldn't figure it out but you're you're testing stuff with AI and you can have better results probably maybe you can discuss a little bit about that yeah so AI has been like a little bit on everybody's radar I think already for a year and I remember about a year year back me and a group of First Mid Journey adopters back in mid Journey version 3 we all sat together at the lunch table we're like what's going on with this we need to try it we need to try it so we' even plan to to make a fake project and try to present it to the rest of the office to see if anybody would notice we had all these really fun uh yeah I don't know everybody's grabbing it trying to implement it in different ways but I think that's also the the issue with AI currently it's a really powerful tool but nobody's really sure what the job is yet for that tool so the whole office is figuring it out at the same time it feels like it's like a big uh Global well maybe not just office it's actually more like a global effort to figure out what what can we do with these powerful TOS so uh so yeah there's about yeah that's great yeah yeah that's great yeah the AI made a huge uh step into architecture and uh idea generation and customization of different type of uh you know uh the Styles and right now it's an unthinkable part of the uh puzzle right now in design processes and it's being favored by everyone uh almost in the industry and we've seen its impacts uh in the last one and a half year and step by step the projects that are designed by AI started popping out as well so yeah I we're so excited about the future and what happens uh in the architecture industry well uh thank you so much for this uh short conversation cast uh I'll do a very short introduction about uh PA and uh upcoming workshops uh what kind of events do we have and then we'll give you the stage so just sharing my screen great uh so thanks everyone for joining this is our uh workshops page and the website and uh we do free events like this one and also member only events which is uh two days workshops in beginner and intermediate levels and also we have ticket based workshops as well uh make sure to check this this website uh this the section of the website the workshops page we have a a subscription which is a digital membership subscription if you become a member you will have access to all the free membership workshops that we have done it's like a 4ur for per month it's like less than a Starbucks coffee and uh you can Jo sign for three months and also yearly based they are all the same features but different timings different durations so as you become a member you will be able to uh watch or watch all these workshops or even join for the future member only Workshop so make sure to check this page out other than that we have advanced level workshops as well one of them is data structure 2 and uh federo borello will lead this studio we had done data structure one before and this is a bit Advanced version and uh more compr prehensive version of us covering the structural analysis and structural optimization uh in in Rhino and grasshopper using also caramba and speckle a couple of different plugins so Federica was also uh before in uh zaha Ed Architects and now they have co-founded encode they're working super deeply on the structure analysis so if you're interested to check this out we have another AI Workshop called combinational creativity it's going to be using generative AI artura teski probably most of you know about Arturo will lead this studio and he's the author of aad algorithms aided design workshop and we have this workshop on 24th and 25th of February almost a month from now we're going to deeply explore in two sessions about the generative algorithms and now how are being used inside mueran Spa stable diffusion you guys all know about Arturo probably we have polyhedral topologies uh and Alexandro zelli will lead this studio uh he's going to use blender and Alexandro is also uh the author of the tissue plugin uh for blender he he's a spectacular and great designer and he has deep understanding of parametric design parametric modelings and make sure to check this one Workshop out if you're interested uh there Al there's also uh special Dynamics this is a member only Workshop if you're a member you will be joining this workshop for free and setat will showcases this uh space optimizations in Rhino and grasshopper and how you can achieve it the last Workshop that I'm going to announce is body architecture it's philippo NTI Studio we have done three two body architectures before which were long for long for workshops and again in here we're going to use deeply uh Ai and mostly Rhino grass oper workshops to create uh body customs or body parts for yeah uh fashion design so if you're interested just check it out I I just realized that we have an army of Italian instructors in here philippo Arturo and also Alexandro so uh yeah just check these workshops out uh then we had parametric intelligence it was last week uh with timu if you're interested to watch this Workshop it was a four sessions parametric intelligence Workshop looking deeply into AI so uh to sum up we have uh discounts right now going on for all the previous workshops that are selected previous workshops not all of them uh you can use this pemy 40 to to get 40% discount this is a limited time offer and we never give 40% discount so if you're interested you can check that out if you are a member you will have the 55% discount for this work all the selected courses so it becomes more than half a discount so we'll just drop the code and in the the link in the chat in the zoom and also on Discord so you can have access to them and this is the limited time offer probably it's going to end two in two days make sure to check it out regarding the communication that we're gonna okay so if you if you if you ever joined this uh the Discord that we have uh this is our membership Discord uh if if you are a member you will get gain these badges uh different logos with green uh gold and also purple if you're free member you will be labeled as yellow and for this event we have this chat in here AI driven design practice you can just come here and uh type hi everyone this is this will be our main Communication channel this is our main Communication channel please type here so uh I see that many people are started uh just writing in here and also there will be a questions Channel please just use that channel to write your questions and nothing more and if you are a member you will be able to see our previous uh conversations and also uh follow the uh member only workshops that we have in here and uh we have also forums about Rhino Grasshopper And also me Journey Explorations that you can just watch them here and explore your stuff inside uh mid Journey well uh as I see everybody has started already using the the Discord it's very great please please use this channel to communication between us and also yourself and also between team members and Cass will have an ey and here or us will be asking all the questions that are written in here the questions channel will be only for questions if you have ever any questions please make sure that you write in the questions Channel WR in here okay and then we will ask them from Cass other than that if you ever couldn't join the zoom make sure that you are following from the YouTube link this is our YouTube it's now yes it's now yeah like this you can follow us on YouTube the chat section in here will be F uh monitored as well so feel free to communicate everywhere but the most important thing is joining Discord this this video will be available Forever on YouTube and also will be added into course section of our website well uh thank you so much everyone let me see if I can find my zoom screen okay stop screen sharing okay thank you so much and thanks Cass for joining us we're super excited about the session let's go perfect exciting thanks Amit for the introduction I'll uh start screen sharing yes so we're here all the time and if you need anything just let us know and right now I see your second screen I'm swapping it over awesome great yes perfect perfect great okay well um first of all thanks for uh to PA for inviting me thanks everybody for joining this lecture it's uh really an insane amount of people see 46 8467 somebody just left that was too boring it's clear uh coming in um talk to me it's not that uh you have anything I'm sure you have better things to do on your Sunday but uh I'm going to do my best to make this an extremely exciting and uh condensed uh talk in which I will take you through six of these design workflows that I figured out that AI can help us uh design um so so that's also a little bit the the the setup I'm going to talk about quickly give you guys an introduction to me then we go through the design process discuss those six points and then dive into those um I do enjoy a lot of interaction so please feel free to send those questions in the Discord and uh I'm sure the PA team will let me know if they're interesting and I'll try to react to them otherwise I'll try to react after the the discussion let me talk so um this is me I'm K esar I'm a project leader at mbrv and uh I've lived all over the world actually in Japan Cyprus Prague uh Chicago in New York and I've worked in a few offices the small ones to really large ones like mbv and B and in that time uh I've had the the luck of doing a lot of extremely exciting projects with within some of those exciting offices but what is a project leader exactly a project leader is a person who is between all the other people so I have my team on one side the budget the advisor more advisers constructural uh engineers and fire safety advisers and on the other side I have the clients the partners directors all pushing on me and I'm the guy in the middle trying to organize it all which is uh is a joke I kind of consider myself the punching bag within sometimes within the design processes but of course it's it's a joke because I'm extremely lucky to get to work on some of these most serious and most cool projects in the world um I'm going to take you through three of these and kind of talk you through the design process and and thinking so you guys can see why I turned to Ai and why it has a possibility of enhancing our design process so much so uh these are three of my most favorite projects I think this is the fourth quadrant in Prague uh actually I used to live uh close to here and then I would take the bus stop from this area and then always I would stand with my back to this big green field wondering what would be happening to it and five years later oh a bit longer eight years later I would come back here to be able to design something new for this area in a part as part of a competition because it seems crazy that this area is empty but as we started digging and doing our research it was clear this that an other architect way before us anonin Engle had designed something for this area but it just had never been realized so our vision was to reinterpret this original vision and uh uh use it to embrace a more modern uh architecture which would be hidden behind the veil or the filigree of sorts that that allows us to do something that not only respects the Heritage but opens it up to the Future in its way of interpreting the Heritage if I can say it that way so we that means that you have this old building which we recreate with the veil and then the new building behind it which sometimes it's closer sometimes it's a little bit farther allowing for vegetation and spaces where people get to meet each other spaces where if you have a coffee bar here you have this nice secluded comfortable space right in front of it and it creates this undeniable uh Beauty where the past and the future merge together so it's a I would call this like a historical research driven approach to design another project which we did was in uh Romania in Yash the Second City of Romania and in this plot where we had another competition with this beautiful historical buildings super old a lot of uh um cultural heritage which was really important to the people there important churches monasteries all around there and then we were asked to build something insane there huge but how do we add something while also being respectful uh uh so we we started by trying different things and this is one of them we tried to change the heights to align with the surroundings then we opened up the middle as a cour chart and then took all of the imported natural landscapes from the surroundings and projected those through the middle and then we started to carve away slowly at this massing with lines of sight in some way you would delete the building where where it kind of covers these old historical buildings which we fell in love with so much as we learned to understand the location better which results in a building which is there and it's large but in actuality it fits really well within this array of monuments and the peak language uh kind of mimics the old spires and towers of the town and the lines of sight are there you can still see and still connect to the Past um so you're adding something without adding something very beautiful image here by our render a company which we often use a and D3 by Chaba I think he's also in the chat here but it's very nice to see these um you know these these this type of approach so we're approaching by removing you start with the envelope and then you remove as much as you can uh that is important to the surroundings and here a fully different approach once again the Shen and Terraces project which I mentioned earlier um we were asked to design the new city center for uh um part of shenzen because shenzen isn't a city like uh like we know them in Europe it's a City built up of multiple cities with millions of people this is for the long gang City District we were asked to create a new living room then we start placing buildings in a way that they wouldn't affect The Pedestrian flows and then we rounded them off in a way that the people would easily come in and come out and be drawn in and through the area along for a really comfortable and friendly pedestrian experience this then we merg with the idea that in shenzen it's either really hot or really rainy so you we offset all of these um floor slabs for out and pulled them in and out and played with them pushed them down to create this seamless pedestrian experience which uh is kind of unmatched in that area where people would much rather sit inside now we're creating a space where people can actually be comfortable outside most of the year so so it's with this design language which we created the slab language we then pull and connected different floors creating separate connections through up and in between uh all merging the architecture with the nature um and what's also really fun is when you start to continue this design language into the larger buildings on the site you see that these uh certain areas like here's a conference center and here was a the hotel lobby but then you could actually walk from your hotel lobby to the conference center you could go up and then in the backside through up to the museum and then up to the roof Terrace without having to ever take an elevator just invites people uh to explore the building in a different way in a way that isn't common wasn't possible without this design language so and of course it's extremely beautiful undeniable I think uh but the the the fun thing in showing these three projects is that they each require a almost bespoke uh approach to the design and also to the design process because it's not like we're repeating the same six steps but the results are slightly different because in the different place it's actually we're we're creating new Step stories or new approaches each time and um that's something that I've fallen in love with and the the that I think is is where real powerful design can be found is in these bepoke approaches and that's why I'm also constantly on the lookout for these next uh ways to approach the design so that when um we started to notice that AI was getting more and more prevalent and the the the things that were being made with AI were super exciting uh I just wanted to just dive in head first I grabbed everything I could watch all these YouTube videos and try to understand it as as much as possible and um it's also so just that's why I'm excited by Ai and that's why so that's kind of the the link I'm trying to put here I hope it's clear but uh oh oh yeah so I also want to disclaimer that um I'm also not really an expert expert in this field I just love it I love to use it I love to apply it within our design process and I push to apply it uh so so the perspective that you'll get from me today is not maybe a absolute truth like this is the best way to do it it's just a very pragmatic way that you can possibly Implement these AI Tools in your design process uh I hope tomorrow actually already so so that's just so you understand my Approach a little bit because um when we look around ourselves and we're Googling what AI is or what the role of AI is you find a huge variation of opinions uh where a large amount of them are actually really negative uh surprisingly um and some are seem to be a bit confused about what it even is that they're talking about but these are the the things that people are putting out like uh I would considered almost fear mongering or really pushing a bad narrative to be scared of something that's new and I think that's uh really unhealthy when we're trying to still figure out what this even is what are we even talking about but these people already know it sucks like Elon Musk thought this when I heard him on the Joe Rogan podcast I was surprised how negative and fearful he was of these AI tools um but on the other hand we have um really powerful uh um visualizations being shared everywhere on the internet and Pa is uh is pushing a lot of stuff on on AI and design boom and these are just really beautiful and really smart uh sophisticated workflows where you know this one was great I thought this is insane that this guy Ste Carlos I think he's also talks here something he just shifting some blocks around creates all these different uh buildings or when Tim Fu posts another GIF which is just completely inspiring and amazing and you think to yourself oh that is cool we need to figure out how he's doing that I want to try that uh let's let's get into that but then I try to find a way to make it make sense within my uh design process when I am on a deadline when I do have to deliver products when I have to make the products that I'm delivering also be understandable to the client and I find in many cases that the the truth between this you know super skeptical point or the super embracing point is slightly more in the middle and you know I'm picking on these guys because I know that they're extremely successful and their work is very cool and inspiring but it's also and sometimes want to challenge them and that's also what my talk here kind of challenge the way we're looking at AI not as a a co-designer but actually as a tool so that's my truth or the truth or my truth is uh uh really in between I think this full Embrace or the full fear um because AI is a tool to enhance your design process and the tool though we're still figuring it out is also in search for the right a good job uh and it's not always really clear where AI really fits in when would you make a little GIF of your building spal spiraling out of the control and trees blossoming out of it when do you make that is it when everything is done and you is this for PR purposes only or or so so those questions are in my head as I continue just talk a little bit because and to do that we have to reflect on the design process I saw this sketch online um which was really great because it really made me feel like that is the design process for me it starts in this big jumble of mess of all these ideas that come through back and you're looping back constantly making iterations iterations iterations and finally hopefully you're peeking down to to the Future as you keep making steps and then at the end maybe there's one or two little loops and then okay what what is my design process currently look like and it was actually way different starting with multiple options but uh it's just just fun to reflect and U but let's keep it simple for now for today um where when when the you start with the project start and as the concept more or less develops U or the sketch des and you go into sketch design design Development and Construction documentation you're basically funneling down to just a few changes so you're not changing the shape of the building in CD uh phase but with that in mind I started to figure out key moments six key moments that we can grab ai's power and and uh use it the first one being in the initial inspiration phase um Ai and mid journeying specifically is such a powerful tool that uh it creates uh hyper realistic images really quickly that are super convincing and can really clearly uh translate your concept sketch or your thoughts or your ideas uh to an image that you can show a client or even or internal or even a client um but be careful for showing that to a client sometimes because sometimes they're better than the renders you make yourself later but um the second moment is uh this High powerered reference generator uh you often find yourself looking for go spending hours in Google trying to find good references but AI is way better at that for you so let's grab it there um then the third moment is uh play 2.0 I call it this is the moment in which um uh we're just outside of the first initial inspiration and we're starting to build physical models and even though with the digital world physical models have started to die out uh after covid we have a lot of international clients here were working on Zoom a lot and then having them look at this physical model on camera never really works but now I think we can reintroduce and re um the love for model making and uh so so through the embracing of AI uh the fourth moment is enhanced massing studies it's clear that in certain phases of the design process um we are um building blocks in in Rhino or SketchUp which just represent massing which we as Architects all really understand but the moment you show it to the client they're lost they're worried they don't see window openings they don't know where the front door is they don't know where they put their mail and where the the trucks arrive so you're just creating a lot of fear in them so the abstraction that we can the abstract level of products that we can reach in earlier phases can be bridged through the implementation of AI tools like stable diffusion like you'll see later on finally I'm I call it materialization uh for now but in later phases in the design process when most things have been set let's say open to closed ratios or placement of certain doors and important things when 95% of stuff is or let's say 80% of stuff is more or less set um you're often as to the very tedious studies where you're you know changing this wood to slightly wider wood planks this one to wider and then let's try diagonal let's try this and so so these tasks are uh take a lot of time weeks sometimes but I think by embracing AI at this moment as well um we can really speed up the design process finally um communication and visualization communication um our clients but also this is to our clients but also to external parties visualization Studios play a extremely large role um in bringing our projects to life however it's I've never been in a project that we haven't at the last day the last three days said okay this day render actually needs to be a night render and then you know we throw everything away we restart and um um so we restart and then we um you know waste a lot of time and throw a lot of the stuff away um I think that is that's not great and not great for all parties so we need to we can Embrace AI also in this phase as a way to um further uh you know simplify the communication process uh so these are the six points initial inspiration High powerered reference generator play 2.0 enhanced messing studies materialization visual communication which you're going to learn in uh let's say three or four hours and um and I hope you can apply them tomorrow already it's your job uh let's talk about the initial inspiration mid journey is an undeniable power tool for this moment in the design process um it's a extremely powerful dreaming tool that remixes from a really diverse set of images but it's very curated and well curated by a lot of artists and it's it's it's really powerful and it takes a lot out of our hands so it makes it very easy to make nice images like this and especially now so that this was I think 5.2 and then here in version 6.0 um it is extremely easy to make images which are almost IND discernible from real life so it becomes a a few words and you have a priter prizewinning design in a minute um but also here we need to reflect as you know as it's starting to get easier U we still need to imbue the images with our design intent we can't just randomly choose words and then you know use the nicest looking image as a way uh to choose one we need to create images that are buildable understandable they need to be fresh they need to be fun and they need to be convincing that when you do present them that you convince somebody that you can make this a reality so um I'm not going to go super in depth into like uh this is how you open the journey right now because I feel like most people have already played with it but I want to talk to you about the key um prompt and I'll run you through how to do really good and really in-depth prompting um so uh The Prompt that we using is you type first slash imagine and then you you have to name a subject towering structures then it's it's good to name and separate these in any way that you want uh an environment maybe in a forest or a landscape then discuss what lighting would want it to create and then color and then we continue into mood composition and a sort of a style or an aesthetic and these are these are the the seven key elements that I think needed are needed to make a really powerful prompt and this also gives you the control because you can leave a lot to a chance uh but I think as a good designer you need to be the one holding the steering wheel in the process mid Journey cannot be leading you to a good design you need to grab mid journey and take it to the good design that you have had in mind or envisioned or that that's part of the because mid Journey doesn't know uh what area you're designing for who those people are what they do you know do they like to uh eat dinner together as a family or are they all separate do they all eat in front of the TV these are these are strange things to mention but the it's really important to to pull the cultural context into your designs and M Journey can't do that so you need to grab M journey and take it take it with you there um I gave you guys all this uh prompt sheet cheat sheet that I've been working on I think it's on the server somewhere PA has shared it with you but um as you're working through these these prompts and using it more and more I really recommend you create this cheat sheet for yourself and share it with people in your office make it a Google doc and make it a combined effort and don't fill it with just trash but fill it with things that fit your design approach and fit your design style uh in that way you know you can really push your mid journeying to the next level the next tool I want to talk about is blend um there's it's it's very simple it just takes two images and makes it into one and I think it's the the important thing here is also so combine this with the powerful prompting and let you change something which is an extremely boring housing building and combine it with this beautiful sculpture into a housing building which has these extremely beautiful light shaft uh elevator shafts um uh blend is a is a powerful tool there as well and the results will be great if you you know grab on to it with that prompting another part of the mid Journey prompting which I need to mention just for you to understand it is this aspect ratio so after your whole prompt you do slash imagine beautiful house in blah blah blah blah blah in the forest gloomy lighting mood um then you type your aspect ratio and to understand is that M how moury works it I'm guessing it has somewhere a huge database of images and by then it starts to draw from those images and each image has a specific aspect ratio that it's drawn from so it's good to keep in mind that as you change the aspect ratio it's also going to draw from different images so let's say the aspic ratio is square and you're asking it for to build a skyscraper um it creates more of a a a Stumpy little skyscraper but as you bring the aspect ratio wider and wider it's starting to have problems showing us the skyscraper even though you know it still could do the Stumpy one in the middle here it decides not to and creates more of a zoomed in image where you have this nice fista so if that's what you want that's nice but in the other direction if you want your skyscraper to be taller it helps to enlarge the the aspect ratio so you can part of the prompt crafting is uh also the yeah sorry I thought there were messages coming but part of the prompt crafting is also understanding the the mid Journey tool so so to enhance your uh desired outcome you have to find an aspect ratio which reflects the the the thing you want to create so if you want to take a really tall tower you have to do a really tall aspect ratio it's so simple but it sounds so but there's also a limit to this got to mention that for a second that after some point it starts to not understand that there's one image here and starts to create two images uh one at the top and one at the bottom try to stitch them together it's very interesting to see that happen um let's say settings it's important for you to always uh keep remix mode on so as you're creating more variations it allows you to change the prompt slightly and for the rest I would just keep it a little bit like this that's just my recommendation but feel free to play with it B um and now an example so we're in real life after a meeting We've Ended up with a sketch like this it's some type of housing building where uh which we're doing for free or a very small budget and the idea is that rainwater is gathered within the the housing building itself and it acts as a buffer for the rainwater that's coming in and then it gathers this water and then uses it to flush the toilets but also uses it to cool the building so by expelling the water over the facades we can use it to cool the building so that's the idea and this is the sketch this is all you have and you have a meeting in one hour so what do you do you just grab your prompt sheet and this basic prompt setup and you just Dive Right into mid you throw it in there so here you see apartment building eight stories overflowing water in a pond so location bright sunny day mood lighting fresh human perspective photograph architectural I did two three thinking I want to the building is not so small so I want something a little bit of space and but they all suck so what do we do just flip the ratio let's see what what we can get and starting to find things which start to look interesting and as we dive in further trying to change little words here and there I think here I added um um I changed the the okay I'm not sure what I changed here but then as I developed that one the one in the bottom left further and I start changing things oh this starts to get interesting convincing but it's not fresh it's not uh doesn't look like a day it looks like a sunset so these are not images I would be comfortable showing a client just yet but I'm starting to grab some of these making them bigger and then okay let me re redo this one with a a large variation but uh put in uh a bit more modern buildings I wanted something wood but the result was horrible so you kind of go back and forwards and just want to illustrate that it's never just type in the prompt once and it's done you're constantly evolving it further and further switching the aspect ratio again then adding more vegetation removing a bit trying to you know your St so then I thought to myself okay this is awesome this image is great it does make sense that if the water comes over the edge it automatically Waters your plants it does make sense to so so all of these things just look this started to get really exciting to me and I thought okay can we find a more distant VI so I tried to uh tried to get it to be a drone image but it didn't work but I then in that Randomness I did find this image which was beautiful and I thought okay what if I just Photoshop it a little bit and then put these two next to each other um uh you have something which is really convincing um so so just wanted to illustrate this more in-depth mid Journey workflow because it's good to understand that you don't find something in one second and you never will it's always about continuously generating tweaking adjusting the prompt adjusting the aspect ratio adjusting the style or the the all of these types of things and then you want to end up with a result that's clean fresh and happy and has a very clear uh clearly respect the initial ambition that you had so if you remember that sketch this is almost one to one that sketch but it just looks super real realistic and buildable and this on the right just makes you understand the positive qualities of it um so um I think mid journey is an essential tool When developing Concepts and bringing them to real life um like these things that we were talking about before before they they can be Fantastical and they can be uh crazy and out there and really pushing the boundaries but they have to be real and I think this feels real and I think images can work really well if the concept that you're trying to push for like here this really highlighted plin area which then pulls you up the building you think okay I see it I get it even though that this image is way too Moody and way too Misty to be uh you know shown anywhere it's it's good to to see that this thing uh which your ambition needs to result in be be grabbed by the you're trying to create um because it does have its limits we love mid Journey also because it pulls us down these weird wacky routes where I don't know pieces of broccoli hair are growing in your between two panels of glass and it's cool but it's too far outside of the real world and that's why images like this will never go anywhere and images like this for example all end up on the cutting floor whereas this one is the one we feel free to show and we want to uh push for in meetings it's clear fresh buildable realistic and it clearly represents the concept that we had as our first ambition yes can you hear me yeah I can hear you sorry sorry to interrupt uh can you can you lower the audio of your Discord because there are so many messages and the background audio is just distracting sometimes sure I didn't even hear it uh let's see you need to right click on the value yeah okay here we are again uh Swit yeah switch your screen again please okay no no reply means yes I I I I still see your notes yeah from display setting on top top left yes cool got it awesome great thank you in general am I am I hearable and everything's okay yeah all Everything is Awesome thank you I see some people complaining about the sound to low should I sit more closer to my should I move my mic a little bit um okay I hope this works better now yeah it's very good the sound is like uh if you can get closer to the Michael I'm just going to continue um yeah let's think right now I would like uh to ask you guys uh to try it once as a little assignment um uh because just for fun we had a project uh pro bono project for the people of Norway who've received a grant to develop a new community Hub the budget is super tight we don't have a lot of time uh we for because we forgot about the meeting just an example situation and we need to create a new cultural hub for them so I ask you guys all to log in to your mid Journey accounts and take a minute and use the The Prompt setup that I created for you guys or this seven point system I think to understand uh to create something new and sighting which could uh be which is clear fresh buildable all of those things that we were just mentioning it's good to just I'm going to do the preg Google for you guys the SII people are the indigenous people of Norway they used to live in all of these types of tents with reindeers uh reindeer hides over it they're really well connected to Nature and they love this red uh blue yellow color scheme so I think those are the key features you need to integrate into your design um to have a successful design so I'm going to give you guys about 10 minutes I think would be enough that would be really interested to see what you guys come up with in that time try to mute my Discord in the meantime what's the name of the culture it's the Sami people I believe you can still see it on my screen see Sean disagrees about the unrealistic images being a a limitation I think um I understand what you're saying it is in a way uh something that can can also push you further and drive your design process to do something more than you thought of before but imagine people come to you they're asking you to build a building they're ready to invest a lot of money and some really exciting things but you're coming back at them with this picture of uh some type of zaha building which is just not within the price range of many people you know in the really basis of it you're here to to design something which also fits um the the brief given to you um so I do really think that that if you you know if you create unrealistic images um you're not going to you know gather the trust and understanding from your clients as you continue with it so uh it's something you just have to be aware of and the limitation is also um solved by being aware and understanding it but um yeah indeed so so so the discuss it's on the Discord Sean just said I'm coming from a research academic context versus a commercial practice indeed I think that's the difference I'm just trying to to where you may be more busy pushing the N the the boundaries we're also trying to create realistic so I think I'll put on the screen is this just quickly when crafting your prompt now really start with your subject so it's this tent shaped uh Cultural Center then talk about the forest so it's in Norway maybe in uh talk about the environment then it's maybe in Norway or something like that then you want to talk about the lighting so bright hard lighting you know I can imagine the snow something like that you want this really Fresh Blue Sky so I think implementing those types of fronts will get you something that you can control way better see some people lost the prompt the assignment so let me resume here I see just a few things coming up so if I were to look at this image and just to reflect on it I think it's extremely cool but we're missing the relationship to the to thei culture so there's nothing that nothing has to do with these colors these bright beautiful colors it's beautiful right but we need to go further than that um this one seems to be quite exciting actually quite cool fits the brief also oh yeah here I was expecting so something like this really uh pulls in my intention more because we're referencing these old Huts that they used to live in you're putting many of these Peaks together in some way so it's talking about these Huts coming together you're talking about the culture and creating spaces for a community to come together so this you know I feel like we can start to tell a story around an image like this yeah I think you this is a nice Hut but it's not exactly the the the resulting it it doesn't really reflect the community center or something that we can add to them so as an architect I'd want to add to them uh give them something of our knowledge as well this is cool give it two more minutes if you guys want to post it in here we can flect a little bit kind of want to know the prompt for this one but what you do have here you know despite the design maybe being something that doesn't directly feel like something I want to build bu or would want to propose but it is warm it is fresh it is inviting and exciting it's a little bit Las Vegas meets the Sami people so I'm not sure that's the reference they want to go for but um it's a good one to see and it really starts the discussion see a few more coming in and then I'm going to keep going Discord app uh can you hear me see this is a cool image C can you hear me it doesn't it's not the most successful mid Journey image or a he image but it does the other things extremely well and that's why I would it it is clean it's fresh it's inviting it's exciting you grab a concept you grab the concept of this tent it's got something of this like nature reservation idea in there so I think these are the [Music] um these are the most important you know features of these images so they don't have to be perfect but they do work very well very cool so I think people are starting to get the the you know buildable clean fresh this is cool I like it a lot um I'm going to continue with the lecture and I'll take a a look at the things in a second at the end again I think it would be cool if I will find find we can find some of the most exciting ones and reflect on those uh also after the lecture so here we go so this is the thing I created myself um in that little time I gave myself 15 minutes um that that though it is successful in some things you know it is it works well as an architectural reference to these Huts um I did have to photoshop in this quick pattern here which is reference to the old Sami uh textures textiles that they do and I added these people in but there's a a quality missing in it which was in some of your images that it's too dark it's spooky it's gloomy and it doesn't um it's not an image you could ever show really to your client or to uh you know in a professional sense so as we continue just that was just a nice first little task um let's continue then the second part is the the high power reference generator I found that in a design process we're often trying to communicate our Ambitions to the clients and to um you know within the team about what we want to do and what type of spaces we want to create and then often you find yourself Googling online office pool office hip office nice child daycare with happy children or Cafe which you know is very good for the communities but it's it's really hard to find good references that hit exactly the thing you're trying to show um that are also from your own firm or or so nondescript that you can reuse them because you know we don't want to be using images from other firms's work and there that's where uh mid Journey shows a huge potential as a way to upgrade your programmatic or style references um because as I started Googling and this is where it kind of tweaked my mind I actually was Finding mid Journey images in Google which then made me think why not just use mid Journey uh without Google at all so um it gives us the the access to these great references so here let's see a super green office Stace that looks great and fun but maybe we wanted to be more cozy and smaller and more of a startup or maybe actually not plan that we want we want more flowers and these tweaks they happen in just minutes instead of restarting your whole Google search or this big database of projects that you have that we all have in our heads but um it's the the power of these references and here students in a cafe or should they be elderly or should they be in China or should they be aliens it's the freedom um to create high quality reference images um that mid Journey offers us is just unpar paralled and yeah it's just something we should integrate in our design process I believe so that's our second one now we're on to play 2.0 and before I kind of can go into play I will kind of give you guys this broad intro into stable diffusion um it's going to be super direct and super uh um uh focused because of the time that we have but um I think you need to understand stable diffusion a bit better to be able to use it unlike mid Journey which is so easy and accessible stable diffusion which is the other program which I hope you guys have had the time to download uh and get set up uh is um you know we need to talk about a little more because what is stable diffusion stable diffusion is a deep learning text to image or image to image model based in diffusion techniques and what that means is that it has a huge database of images and you can provide that database of images to it and it starts at a faded noise image and slowly starts to develop a realistic image based on that initial noise image and that that noise is random and or can be generated by you or or whatever but that's the the base of the the the That's the basis of it so what are the the powerful parts of stable Fusion is that it's completely adjustable unlike mid Journey where it's all hidden behind some type of window every part of stable diffusion is adjustable even the interface that we're using now automatic 1111 is um is also adjustable and completely open sourcely uh available and people are changing it some people are using comy UI for example um um and you can tweak the inputs you can tweak negative or positive inputs you can tweak how much the stable diffusion listens to your inputs and then you can tweak what imagees it uses as a reference to create the new images so you can give it a million pictures of cats and zero of a dog and then if you ask it to generate a dog it just can't do it it's it's very interesting to see but but stable diffusion gives you that freedom to to um to to to tweak but on the other side the negative side the downside here the sad person inside is that um creates a huge barrier to entry uh it's extremely complicated sometimes and um really hard to grasp what you can do with it and what's going wrong when you're not generating images which meet your standards so it requires a lot of testing and a lot of time spent in it so um that's why I will also kind of run you through it and get you uh uh going so you know which know you should be playing with and which knobs you should just set and let go or which knobs you should just never touch um so let's start at the model you open your your let's start right at the top when you open your stable diffusion and I'll go into it in a in a bit live but now just in a in a slide setting at the top here here you have a model selected which is the basis from which it draws all of its images and the models are completely downloadable from online so there's all these websites where like Civ I AI where uh people post their models that they've been working on seems like they're only posting pictures of beautiful women but luckily there's a buildings tab which you should go to and in there you can find a few key models and these are the ones that are also shared with architectural real mix it's a really great selection of bird's eye and eye Lev views of buildings in in perspective and the two-point perspective which is really important and it gives a really good clear architectural r style another model which works really well is epic realism this one is just insanely powerful in creating really realistic images so it helps with Reflections and those types of things and then there's another one dream shaper and all of these you guys have received in this file it's a it's another model which can add a bit more artistic flare to the images that you're generating um and these need to be downloaded and installed in your stable diff uh folder on models and I hope you've done that already but but U if not don't worry about it I'll run you through it and you'll figure it out after the session and then if we let's just T Test them right next to each other I put in this very simple prompt I hope it's readable but photograph of a cube wooden building in the woods on a sunny day very simple using the architectural real mix one uh I locked in the seed just so yeah then here we're this is the depth again of stable diffusion but then just want to mention that I lock in the seed and the seed means that the starting noise from which the image is generated is always the same so if you lock the seed and hit generate once and you don't change anything and you hit generate again you'll get exactly the same image but the moment you unlock the seed or what here randomize it each time you generate the starting noise will be different so the resulting image will also be different um so here we go let's change the model to the Epic realism one here super realistic I would say but it doesn't really hit what we're looking for right now nothing exciting and then here with the dream shaper one very beautiful greed but it's interesting to see that um these three models with the same prompts and the same noise creates slightly different images and that's um uh so that's the basis of your whole design process more or less um a further development of that and I'll just give you what you should put in or what I think you should put in is that you can mix these models because some of them are better at some things and some are others and we are looking to create beautiful realistic architectural images that are also a bit dreamy or sometimes a bit more realistic so if you just combine these so you go to your checkpoint merger and you combine epic realism with architectural mix at 65% which means that architectural mix will be at 65% and the rest will be epic realism you will do the same for um dream um then you have two really great models which you can be using in the coming you know forever I hope uh until they come out with new ones um and that mix gets us an image like this which is bit dreamy a bit realistic a bit beautiful and um here I just unlock the seed and start generating a few more things but the the consistency and the image quality goes up because we've created a good basis on which we are going to be generating our images the next one The Prompt window uh The Prompt window you probably saw up here unlike mid Journey you don't have to do too much complicated stuff in here you can just type using the same uh seven points a prompt into here and then the top just prompts that you want and the prompts that you don't want it to reference you put it in the bottom here you have to imagine that it's just like a Google search so it's going to look into its folder the the model that which you created it's going to look for images with the same prompt tag and use those to generate the new image from um and it does the opposite for the negative promp so that's the the freedom or the tweakability you get when building this and I'll show it more detail a bit later as we dive through um what you want to do in stable diffusion and these are things that um uh you basically just set up once and then you forget about it and you keep using it um what you want to do is you want to set up a style and this is also one that I've shared with you in that folder where um it's a basic style where every prompt gets these words added to them photograph photo realistic hyper realistic hyper detailed and these ones will just support your image and bring it to the next level we always want to have the small crowds because that's what we're trying to create exciting inviting spaces and what we never want to have is a watermark or we never want to have text sensors deformed people thick lines or broken or floating limbs cuz that's an issue stable diffusion has it creates all these hands floating all over the space so um I also tried to ignore face up close or selfie or faces really close I think those just really interfere with the stable diffusion uh model so then we have our original PR and then here with our style included you can see the resulting imag is like 50% better way more vibrant and way more realistic so by implementing this basic style each time you upgrade your image uh a lot um and then here just just I added the so I didn't like these columns right in the front here so I added as a negative prompt uh the column and it removes the columns right at the front here changes some other things as well but it's good to know the the tweakability now that you have in comparison to Mid Journey where you could never write remove the column of this one and it'll keep most of it the same but just remove that one feature the the you know the power of stable diffusion is just you know unmatched so I really want to push you guys that's also with the previous majoring one is uh push yourself to make thoughtful prompts and uh even though the simple ones the result is fine a thoughtful well thought out and well you know designed or engineered prompt Garners way better results and very clear to see um sampling methods and steps this is the next little thing in your as you go down uh the sampling method is basically the way it creates the image and this is the clearest image I've ever found that it is from a YouTuber online and he just generated all of these with the same prompt dog for every sampler and then with a different amount of steps and our goal is to create uh use a sampler uh that creates an image in the least amount of steps and creates a consistent image each time so you could see that this UL a for example creates a really nice image in 10 steps but it's very different from the one it created in 15 and then very different from the one in 40 so you know you are hey sorry we can't use you because we're trying to look for consistency we're trying to get trying to get to the image quick and the same so so that's when you're looking through all of these it's from practice and and testing it becomes um clear that see should highl here a DMP Plus+ 2m cars horrible name but it's one of the better models because very quickly within 15 to 20 steps you have an image which works really well and it's consistent or ever if you keep sampling more steps which just takes longer so you're looking for something usable um then the CFG scale the next thing down it's kind of hand inand with uh the sampler with the CFG scill is basically how much it listens to the prompt that you give it uh and how much it grabs onto the images so in the on the left here if you put it really low it'll be like okay thanks for your prompt but I think this will work better here or this will work better there and if you put it all the way to the top you'll see that the it'll really try to grab onto the prompt that you give it however sometimes it really messes up the image because if you look here examp rer a completely wrecks the image by increasing the CFG skill too much but here for the one that we are using the mp2 you have this nice zone between three and seven CFG that works really well but it doesn't break it to go higher it just doesn't help us in any way it just makes it more complicated but so this is The Sweet Spot run you through the prompt that we have cg2 cg4 cg6 7 8 10 12 and this starts to get a little crunchy a little bit horrible and you know if you as you push up the imagees gets worse and worse so The Sweet Spot my experience and also what I've seen online is between this three and seven for the dmp2 I'm just going to call DMP uh um sampling method batching good to understand this is we are busy people you don't want to be sitting there clicking generate generate generate generate every time so just increase the batch count randomize the seed and increase it to something like 12 gives you a nice Square image which is full and then we can zoom in and we can look each one uh look at each one and evaluate them see which ones we like and which ones we don't and then when we have one that we like we can go to it in our viewfinder window here and lock in the seed while locking in the seed like like I said we're starting with the same starting noise so if you hit generate now it'll just keep generating this image over again so now we have a locked in seed for an image we think is really great and we can start to work on it I feel like the dirt path here is horrible so I added dirt to the negative prompt and it removes the dirt uh path and then you know just for fun added red paint to the building here gives us a nice image the dirt path came back so so which just so it's understandable I just want you to understand it's not science but it's in some way you know you're working on this image slowly defining it further cutting away this part adding this part changing that and then you know you can do a batch again and you can adjust the prompt again and you can batch so that's a little bit the the the methodology that you're using so I think these are the points these are the key things that you need to be playing with the prompt the style set it and forget it then you have the sampling method set it and forget it sampling steps same thing batch is something you want to be playing with the CFG scale something you can play with but the main tool that you should be using when going from text to image is this prompt uh window a bit more depth image to image because as we are this window you can click here and send this image directly into the next one which is the image to image and it will transfer all the stats over except for your style which um just good to mention that and then if you hit generate it generates a similar image based on it generates let's say an image based on this image using this image as let's say the starting noise for a new image um and here you have a few so if you scroll down you have very much the same things except for two which are different I believe the D nois or strength and the control net these are the ones that you should have installed or or I think the control net I hope you have installed because it'll be important um so the D noising strength I'll go into control net I'll go into um the control net you open it up it reveals more stuff there and let's click all of these enable low vram allow preview upload independent control image doing these things gives you the ability to place uh images in here and that's let's do that so we place an image in there and just for now I'm going to tell you go to the depth one and explode that and then you get one let's see then you get this control net image to use as a base I'm just going to run you through it and then kind of explain what it's doing and now when I change CH the prompt to Orange the building changes into an orange building while everything else is kept quite the same and here I changed the building into a jelly building well everything else is kept the same so and here a rainy day for example so the control net gives us the ability to keep the image that you have in uh lock it into the to the framework that you give it because of the input you put into the control net to generate a new image and then 's few ways to to use the control few different types um one for example is canny depth mlsd uh and scribble these are the four ones which I would recommend using a little bit and getting used to where Kenny gives you a very detailed uh line workor based on your input um the depth one is a bit so so that'll also effector output as you see on the right here so the same jelly um prompt gives us a prompt which is way more related to the previous image because the control net that we're using is so hard on that the output image whereas when you ask this jelly uh prompt with the depth uh control net there's a lot more freedom there to play with uh for stable diffusion since it doesn't show you know it doesn't have all these vertical lines in here for example so the stable diffusion has more freedom to play with it uh Ms mlsd uh is one which just generates the straight lines which looks great on the box but then the trees and this weird UFO things that start appearing are just a result of you know having no control net input on that area so it works well and it doesn't work well and the same with the scribble sketch where you know it works great for the box but now we have these strange artifacts on the ground floor so it's good to be aware that when you're using [Music] um when you're using these different types of control Nets the result will be different and you can play around with it and get a bit used to it but good to understand that if you want to keep things like Line work really strict or where people are placed let say you have some people placed or you want to keep them there the canny is not a bad thing to use and if you want to give more freedom possibly the depth is more more uh great to use so so those are the the few different um yeah just to understand that the denoiser um as you know you scroll down here this is the den noising strength and the den noising strength affects the starting image so this one on the left here and by increasing the den noising strength the input image which is put into the stable diffusion is either made more noisy or less noisy and when you reduce the noise like a three like you see here the output image is going to be very much the same and when you increase the noise the output image and stable diffusion will have more freedom in generating the image so that's just good to be aware of here so when your input image is in one one way and maybe it's really bright red and you don't want it to be bright red it's good to increase the the noise so it so it forgets it a little bit that that was a bright red thing so then so so for different functions there's different sweet spots for your noise level so if you're generating images like we're doing now the noise between 6 and 085 or 08 are really useful but if you're just upscaling or tweaking images little things refining them I would keep the noise way lower to keep more of your original image in there uh let's go to the upscaler that's the so that's a little thing right here click on that you can send your image straight to the upscaler and this is a little bit self-explanatory scale by or scale to you can also by the way go to extras and scale image and load your image into here so that's what all these online ups scalers are just doing right so you upload an image they put it in stable diff I guess or something and then they send that back to you if you ever used some of those but um you want to choose an upscaler just use this Arison 4X plus that one this works fine and then you can upscale your image it clearly goes from the weird pixelated artifacts to really clean lines and works very well by uh so just so you know uh next thing I want to dive into quickly is inpainting uh inpainting is really powerful tool but it's also I'm not really sure where to use it yet but I wanted to let you guys see it and so you can kind of understand uh what it can do for you um okay but okay yeah so here I as a joke I said okay then you accidentally refresh your window what do you do now you have nothing left you can go into your PNG info tab right here at the top upload one of your previous images that you generated from your AI folder wherever your images are generating and use the send to image to image or send to in command to refind all of your previous work that you did so you don't have to you know re-enter all of the commands and all of that stuff but it just puts it back in there for you you know re restarting your workflow more uh let's go to in painting ah sorry first I wanted to mention two things good to know that the Styles tab is clear then as well and also good to know that your control n is clear then so you cannot find that from the previous image so you'll have to reset those uh before continuing if you want those in but just want to mention that so in the inpaint tab image to image generation in paint you can start to do some quite exciting stuff by using the tool here you can draw in a surface here and just regenerate that area while keeping the rest of the area the same so here we can change this colorful jelly facade which we've made before into this beautiful green facade and then we can go down you know it's not that easy as made it seem because there's again a thousand little tweaks where you can impact the area that you masked or you can impact the area you didn't mask you can consider the so the the image which stable defusion takes as a starting point to generate the new image to be the same as the original or we can uh use latent noise or just some random mix of noise to generate uh a new image so just to be aware of that if you wanted to fit the context would just keep it on the original if you are fine you want something completely new like a person which wasn't there before F lat noise uh and here two tools you can impink the area only or the based on the whole picture and if you only do the mased area it means that it'll zoom in and regenerate that area based on all of the knowledge it has of pictures of similar size so it's a way to upscale certain part of your images like faces or hands if those are weird you can grab one of them and adjust it um so here we go let's try another tool we go into in paint sketch this is even cooler I think I think this is really exciting one to play with so let's just paint in using the same tools you had here before now you can pick the color and a brush size paint in and paint in also as a tip paint in with a little bit like where the Highlight would be and where the Shadows would be so here you have the the two people I thought two people walking it's in the rain they're walking towards their little jelly cabin and um then so I put that in the prompt people in rain coats from the back Sheltering from the rain and then hit generate and it completely does something insane but as you're playing with the noise with the CFG skill you can start to get really convincing results or let's say let's add some trees or let's add a little uh child here then you know let's full scale it and we start to see the face is really weird on this girl if you saw it here so that's where stable diffusion has a lot of problem and then we just go back to the in paint tool draw a little dot on the face and ask it to to generate a child's face okay the child's face looks more realistic and the image becomes more acceptable then with a little bit of tweaking in Photoshop we're able to go from using all these tools image to image uh upscaling it inpainting it um you know all the stuff we did we're able to create a whole new image based on this basic image which we started with a few minutes ago I think that's the power of stable diffusion so that's very cool but how do we apply this within an office um not used to talking for uh two hours straight but um it's let's talk about how where we are in the design process can you hear me 2.0 initial dreams have been set clear sketches and Ambitions have been been uh discussed and the team is aligned based on possibly three main directions and uh We've created some cool models and want to show them to the client so let's say that's the phase we're in so we're in this little bit sketchy phase you have your models and boom you generate one great great image it just so easy well actually it's not it takes a lot more work than people will let on online it's it's a a process in which you develop your input you upgrade it in programs like Photoshop and then you work through it in things in stable diffusion and then finally combine the elements of the different images that you love um that you love in uh in stable diff generating nice image for you um let's do a real world example after a productive meeting you know people yelling your partner or the the the director in your office or the project leader as youve worked all day on these beautiful models that pratically grabs them smashes them up and tries to make new pieces and you're trying to figure out what meeting luckily an intern on the team is taking a few pictures during the meeting these are the only images ah did I mute you guys hi yeah hey sorry can you hear me Cass yeah yeah sorry okay that's fine so uh you lowered the entire computer's uh audio that's why you were not hearing me but that's so far it's all good and we're on a good track if you need any break uh just let us know and yeah we can go for a five minute break and have a coffee back I was just talking forever and you guys oh we were just enjoying what what what you were showing yeah thank you okay good good it's funny I was just I was going I was like nobody's talking with me but that's fine no that's good that you were not distracted with us so it was super awesome let's continue give me two two more slides and then I'll take a little uh five minute break if that's okay yes yes for sure let's go perfect so um let's see let's see you guys can see my screen again on the left so here we have the images which the intern yes we can see your screen perfect so we have the images that the intern grabbed for us and from the bird's eye and the eye level and a few key sketches related to them so one of them is the green growing up on it the other one is the landscape which continues up on the building blots and the other is two buildings separated one towards the park and one towards the city and with this in mind I will uh run you guys through in after a 10-minute break because I need a fresh coffee and I think you guys as well uh we'll run you through how to build it up so we can create a really nice image so give me a 10 minutes I'll be right back all good thank you so much much so let's get back in 10 minutes okay I think I think we back I hope everybody's still here welcome back everyone yes we're here cool let me resume the share and we'll just go back right back in So if you remember your intern ran into the model shop and uh just took three pictures of the three options which were there and they're all based on different conceptual approaches and um when when taking these images and you're developing them um in stable diffusion we need to do it in four steps I would say first one is prepping the input for Success second is generating your starting image and trying to understand if you're going the right way and then then tweaking that takes about a million times and then finally finalizing it in Photoshop and then this three step is um is kind of this fullon repetitive one that you'll keep doing forever and ever um and I'll go in right now I will switch the screen set my other screen now and I will kind of walk you guys through this first one so in the the images you guys have received let's take this blue foam or I'm going to take this blue foam 01 and prepare it for Success putting it into Photoshop key is that when you see this you you want to return the focus to the main center stage of the project so just press the C tool crop in around it and then imagine yourself being stable to fusion and looking at this image SP for a second looking at this image there's weird stuff happening up here there's weird stuff happening up here and we need to get rid of that so I'm going to quickly create a new layer not it's not really a Photoshop tutorial but I'll just name the things I'm doing so you can kind of follow along if you're not so used to using this tool but um so with the I tool just press I click on the here grab that color go to my new layer and then I'll quickly color in this top Corner doesn't have to be perfect but just get rid of all of this weird stuff happening around the table and then here this model get rid of that completely boom I think now we have quite a good uh starting image that we can use as a b so I'm going to export this one quick export it and in here I'm going to create a new folder call it say The Green Tower because that's the goal of this one I'll call this one one this is our base image which we'll use for the control net and now we need to of course make it a green Tower and we can try to pull all of that input from stable diffusion but I've noticed that in the work flow it's way faster if we start to already Implement some of the features that we're looking for at this moment and it can be so quick and dirty so if you just saw I just Googled Park top VI I place it in here looks good I'm going to put it in at 30% or something take my uh lasso tool click right through it here I see this building here I want to kind of allow for a little bit of Road there I would imagine get all the way around here and then click the masking tool and here we have a really clean clean not very clean but we have a park which leads up to the building and now we want to help stable diffusion to understand to GRE facade that we want the green wall to grow up onto the building and up that facade so let's see if we can find one which just represents that uh for us I think this image is a bit too zoomed in maybe this one works well it doesn't have to be perfect but just has to get the the goal across so I'm just going to quickly crop this one out because we need to so I selected it with the m tool contrl j delete the background control hold control then I select the right thing here I'm just going to scale it up so I know the whole thing is covered more or less sorry hit ENT hit enter put the Press V five so I'll pull down transparency so I can kind of see what's going on behind it and here I create the vertical wall which climbs up the building and I mask it out a basic image that we can now use um to inform maybe just out of experience I know that this image feels a little bit uh it's not very detailed so I just pressed contrl L and I'm going to pull up the dark tones in here just to get a nice clean base uh image and here I'll grab all of these put them in the folder I'll just call it one and file export quick exports okay two fine so this is our base image that we've just prepared for Success this image which has all of the information we more or less need for it to start generating images and stable diffusion that will create good results for us so here I have um the stable Fusion window as we just kind of walked through it um I know I want to drop in the image here this is going to go create our base put that in here I know I'm going to go on control net like we just discussed before enable all of these key features enable low vrm allow preview I'm drag and drop it right in here for now I'm gonna just from my experience I know we don't let's say I generate this cany example the resulting image on the screen is so crazy it's going to be really hard for stable diffusion to work with this you see all these squiggly lines it just doesn't allow stable diffusion enough freedom to create something so I would recommend going for a depth uh one right now and that's it we're almost there now we just have to put in our style base that we uh put in before Arc realistic basic and let's just see what happens if I do hit oh one thing put it at the right scale so this build this project is a little bit higher than wide so pull down the height pull down the width I mean sorry here we go let's generate now let's just see what comes out if this is something that we want to be working with it's going to take a second I'm not working on a really great computer here but it's it's fine um what do we see here it doesn't look great at all the trees are coming through so you can see that the base input that we did in we put in um really affects the outcome of the image so what we want it to be we want it to be a bird side VI it's going to be a housing Tower next to Park um let's say it's in New in New York and um maybe just emphasize the fact that there's lust vegetation climbing plants something like that let's see what happens I'm not expecting the result to be great yet or even close to great and it doesn't have to to be great just some parts of it have to be good enough for us to work with so here instantly when I look at this I think to myself okay the green wall here it's working it's looking good the buildings and the surroundings not so great yet I don't like this Lake doesn't feel Urban I know why it's doing it but let's get rid of that Lake Pond sea I don't know I just don't want any of that I just want it to continue random uh buildings kind of continuing through It's seems to have a hard time with that maybe we need to so now I'm thinking to myself okay it's not really changing this so it's having a hard time filling in something new here so maybe we need to add more noise let's just pull it up to point8 so the original image so the image that stable diffusion is taking is less looking like this original image doesn't look like we can get rid of the the the the the water and it's still not climbing plants up so much so let's go overgrown maybe so this is the the moment in the process that we're just tweaking and tweaking tweaking even though we already know we have one great facade in the starting prom maybe we need to park just emphasize the park one more time see okay now we're starting to get somewhere cool so I start to see these things and feel like we're going in the right direction so what I'm going to do is going to start a batch here for 12 images and at the same time I'm going to start preparing a new uh base that we're going to reinsert into the project so let's say we did like this one this one had a great uh a green facade this one had good surroundings did any of these have good parts did there's a new one not really nothing really exciting yet but that's fine then I'm going to grab these two at the same time contrl T just resize them to fit the window it doesn't really matter so much we're not doing science here or something it can be a bit Freer let me know with this great tool the wand tool the object selection tool see if I can select the tower it would be great if we could get a mask on that so now we've just masked our Tower into the old surroundings but we still have the issue that this Arc isn't really growing up like we wanted to so I'm just going to quickly merge these you can also just rasterize the layer that's my little trick and then I'm going to increase the brush size go to press s so you've got the the stamp tool and I'm just going to stamp in here look at that I'm just going to stamp in a little bit of green and then I'm going to go to the tower here and do the same thing uh going down ruze it again just so I have a bit more freedom push it down that's not going going right but let's just grab the last hot then duplicate it it's completely um I don't know how I would call this like uh a little bit a a gangster or something you don't care you don't care you're just going to do something you're bringing it to the next level you're not sure yet we're pushing the things around a little bit and we it doesn't have to be clean it doesn't have to be perfect so here we go this is our third rendition of that let's see if something cool so now we're done waiting we can kind of take a look at this does anything jump out okay we have a better background here this one looks better I think we don't really have one which really does the overgrown stuff so well so what I'm going to do is reinsert see a lot of chats use autofill can sample many layers New York okay not sure yet but let's grab the image we just created play Green Tower three let me throw it in there and let's just pull this badge count down and just for fun see what happens now because it's gotten different input now which will once again adjust the output okay this did not work as well as I was hoping it would maybe we call it covered in Plants let's see what happens then so it's all about little bit playing with the prompts try limiting the control net strength I think that's also possible but I don't really um it's it's a little bit of a play right you can tweak little commands here and there it doesn't really matter um so sometimes you can kind of start the control net a little bit later give it a little bit of Freedom at the end and that results in more more freedom for a stable diff so so this is that tweaking part which I was worried about uh showing you guys I don't want to end up at a moment that you don't have a good good final result but it's all about just being open to and playing with it constantly think I'm not getting something I'm happy with yet but what you can do then is ADD weights to certain prompts so I was hoping this would get us there um blush vegetation overgrown maybe we want to and just by putting parenthesis around this PRP starting to lose the shape a bit too much you see here with that curve so I'm kind of wondering to push back on the the control net so as you can tell it's all about tweaking and finding and the the example I showed you guys before was pulled out of the process so you're never really um because it's little bit at the mercy of uh the stable diff Gods yeah L you're right it's a bit about a trial and error and sometimes it just doesn't care what I want doesn't want it to um maybe let me try to just put this one in the control net I don't know if it'll affected so much but let's see to birds I I'm going to try to get rid of New York right now maybe that's giving it too much of a too much pressure on the the building side making it too much like a real building covered going somewhere going to get rid of too much glass knowing that we already have a good backside we have a good surroundings we're only just looking for one image it can be the whole building basically covered in uh green but it just needs to be we just need it you know it can't uh I'd rather not try to photoshop it in give this about another three minutes and then you can push uh how Tower just go Tower emphasis on that see some raised hands indeed we could [Music] in indeed we could try to in paint but it just doesn't matter so much Tower covered imp plants let's try it it's all a little bit trial and error and trying to find uh um the right way to see Tower covered imp plants it feels like you want it it should know what you want but it's having a hard time maybe we need to add a little bit of trees there we go this is great here we go so just by grabbing that one gra that I think this is just a perfect example of this um let's get that Tower it's nice it's exciting it looks green let's get the surroundings from the other one [Music] and now it's also important to add a bit of this white just saw back here good to add a little bit of depth to the image to to you know show that we know so just by the O tool going into the the background image adding a little bit of lightness behind the tower letting it pop a little bit a little bit of dark n to the front it's bit too much the idea that it's it's a two-minute workflow I'm trying to get in here there it's maybe it's a two hour workflow or one hour workflow I'm trying to squeeze into two minutes here just to get you guys along with it so it'll never be as smooth as I uh as I hoped but I hope the the ambition is clear that you just try to get the image for somebody to understand the option that you're uh creating I think this is it's not great I'll be honest but we're getting somewhere and we can imagine that with this um crawling plants great climbing facade it's you can imagine that with um so just very quickly we've gone for something which looks like this a real Snapshot from an intern and we're able to transform it into something like this and put that on the screen and just discuss it together is this what we want what do we think of this and stable diffusion automatically adds in certain details like these horizontal stripes on the back wall there or the green on the roof or the the you know these types of little features which you can start to discuss then so it really grabs the conversation from a really abstract level this one where you know everybody can say I get it but in the back of my mind I'm thinking of a wooden Tower where um the green is only on the balconies here and maybe somebody else is thinking of this whole facade is completely gone you don't see any windows and it completely overgrown and by testing these things um we're able to um by bring by bringing these things into reality through stable diffusion we're able to move the conversation on faster and in a more structured way so so it's my yeah I don't know it's like a little fun uh first uh uh thing I would recommend first first it's actually the fourth one way that we can reimplement stable diffusion into our design process by this way and now I would like to ask you guys to do the same thing maybe will take five or 10 minutes I don't know if you've been following along but trying to do the same thing tweak one of these options and trying to create one design direction that clearly matches the concept um that you're going for in a clarifi clar clear way that clarifies the concepts I hope uh you guys could take about 10 minutes for that I'll keep an eye on that I see a bunch of questions that come in I'll answer answer those I'll answer the questions out loud as you guys are kind of testing these little things yeah indeed the the result so here I'm just reading the Practical and interesting nicely done for the time that was spent the result is very good and I AG agreed T it's the the funny thing is just the the the time spent in the results are just in saying the the amount of communicative power that these have are uh is really powerful how many people on a project team spend time with AI a dedicated or two to three people working together on the entire team so I can it's more of an MV question but um in principle we have teams varying from one person to eight or nine or 12 people and sometimes there's an assignment that we're all working on and that could be uh fixed by Ai and then people are doing that with AI so it's not like a dedicated person to AI or something like that it's really um uh really you know integral to the design process if you could do your graduation thesis project I think I answered that already can you provide a link to access stable diffusion I'm sorry I can't give you guys access to our stable diffusion because that um [Music] um but it's too much private stuff on there from from our office wouldn't be useful to you guys let me see are there any more questions otherwise I don't know I'm sure I think stable diffusion also sobody somebody asked if St stable diffusion also works on Mech um I think it works on both yeah I wish I could give you guys access to the stable diffusion on our server but uh it's not um not something I could do for you guys our computers couldn't handle all the requests people are using multi so are you using are you combining the images from mid journey into stable diffusion or using it as two separate tools I think in the design process M doesn't allow for enough um accuracy so what I just showed you is a way more accurate uh way of using uh stable diffusion so in the process I can't be showing mid Journey images at this point anymore that because um that's because let's say the programs not in but with this stable diffusion image I can be sure that if we continue um uh developing the project in this way that there is enough program in the building just simp simple things like that or we do have the maximum height see a lot of people asking about the Laura some people are training luras in the office ined but I am not currently using it so much try to keep it there's so much to learn in this I try to keep it focused you prefer mjor stable dip I'm going to give it three more minutes before continuing wait um I think like like I'm saying is mid journey is unbeatable in the first design phases but stable diffusion is just way more powerful later so it gives us way more control um do you see these tools as more related to the day-to-day office Life In your experience you share these studies with clients as well I'm curious how the F client relation could be handled there is um I love these tools AI tools and but they are just tools so um if you can get an image to a point that you are comfortable sharing it with a client because it is clean fresh realistic buildable we should be uh completely fine um doing that but you have to I would always mentioned that they are AI images just to be able to um you know reduce expectations and make it uh you know reasonable is it useful to integrate stable diffusion on grass over sure some people can do it and it works really well other people have a lot of hard time with it um I would really recommend the six workflows I'm trying to push short today I think those are just things can do tomorrow in in a way that you can actively integrate stable diffusion and mid journey in your real design flows um but um you know there's way more depth there's a big plunge you can take to the next step that's for sure just going to continue to the next one it's pretty clear or I didn't see anybody coming in with images yet it's pretty tough I think to work with this uh stuff live as you saw I'm trying to do it in five minutes and just doesn't always come out the way you want it f the screen so I think the next step that we're going to do and I'm going to do it live again let's hope it's a a more beautiful result this time and that's the enhanced massing studies I've kind of created in a rhino model which you guys have seen um which I hope has shared with you guys I think saw PA posting the links to those things um where there's multiple options within this rhino model and we're going to do the same thing again prep the input for Success generate tweak and finalize and let me quickly ah I was I flipped it over to the Rhino again here we go I believe you guys can see the old thing again here we go this is the one I'm looking for um so let's prep this one for Success because we want to make a cool stable diffusion image we're further on in the process we have now created a few options two options specifically which are both um very cool and open uh uh you know to be developed by the client however in trying to convince them we need to pull them more from this abstract abstract model that we have here where we can show many screenshots of and we need to pull them into the real world and we can do that through uh stable diffusion I think so I'm going to Quick go into options and I've shared a display mode with you guys it's where you can import it um enhanced Ming studies fuel Port diagram simple this is this display mode is just amazing and I really recommend using it it'll I remember back when I started we would be doing um vray renders just to to get to this quality of images and that's just you know doesn't have to be done that way anymore I'm just going to set up a quick viewport here uh a viewpoint aimed fuse save this one I'll call it perspective one we'll keep that one here for now and when I look at this image and I try to Envision stable diffusion what it's going to do with this it's going to have a hard time a lot of these lines are in in strange colors um another issue is is that the perspective is in perspective mod not two points so I'm going to quickly adjust that and update that it's because the the model that we're using is trained on a lot more two-point perspective models uh these buildings are uh uh purple or Lilac which doesn't work very well and then we have these sidewalks which are bright red these are fine colors to use during the process but don't really work now anymore so I would for a stable diffusion so let's just pick nice soft beige colors let's say this project is taking place in Lebanon or something that's what we're talking about a little bit more Stony a little bit more beige maybe our surrounding buildings are a little bit desaturated than our design actually take that same color and lighten it up a little bit make it fresh here we go I see that all the curves are white that's going to be something hard for the control neck to deal with so I'm just going to cell curve change all of those to Black this is once again just a bit of a um a trashy method to to deal with this but it works because I'm just we're just preparing it for um the the the rendering for stable diff so here we go this one and then we have a few people in and uh trees that are built into the file but we actually don't want those because they confuse stable diffusion and from this we go to few capture want view capture to file and then um let's do the a custom view Port of about a th000 by 500 and we make the scale four just so we have a big enough image for stable diff to work with I'm G to once again save this in our enhanced massing studies I'm call this city BL one so this is our base image the first one we're working with I'm going to go go right in here into Photoshop again so we're prepping um the image for Success um it's not a bad image in base but I think we need to enhance the park here a little bit add a bit of green to the top and give a soft color to the background maybe let's just put in a real Sky there just to give stable Biff something to work with trying to to enhance the chance as much that we get a good result so I just grabb the magic W to grab the sky um put um a mask on this folder that thing throw that so then I throw the sky into there working on a server it's not always as fast as I have it when I'm actually there there so here the sky looks okay I'm let me just soften it just so it doesn't impact the image too much I think now we need a grass or a field maybe something that looks like a grass in perspective here we go pull that out of the Mask region and I'm going to mask MK this green stuff here and just by adding these little things it's not necessary to to but it just makes it way easier to receive to generate good results tree PNG just the first one doesn't even matter uh pull that out again so this like the I don't know shooting from the hip type of workflow that I always really like because we're going to get results quick and dirty but they're going to be just good enough uh in the time that we have to to get stuff clear and communicated well to the client so okay something's going wrong down there just add a few two trees through trees maybe one up close just for the nice just to get a bit of depth into this image cover this building up front because we'll never get stable diffusion to hit that image as be as realistically as it will be there unless you start to do more complicated workflows but um T we in family Lebanon right maybe just do family in park PNG see what we're getting okay not there we go that's not a bad image just to get some a little bit of happiness into this image we'll just grab the object selection tool again select this family I think this one's just a bit too complicated for it all the stuff happening behind it this one it's also the same family actually very funny it's up here and I see I cut off the head of the the girl which just doesn't work but we're going to keep going by just just replacing it with this one so here we have the image prepped for Success almost because I want to add a little bit of love to the plint here I would think the even though we all understand it's something like a really exciting and inviting um space um it's not being shown yet so clearly so let me just quickly create see see my spacing is too high just quickly create a set of entrances merge them and I cut it open Just sh it down and then shift them out a little bit like this it just gives the a hint of some type of entrance happening that maybe not that hard it's not perfect but remember we're on this fake five minute deadline needs to be done in two minutes right so the idea that we're working under pressure here trying to get stuff done as quick as possible with the highest level of quality that we can get in the short time so I'm just going to clear it up by refreshing so so now our B image has been prepped for Success going to bit of tree on the rooftop it doesn't have to be exactly right but you'll see that if I just take the lasso tool just mask out this stuff this big stump importt The Mask unlock it shift it down a little bit starting to get some hints of little bushes just do the same thing here remove the mask quickly this starts to be you can imagine there's a lot of nice trees up there right that's what it kind of has to hint at as you're that's what stable diffusion needs to imagine there it doesn't have three perfect and these are now we've kind of created an image looks like it's from the 90s or something which is quite funny but um it's just an in between step in our workflow so here we go clean works at ARC realistic basic I'm going to insert the see a lot of comments coming in I forget something let's see so indeed we're basically just prepping the image or that's how I like to do it I'm sure other people do it slightly differently but we're prepping the image for for Success so that stable diffusion can fix it up for us in a way that we um you know bring it to the next level so let's go I level um photograph housing building next to park again next to Park beige Natural Stone is kind of the material we want to go for I think it should be a bright sunny summer day Coy something like that historical buildings I think that's kind of the context we're going for and uh Arch arched plint uh with shops shops Co let's just see what we get out of here and I already know I made a mistake this is completely wrong we didn't actually hit the candy so you can see the results not always great especially when you don't enable your control n there we go you see it's got more control than we had previously but because we want it we want it to have exactly this number of floors and we need it to have that and here we go the width and increase the width slightly sure it hits the whole image okay okay we're getting something but it's not there yet I think it's still taking too much of original image but you can tell already it's understanding these arches it's understanding the buildings back here I need to add in the city or something here look we can see the road is great here and the trees are fine the lighting is good to create nicer facad so R facade Windows let's see what it gets out of it not really I'm not really giving it words that what I want but I just imagine words that people have used to tag the images that relate to our project so this is not bad I think the so yeah this is kind of funny that we got here so quickly because I think this is actually quite good see I'm just going to drag it in here my workflow is bit messy and here we go I'm just going to drag it in here and just just to [Music] understand what we're looking at and what's good and what's bad I think when you look at the original image showing this second one to a client after let's say what was it now 30 minutes of work it's just spectacular the client will go crazy but we need to do more because conceptually it works up top we understand those are levels we don't really know the facade yet but there is some hint of it we understand there reopenings in it but the plint here still looks a little bit dead to me um I would honestly just fix it in Photoshop by a little bit of an orange glow let's see if I can quickly do that I wonder if we can then reinsert that image into stable diff and it will kind of fix it for us or if it needs to be more than that people happy with it but it's the the the funny thing is the how fast we can go from this to this and the amount of excitement you can give a client or a you know Studio director or whoever person your boss um by showing him or her something you know which is usually takes a while an image like this isn't something you get in the first week of rendering U of of the of the design process but now we can get there in first week we can get there day two if you want you know okay it's getting nicer I'll be honest but let's do small shops just let's just add trying to find an image where it looks like there we go an image that I can just throw in just put it over this orange stuff so it's not really there but it's a little bit there I can imagine a little market underneath here you know people having little food market quite nice because we have the control net from the pre previous um one I'm not too worried about I'm not too worried about little overlaps like this of course you can erase them and the result will be better but because we have the control net it won't screw it up too much you [Music] know so now you like the fact that I'm not using Photoshop AI I don't think the Photoshop AI is so great yet that I can use it reliably let's see if we can get an activated plint in here if not we could even use this image more or less so yeah you need to dive more into the plinth with um let's just do this decrease it a little bit act to like there's a little bit of stuff happening there I wouldn't be I would be quite comfortable showing this image uh in a in a client meeting so I think that's quite fun did we have an updated one of this this one looked a bit better right SC that up a little bit I think so now I just see that the road here is not so great but so you can imagine now that I would go back to that first one where I said the road was great let's [Music] see you just go in put it in we're not here to make a perfect export straight from stable diff or from it it's just impossible you're G to Cy yourself to sleep because you need that perfect promp and it just deleted that road when you were wanted that it's all about just I delete it's all about being smart with your time so here we go I think this is a nice image it's we can see a few key elements which we were intending to keep in the plinth is open fun exciting it keeps these arched shapes in there we've got this activated park where you see a family or a hint of a family it doesn't look great but we get it the rooftops are green and it feels like we're in the historic City Center it's not perfect but we get the hint we understand that it's not a modern shopping district or or something it works really well in uh to get us there um I think I will quickly go to the presentation and you I think you can imagine what my next slide will be because I would like to ask you guys also to take the time to uh to try this um by yourselves I saw the previous one didn't get that much of a reaction anyways I need to get a little drink of water but let's take 10 minutes or something you guys just try to play with it see what comes out I'll also answer some questions in the meantime um you have the files in the folder I believe so uh go for it let me screenshot this get that to you guys as well in the chat oh I see some already coming through that's awesome I see few questions you can always add a bit of detailing for the facade with the script indeed you could do that really easily um I'm just trying to show a workflow where we're going so fast that we don't have time for that you have the meaning in two hours let's go for it let's push this this crazy simple model which you know was made last night we have the meeting in two hours can we give it that final push that we get further so maybe you don't have time to write the grasshopper script or panelize the facade and add window openings this and that but we do have time to just knock it through stable diff um because the screenshots that we normally are able to share just won't reach that quality I'm just using stable diffusion the sdxl can you invert your shared screen your sharing your notes was I sharing my notes the whole time no uh I see assignment three channel in 15 minutes take one of the yeah I don't see notes I mean probably uh while you're switching to Discord maybe you can just share your entire screen so we can see the Discord as well sure show now I'm kind of reading is this which one I'm sharing actually let me sh the Discord here we go so here I see if no no I see your presentation right now the PowerPoint file are we good now yes great it's because I'm using I'm logged into a computer at the same time as I'm zooming on the laptop so I'm missing the green border which shows which I'm show sharing but uh that's fine that's fine right now I can see it perfect this is very cute whoever's making this but you can tell instantly that for example back here by not photoshopping out those edges it'll always affect the image and um stable diff will always think your thing is on a table or in front of something it be very hard to get rid of that without photoshopping and preparing your input beforehand do we have any interior examples I will not show you any interior examples I will only do exterior um but you can imagine the workflow is very very similar to the to the uh exterior workflow would it be a good idea to get the reference image from mid journey and tweak it in stable diffusion yeah it's really really fun idea I think that's also what I did on I [Music] LinkedIn on my LinkedIn post about this um returning to playing with Legos I didn't actually make the Legos I just generated Lego towers with stable diffusion with the mid journey and then put those into stable diff to get a lot of fun uh results let me see put that on the screen quickly for you so it's it's just a a a fun way to test yourself and you don't have to go out and buy Legos but it's I just yeah use mid Journey as a base here here you can see starting with this mid Journey image of a folded piece of paper generating a little house it's it's the they work well together these programs when you design the buildings are they typically made in Rhino or Revit or in a combination of these while using AI tools you got it right so uh each tool has a really clear um use Rhino is quick and dirty and will'll get you through a lot of stuff and C really high quality images uh and works well for in that way but it also helps you uh works well with Grasshopper and stuff like that and then slowly as you dive deeper into the process and the less things are open you can start to use programs like Revit which are way more powerful in exporting big sets of drawings and stuff like that and carries a lot of information it also works well with advisors at the same time but so so as you go in the process your tool sets changes from Rhino and Rhino and CAD maybe to or SketchUp and CAD I know people use SketchUp more um to Revit or rev it and Dynamo and just goes with the process but stable diffusion I think has a role and these AI tools have a role all throughout the process this one is cool which is this one I'm trying to think it looks cool not seeing anything's in here so I'll just keep my eye on the the chat how long would you keep trying before going back to your old ways of doing things I think think you're right the that is the toughest question when you're in a process you're in you have to that's the the hard thing you know you can make a beautiful render using vay or whatever but [Music] um um it'll take more time and but you know for sure that there will be a nice render at the end of the the journey so that's that's just like a little bit of a gamble currently where I'm still trying to convince people that it is the way forward and we are testing it and sometimes it goes wrong and you just get screwed over and sometimes it goes great and everybody is super happy but there's a little bit of a play there can it be used a stable diffusion from free from which the site can be accessed stable diffusion is free and you but you just have to download it and run it on your own computer AR Armen yeah I love this this is good this this is how you should use stable diffusion but it's it's funny though the the freedom it gives you cakes and stuff so but does other things Max aquat town can can you try with a building or a result that works well for um a result that gets would work well for let's say a client Nancy Herman I agree there is no going back humans must adapt now to the AI that's coming in keep track I'll give two more minutes and I'll continue I can tell the stable diff stuff is just not easy to do as a little assignment so I think it's fine too but but I think this one is really interesting to see as you can see it adds so much beautiful detail here it really grabs my attention but you need to still do the postprocessing so the pre-processing was good your stable diffusion result is good but you need to do that last step where you're still blurring out the background slightly you know you got to add that depth in here you got to pull your building out and separate it from the background here we have to do something about the if this is your concept right just a normal housing Tower and something about this big roof space which is drawing so much attention so I think just a little bit of post-processing got this image you've got it down you know still see some Sami dwellings coming in see Luke is asking what AI tool do you think is coming next do you animate your AI visuals I've tried animating the AI visuals it's super cool and it works sometimes but when do you do it when do you do when do you take the time to animate these things and and because it doesn't work on posters if you're giving a digital presentation it won't work um because you know you've got Zoom there's lag they're only getting three of the 10 frames it just doesn't work if you're there live it's always screws up It's just sometimes you you have to be better safe than having put in all of this time to get this beautiful uh render but um so I would yeah I would stay away from it for now it just doesn't work yet as good as we would want it to work the only thing I would want to animate is just a little bit of trees waving in the wind maybe and that's it maximum I don't think this one's coming out out of any of the images I shared but it's cool okay very cool let me pause the screen see one more question have you tried Imes from your own projects have you if we've tried retraining stable diffusion model using images from o project yeah we're working on that of course it's super exciting but very hard get this I'm going to share my presentation with to be sharing the assignment now again and go to the next slide so as we dive deeper into the process we you know less things are open and um but we're asked to do these really repetitive tests as architects in the process testing hundreds of different materialization options and I'm calling it materialization now but it can be different things openings balcony types uh where there's an infinite number of solutions um where 90% of the things are already set in Stone the amount of percentage open to closed for example the window placement is based on the program behind it um and and the floor to floor height the balcony types maybe 90% of the building is set but that last 10% is creating a lot of discussion um then you as an architect are asked to dive in and make 600 different options for materials and those things uh cost a lot of time so I think here once again stable diff can come in clutch and help us let me go share screen here come in clutch I think you guys can see my screen now um can come in clutch especially when doing those repetitive studies here I'm just going to go in so once again we start with extra tabs pop open sorry once again we go in and we we take this model and we prepare it for Success just like before so using the same diagram white VI mode go in set set VI set camera [Music] camera Orient to Surf sorry forgot the command for a second M few save it here and then quickly adjust it to twoo perspective I wanted just a little bit of tilt on it so we don't lose the realism here and I want a little bit of blue color in the glass so I'm going to select these two layers select object set object display mode to shaded and then make sure that the material that we're using here it's transparent enough so we can see something behind it this is more or less okay a little bit less transparent this one's still doing something I'm not super all I'm turn off this okay let's it's fine as a starting point just this dark color which is now a brown color will really affect the outcomes so I'm going to go to a nice light gray that will help us uh you know uh help stable def make its own in uh choices when choosing the materials and have more freedom and not make every material this dark brown so I'm going to go VI capture to file the hardest command to type te um go VI capture the file custom 4,00 500 again the scale for it just get a lot of quality there talization that's study here we have the first image and to PR it to for stable diff is a lot easier now because we don't want to add so much information but we do want to give stable diffusion a little bit of a hint of where the sun's coming from so I'm just lightening up this side of the image it's going to darken this side a little bit just with the O tool the Dodge and burn tool and then I'm going to export it again I think that's about all we need to do for this one then we can go straight into stable diff I'm just going to refresh clears out everything here at the realistic base image to image that study that we just created enable all of these things and cany I think for example now since everything is a straight line we could also do the MS mlsd but it just I just in theory you could do it but you can see some lines are just weird right now so I'm just going to go back to the Kenny we can try it with depth as well but here it's very simple prompt I think would be the work would be the best facade and let's do wood and this one you can tell it works quite well we have a wooden facade now um in 5 seconds vertical slats then let's do vertical wood slats that are very thin great some parts are working great some are less great great but it's it's getting the the goal uh across bricks let's see taking requests now so it's doing some type of concrete blocks now let's still uh shiny red um suram tiles so just it's it's insane how easy this is it's like not not reasonable to do it in any other way I would almost say it's doing something crazy it's not sure what to do there but you can tell this is working quite well and it automatically fills in some of the other details it makes the window frames red which you know maybe not what you're intending to do but it is something that's a cool result um let's just try uh metallic red let's go blue blue I think especially images like this um work really well when upscaled yeah this is just really cool uh I don't know how else to say it um I think the only thing I do here so let's just take this one because um the quality now is so low because it's so zoomed out I would go into the extras for example and use that upscaler so I'm just going to turn on the uler or the Aran four times but I just want it two big enough throw this one throw this one straight into Photoshop here and the only thing that's missing here in how I would explain it to anybody is that the balcony here just doesn't work as a it's silver so I would just try to let's see if it works if I just grab it here I don't think it will you can be optimistic just do something like this really quickly to make it clear that that's our ambition maybe I would just erase the in between things but this is just uh you know instead of trying to go through stable diff and um trying to fix everything so it comes out perfect I would just go in and do these little tweaks by hand just saves you so much time and gets the results done so export material study three PNG done you've just done six different options imagine having to do that in uh nscap or V or or whatever each time you have to find the right texture adjust it put it in make sure the lighting's fine it takes a few to to render and then you're putting them out you know it's just the time that you spend there versus what what we did now is just insane and the result it looks good it's not perfect but you can get the first discussion in and get the team aligned in One Direction um before uh you know pushing through and fully detailing it checking you know then you can so so then we agree we agree blue bricks in a gradient to the top right very cool we need that how do we do it we call up the tile people we ask what their sizes are then we start to model those things then you can start to realize it but you don't have to do that for all of the different materials we can just get their Ambitions really clear and then knock out the rest uh in one go I think that's uh I think that one that one went really smooth huh um yeah I'll give you guys a moment also to to play with that one let me put the assignment on the screen this is the assignment for just take 15 minutes the file is in there feel free to just test it take you okay I we won't do 15 minutes I don't think it's necessary um ni else came in in the meantime if you guys created some amazing stuff I can share some of the the prompts that I've been using just today oh nice so I see here you use I guess this is the Arc view Port um I would use the other one that I that I provided I think that helps a lot and here would you see very clearly is that the trees I put those in there kind of as a red herring uh because I knew people would do this if you put the trees in that nscap gives you or something or very simple ones uh stable diffusion just doesn't understand them uh you're going to keep trying to use them it just doesn't work back here it seems to work f here but these up close it just thinks it's a part of the building so good to be aware of that but it's good that you tested it this one is really cool did you SI I these are really good I think this one so so mark You're testing them out that's nice but you see the Arctic mode it's not giving enough input to the stable diffusion model so it's missing these lines so I in the file you can also install the in the file I provided you guys there is um a viewport mode viewport simple diagram and you can just by in Rhino typing options go to the Views display modes and imported in and you can have this diagram white mode this one works really great for stable diffusion but also for diagrams it just speeds up your workflow like crazy yeah so this is clear you see the control net so I would really use the this looks good too these are really fun it's fun to see what other people are doing with it you can see I think the the base that you're using here the model so let me see the model that Hugo is using here is way more artistic and artsy which means that the results will also be a bit more artistic and artsy and it's a little bit more free surprisingly similar to mind great I think you just use a screenshot from mine right that's what I'm guessing these are also very cool but you have to imagine so this one I would never be confident showing this to anybody the outside shape is just too crazy um it's also because of my knowledge of this more wavy uh architecture is just lacking I I've never done anything like this um so I wouldn't be confident showing something like this if it's possible or not the inside on the other hand looks amazing and I would maybe try to use it in one which looks straighter but very cool I list the models you recommend for St diffusion yes sure I did in the link that keeps being shared by PA Philip B you can find all the models that I'm using it's epic realism dream shaper eight and Architectural realism something d10 I don't know but um those three they work well in combination so I would always mix um the the what's it called I would always mix the architectural realism one with one of the other ones about you know 6040 65 35 something like that and that way you can uh get the results that you that we all think work well that work well I think especially when you just saw the material study it just works super well okay seeing people still love the the Sami one or maybe they took a break they put it on pause that could be it to but it's good to see um let me see if there's any I think these are also really good to see um I pause for a second and I'm going to try to start the last Biz communication one is this one within you guys can see my screen again without all my notes um but the final step and it's kind of what all of these are doing is they help us communicate with uh other people um where we are so used to using just these really simple massing models to communicate um they're often too abstract for anybody but Architects or people or even within Architects within our team sometimes even I notice that people don't always understand the same things from the same inputs um for us uh because we're all coming from different knowledge bases it's really important uh to communicate well with each other and I think here as our stable diffusion and really push us further as a communication tool so very specifically now want to give an example of that we're working with visualization Studios and often having to push them to render an image from day to night in the last three days it's literally what happens all the time here um because we just don't communicate it clearly so I'm going to run through that process in the way you can use stable diffusion to communicate more clearly with um your visualization company or within the team about what renders you want to make or the direction should be of the images I think I'm sharing the Rhino model now yep should be good um here we have a beautiful little stack building I'm set it to the same display mod as we learned the the architectural realism uh model is based in a lot of two points perspective so let's set it in two-point perspective then look at the model now I'm going to turn off these trees cell color right I'm going to go into these files once again let's say our building is a wooden building this time so slightly darker one it's it doesn't hurt to give a bit more information to stable diff Fusion especially in this stage I'm going turn off the masing on the inside the of the windows and these buildings in the surroundings they're just boring ones we don't want them to focus on that at all and now just grab our glass set object display mode mode shaded gives that nice transparency to the glass see there's no floors there turn on the floors which course put in there to the right there let quickly grab that it's a mistake I made sorry guys just grab it unlock it grab it on Gumball and just drop it right in there that's be perfect for right now [Music] and you capture it to you capture to file I'm just going to leave it this for now ai just F communication F come one so we're do the same thing again we're starting by preparing our model for you know optimal uh functionality within the stable diffusion area so let me just grab the sky again this one great well before have the wand tool I'm going to select the sky here folder mask it out put it in put it up and add a little fade out the back in white because it always gives a better effect just to give it a bit of a realism there I'm going to turn this one softer now I think all we need is a little bit of a take whatever you can grab from this IM this tree looks great see if I can just some it out doesn't work [Music] yet and delete all that stuff I'm contigous and turn that off delete that just need a few foreground trees and then I think we're there let me also put a little bit of a par with some people on the ground here I'm just going to get the LA one tool select the round here and mask it out of this get some sort of that effect I think so all I'm doing now is just once again prepping the image for your stable B get the most out of it we just need a few trees on the left here I'm just going to do again tree tree else grab the object select tool move this over do the same thing here it's not a beautiful image at all this is maybe the worst one so maybe we're pushing it a bit but let's see what happens when a file exports as going do the Fizz communication to then again just like we' done so many times it seems repetitive but each one just requires a slightly different approach a slightly different amount of prep work throw this in here control net turn it on again here you can do a lot more with the cany now and I think I want it to be off right at the beginning so here you can kind of adjust when it starts and when it stops listening to the control net think we want to give it a bit more freedom and then um just see high level uh level photograph wooden wooden turn basic see if we can get this looking okay without too much work okay it's getting some things right something's really wrong the ground is wood so let's do grass in a park seems to be our theme today in a park in a park everything's in a park at someth some think I don't want that so I'm just going to turn this down a little bit see and I think the image I put in here on the ground just doesn't work super well but let me quickly adjust that then I'm just going push this over stand eras it's a little bit trashy I would say how I'm doing this but it's about the result I think so I think that'll help us a lot here again tweaking the input here I was getting a little bit too too casual with it first we want to just get this base image done and clear and understandable for some reason he thinks it's a it's a lake so let's get rid of Lake yeah in the negative PR because the rest looks good right now so park is good the trees are good their Building looks fine don't think we're getting rid of this little sand patch for now but the idea here is now that we can for example say snowy snow day snow sunny day let's see what it does this one see it's trying to fight the green grass here a little bit too much but in this way let's say now we can just export this image and say hey do we want this render to be snowy do we want it to be a Autumn evening Sunset see you can just push the you know all the unclarity in the the images we can get rid of it and we can just show these beautiful atmospheric images the quality is low I'll admit there's a lot of abstraction here still but this is just helps us communicate with visualization Department summer day um without having to uh you know lose time there and you can see the it's just works I don't know it's really surprising to me each time how well this were summer day and let's say one more a rainy stormy night but we want it to be cozy as well you see how crazy this stormy just now see how beautiful that becomes it just works I think um I think that's pretty clear um I would like to ask you guys to do the the same thing as we've done now each time I'd like to ask you if you could also try for a minute to to play with that so as as again you know we prepped the input for Success add a little bit of activation here and there to generate it okay we see some things are coming in you saw the people that we put in the park weren't working well let's just paint them out and then we can finalize them I'm going to take a like a 10 minute break I think and then um we'll be interested to see what you guys have made and then let's uh that's it then I'm going to wrap up I have a nice little reflection at the end but this is very exciting guys I'm looking forward to seeing what you guys can thank you guys let's go for a break and 15 minutes we'll be back some of these are looking so good so tell us what the prompts are that you used here I would be interested also I think um the people that you hire to make these renders these beautiful final images it's not the same it's just a a quick click and it's done uh render they it's about composition getting the right things in the right people the right you know it's it's there's so much more detail in making a final image that stable diffusion can't yet compete with that I think the that's what also what you saw at the beginning for example in my presentation those images that were made um maybe you know now knowing that that's the final result stable diffusion can recreate those but it could never hit all of the conceptual markers to such a detailed level that we would be happy with it so I think currently there is not a replacement yet for like a visualization Studio by by using stable diff for example but so yeah you're right in the end we would still have someone make a render um cu as you can see the control is just you know sometimes it's just slightly missing like we want to show that the housing here is single family housing we want to show here that these are luxury apartments which are full width we want to show here that this is a 247 uh disco or something you know just the freedom and that's something that we wouldn't want to wouldn't be able to do or wouldn't couldn't do in stable diff I think also when you look at renders like done by Mir or something that the depth and the quality is just Beyond uh what stable disc can do currently currently it cannot be used to create a 3D model is it possible same building from different angles yeah so that's the the thing I think once you have this one render set up you could for example lock in the the seed and just move it around and then rerender it and lock in the seat and render it and I think it'll be fine uh it won't be perfect that's just stable to it's not perfect yet but uh we get there it's workable I think yeah so indeed it is possible to use this material ID I think it's in stable if it's a different type of control net um tile I think or segmentation I'm not sure I just don't I use it so much I think it's a different input or some something but then you um can literally give uh stable diffusion the information like hey this stuff everything that's blue has to be glass everything that's green has to be grass everything this and and it has like color code set up for that so there is a way to to do that indeed it's just super time consuming um and the result is really good but it's super time consuming and that's something I would try to avoid just by looking at your image Mr Mr L I think your CFG scale is way too high or your control net is too detailed I'm not sure if you want to share a screenshot I can look with you so here you can see the guidance strength is too high I would set it back down to to seven I think then the result will be better and the prompt strength you can do uh8 damn some of these are looking so good it's crazy it's you can imagine sharing this image with with the render company or even inous saying hey this is what the vibe of our building is about it's it's a wooden stack of boxes each with a different uh you know U material slightly and it's next to this Lake it's beautiful and active and even in the winter people will come out here you know this image just speaks to uh oh sorry good that you're letting you know this image just really speaks to the speaks to your imagination we're not using image to 3D so much or I haven't uh I don't think it's working yet to a level that would help me help us in the design process currently sharing the Discord screen I hope it's all right now we see that disc score the screen uh I think it's all fine if you switch to anything else yeah maybe you can just switch to the presentation or something sure I can do that just thought some of the work is it's being shared is very nice especially by AO some really nice ones so just because I do I think I have about let's take three more minutes I am very interested in how you and try how you try to do this what what control net did you use I think I won't we won't dive into this so quickly but if I were to do to change this image into a summer scene would you insert it also insert it into the control me and add the depth filter and then just put summer in the in the prompt and I think that should work if you're using the correct models um but I'm so that's also the issue jump I think you you don't have a control net enabled currently this is creepy okay guys I think I'll continue is that okay hit really remember when we said it last okay I think I'll continue is that okay can you hear me am I M yeah we can hear you let's go perfect let's go um and I think we just hit all of these six key points and I hope uh all of the people here which have been a huge amount of people now 320 still listening which is amazing uh have hopefully learned a set of skills which they can start to apply uh even tomorrow and um you know just stay 20 minutes after the meeting or or at the end of your day tomorrow knock out some of these stable diff images and see how they can support you in your design process would be interesting to show those around I think it's a extremely exciting uh moment in the in the world of architecture where not many people have this skill set and it's not that hard to use and will really enhance your design process um you know from the initial Inspirations the high powerered reference generator the reintegrating uh architectural foam models into the design process as well as um enhancing your Ming studies to the next level and materialization studies it's all ways in which we can uh enhance your design process and allow you to efficiently use your time so you can spend time on the things that really matter um these things should be used in that way and now you can for example take the extra time and go sit to go go on a site visit and don't just stand there for two minutes and take three pictures and keep going but take a few hours because you can do all of the production work faster now take the time to dive into the analyzing the location and take the time to do a real contextual analysis don't just you know rush over those things I think the what I want to push for I want to preach a little bit about this AI integration it's not a tool that'll just put more pressure on us as Architects it's a tool or or raise the bar I think it's a tool that'll give us more time and allow us to do things that we're actually trained for and that's to be good design so that's my closing thoughts a little bit like mid journey and stable diffusion are extremely powerful tools and they're super hyped up right now and we should you know embrace them and love them but don't forget that they are just tools they're part of your like SketchUp and Rhino and Cad and and Grasshopper And andscape and whatever you're using in the office they are just tools that um part of your tool belt but make it part of your tool belt and in state it is an extremely powerful tool and a remixer or an enhancer that'll uh you know Blast Your design process into the next level and it will help you create extremely visuals that will support your designs but don't forget that architecture is about the real and it's about the build it's about these sketches where ideas go to the next level architecture about humans architecture about Heritage and culture it's about the past and the future it's also about creating spaces that we we all um want to live in and that we want to love uh that we eat and enjoy and we live our lives and and um these tools are EXT extremely exciting and fun and they're really you know really but don't forget that we're here to do uh architecture and their means to our destination so I really want to thank you guys for taking the time to listen to me today I've feel like I've talked for hours and I have actually um I want to thank PA for inviting me and allowing me giving me the stage to to talk to all these people and you know share this little bit of knowledge and uh it is so much in development and I'm really excited to see what all of you students and and other Architects and whoever else is in this call will do with it and feel free to share it or send me a message on LinkedIn or something I'd love to see um where where this this is going and uh it's an open discussion we as the world now are trying to figure out what to do with AI and it's not really clear yet and uh so join be a part of this discussion and don't be just a listener and also the talk and would be really my last thing so thanks a lot thank you guys so much it's really been a a pleasure thank you so much C this was really brilliant to see the workflows and also the way that you're putting things together I think stable diffusion and M Journey are great tools and uh there they're we're we're waiting for their revolutionizing the design process and the workflows that we're having so it was a brilliant session to have you in here uh we were super excited to see all these amazing works that you guys are doing and this was the moment to uncover them all it was not quite Qui enough I think we need to bring you back again in the future so uh with maybe a more extensive or comprehensive Workshop so yeah appreciate it thank you so much and days for my voice to relax yes yes I can imagine I can imagine well it was a unique session uh and appreciate your time so guys uh the video on the YouTube is a ailable for everyone to watch uh it's going to be there forever and also we're going to take this video added into your courses inside the course section on our websites so you can get a certificate of attendance with that so uh also don't forget to join the Discord the Discord will be active and it's only for members and whoever wants to join and be a part of our community and uh don't forget to share to join our membership program uh it's on our website and uh check out the future workshops that we're having with uh prominent names and Cass is going to be one of them as well in the near future and uh also check the discount code that we have on the website it's available for the limited version of the courses uh selected courses let's say it's a 40% discount and other than that yeah we look forward to having you guys back in our Academy again and I appreciate your time cast for joining for this precious 4 hours and I I feel that you you are also tired but we're GNA have you back with a much comprehensive format of a workshop and yeah until then uh hope to see you soon cool thanks Amit thanks guys thanks everybody thank you so much and see you soon goodbye bye everyone bye bye okay thanks everyone uh this was a really great moment to learn about the workflows in mvrdv yeah looking forward to see you guys again uh in our upcoming events make sure to check the our digital membership program and also the future workshops and upcoming events that we're having we have a couple of membership workshops that are beginner and intermediate level so hope you like like it and yeah see you soon goodbye

Channel: Parametric Architecture

Views: 21,081

Keywords:

Id: FCKhn42cxHk

Length: 203min 17sec (12197 seconds)

Published: Sun Jan 21 2024