Workshop: Getting Gendo Right — Consistency & Control (with Edward Meyers, Settle)

Perfect Hi everyone, we'll just get everyone through the door Excellent. Well, hi everyone, great to have you here today. I'm Joe from Gendo and today is all about helping you get the most out of Gendo. Whether you're brand new, or you've been using it for a while and keep hitting the same walls. So we've gone through hundreds, if not thousands, of questions from past workshops Built this session around what architects ask for most. How do we keep consistency across views, stopping things from changing, holding image quality through multiple edits, and building Gendo into the process from day one So, George, will Ismail from the Gendo team will take you through the fundamentals right through to more of the advanced stuff And we're very excited today we've got a special guest, Ed Myers, who's co-founder of Settle. He used Gendo with his co-founder across an entire project stage, replacing their Unreal Engine and Photoshop workflow, cutting visualization time by around 80% per image But he'll share why Settle chose Gendo, how they went from SketchUp screenshots to final presentation imagery, and what they've learned through producing a competition mission in Gendo as well. You're also going to get an exclusive preview of what's coming next. Really exciting, including custom styles and project memory And of course, an exclusive offer. So maybe just before we begin, couple of quick housekeeping rules. This session is being recorded. We'll share it with you afterwards, so no need to worry if you miss anything. We've got plenty of time for Q&A. So please drop your questions in the chat as we go And they've always let us know where you're watching from. I can see people saying hello already, which is great. We'd love to see where everyone's tuning in from. So with that, I'll head over to George to kick off. Thank you, Jo. Thank you, everyone. Pleasure to have you again and to be doing another webinar. I'll just introduce Gendo briefly for those of you who are not totally familiar, but Gendo is a browser-based platform for architects and designers to leverage the latest in All the crazy AI and ML technology that is emerging around us. And our purpose is to tailor that, productize it, train it to be best in class for architecture and design, and in an interface which is very easy to use, so that anyone in any stage of design can be leveraging it across all sorts of types of architecture. It's free to try. If you haven't already, please log in and just have a go. You're more than welcome to sort of play along with us. We'll share a couple of links in the chat as we're going as well, if you want to jump into the boards that we're demonstrating in two So to introduce myself properly, I'm George. I'm one of the co-founders of Gendo along with Will. My background is in architecture, so I studied here in London and worked in a few architecture offices doing mostly residential architecture from individual family homes up large town expansions And I got steadily more and more into viz in the traditional sense, sort of 3ds Max and V-Ray before then meeting Will and co-founding Gendo. And with that, I'll pass to Will Yeah, thanks Robert, George. And thanks to everyone for tuning in. Really excited for this one this afternoon. Yeah, my name is Will. I'm George's co-founder. My background's fairly traditional software engineering. Worked in a few different types of company, but I always had this special interest in applying technology, specifically machine learning, for image production and design purposes. Hence, meeting George all those years ago now, and talking about the new generative AI world and how that could have an impact on design in general. That's where it was where I was born, and yeah, ever since we've been building Gendo and I've been running the kind of technical side of the company since then. So yeah, thanks again for tuning in. I'll pass to Ismael. Yeah, hi everyone. Great to see you again and hello for the newcomers as well. My name is Ismail. I'm the head of Applied AI here at Gendo. Like George, I've been an architect for 13 years before joining Gendo earlier this year. Over the past three years, I've been really testing practical AI workflows for architecture and design. And this is why I joined Gendo I think that this is, like, the best fit for me right now, and I'm excited to like, do some live demos with you guys today, trying to do some consistent design view iterations. So yeah. Awesome. Without, yeah, we will pass it to Ed, I think, yeah. Yeah, yeah. So yeah, just to say that as the title suggests, we'll be focusing on some consistency and most of this session is born of the, as Joe said, the questions we've received over the many months about how best to get stuff out of the platform. But more excitingly than our sort of demonstration, we have Ed who has used it in real world context on an actual project And can talk us through his process, his design process, as well as then how Gendo was a small part of that. So with that, I'll maybe pass to Ed to introduce himself and take us through it.

Yeah, hi everyone and just again, thanks to the Gendo team for putting me on the spotlight here. It's an amazing opportunity and an amazing platform that has really helped our workflow as George was saying, but just a little background on me, I'm from Australia originally. I worked there for six years and became registered there And then made a leap to the UK in 2018 to study at the Architectural Association And did some more sort of computational robotic kind of work and workflows in the DRL And then from there, I've worked for a couple of companies in London now at Hediwick. So I'm kind of on the next chapter now, and working towards building a small studio with a great friend of mine who can't be here today, Andrew Friedenberg. We're doing kind of small Timber-focused architecture, anything from a small house to a bigger hotel all over the world. So that's me, and that's kind of where I'm heading Yeah, again, thanks for the opportunity, guys. Really appreciate it. Okay. Perfect, I'll take this one up George, I'm going to share my screen and go straight into it. Thanks. Yeah Okay, great. Can you guys, can George, can you say that one? Okay, great. So yeah, as I said, sort of we're a small studio, just two of us, basically a startup. We focus on smaller residential and we've kind of jumped into the world of glamping, camping, small cabin design And how architecture and NYCHA can kind of become one over the last couple of years, which is from both myself and Andrew's background, we're from green outbacking places of Oregon and Australia. So we we we thought, okay, let's put our skills at making beautiful images and beautiful buildings and pair them one with nature. So this is kind of a A selection of work that we've done over the years But more in focus, I guess, on the project of highlight, which we use Gendo to really excel to another level. We couldn't believe. It was an amazing one, was a competition in Sicily near Mount Etna, which this photo was taken We went on the competition earlier this year, and we basically had I think the full-blown effort was only about 4 to 5 weeks to really pull this thing together So the timelines were tight. But the the passion was high, and we had a great tool at hand. So We had a really quite stringent brief, though. This is the thing, it was a very real-life project. The client was very keen to get a result that they could build, get a result that was affordable, and get a result that would ultimately be economical for a small hotel startup in… in Italy called Open Air Suites, or they're… the guys at Open Air Suites, again, are amazing. Please check them out They are beginning to go on a venture of doing open air hotels, so small glamping to cabin to bigger kind of Hotel style applications in all over different parts of different rural areas of Italy And they're one kind of message is to really provide a place that connects people with nature, and we kind of sat between that to try and deliver a strong architectural language Which could be rolled out across all their sites, across Italy, but have a very, kind of, nuanced, local, approach to each one of those sites. So, a part of the architecture that, I guess, could change to be recognisable of that place, from Sicily to Tuscany up to the Alps, we wanted to develop, I guess, a prototype that could fit all those but be somewhat changeable So straight away we looked at the site. I mean, Mount Etna, I mean, it's an amazing case study area that they have. It's part of a winery Just in the base of Mount Etna on the western side, it gets amazing views, so there's not much you need to do. I think architecture kind of needs to get out of the way a little bit, you know? There's a lot already there, there's a lot already to see And I think our thing, or our architecture was about framing that, hence the name frame, which is what we call the project in the end. It was quite simple. And that kind of came as well from an amazing history of agriculture and architecture in the region as well, which is very kind of heavy stone masonry buildings that have these insane kind of vistas outward all over the landscape. I think that was

You know, the ideas were already there, we didn't have to do much. It's quite an amazing place. And I think there's something else about Italy that you all know about, which is food and their production of food and the way that they've done it for many thousands of years. Sicily is one of those places that produces an amazing citrus and wine, industry, and part of OAS's venture is to really get those small kind of nuanced agricultural industries and put them on show in their suites. So that was kind of one thing that we started to work with So we started without a computer, as always, it's just pen and paper, ideas, plan section, elevation And really started to digest their brief, and really digest what they wanted, which was… which was very much in the DD level of design, i.e. they… they knew how big this thing needs to be. It needs to be a rectangle, it needs to be made from this, it needs to be made from that So we're like, okay, where do we… where do we fit in here? Again, we just looked at these guys and we're like, okay, that's pretty simple. It's a bit of stone and some some beautiful interlocking terracotta roofs. And guess what? The landscape around it makes that place, so let's not try too hard here, let's just let it happen. So we gathered ourselves and we gathered the plan that they gave us, which is 4.5 by 5.5 meters and basically created two living opportunities or two caverns for a small hotel, one cabin being a deluxe, one cabin being a standard. They're both for couples Or two people With a small dining area, study storage area, bath and sleeping area, very simple We we just took a basic approach of compressing them through the entry and putting the utilities at the back and opening them to the view of Mount Edna. And then those series of functions in between themselves and Mount Etna would open up as well to Mount Etna But there's one thing in the way of that The architecture itself, what do you do? How does that piece in the middle Respond to yourself and the landscape out there. So we started with a box and then we began to add these apertures on top of the box. Which are a kind of direct… had a direct relationship with their use behind From anything from a kind of corner dining area and space in the interior, which could open up and we could sit on the edge of the facade to a private kind of window, which would align perfectly with the bed when you're lying down, you could look out into the landscape to entries and exits of the buildings as well. And at the same time, we created this depth that we're really kind of intrigued by in the kind of heavy masonry buildings of Sicily. We got a building that could be shaded. Sicily is very, very hot, very hot in summer at this time of year. How do we protect? How do we make it comfortable? How do we do all those things? The architecture isn't just an identity piece, it needs to perform, it needs to be doing Something for the comfort of the interiors. So we got to the kind of this level of planning which we're happy with It worked. It was compact, but somewhat ticked all the boxes. We kind of got there. But for us, the facade and, I guess understanding how these interior and exterior Relationships could be viewed through was really important, and we… what was really important was understanding how they could be built in a quite efficient manner. They're an OAS or a startup, they're not looking for anything I did crazy design. They just want something that works, that's subtle, and again, that respects you know, the nature around it. So very kind of simple diagramming and work here. Again, this image here, thanks to Gendo. I'll show you in a minute how we got to this. But yeah, from anything from bigger visuals to detailed sections we kind of… we really start to flex that in Gendo as well. It was something that we were surprised that could work very well. So we created a number of different units. And again, I think the beauty of this was how the kind of apertures and facades could react to different landscapes, different types of agricultures And now I'm going to go into some of the videos. Again, a great feature and something that really took our visuals to another level in Gendo and something that I've say you guys need to use as much as possible. I think video editing and video production is such a… well, has been such a lengthy kind of Adobe Premiere Pro, kind of stitching things together sequence, but now this has just taken one of our images and really brought it to life

And just to see and really feel these spaces and the different times of day, and add that level of movement for us just brought so much more life I think this, yeah, the client was very, very excited when we showed them this. They couldn't believe it. And as I said, being able to go from an image to this was It's quite incredible. Yeah. It really sets the time. So That was that. Oh, I've got more. But there was a there was kind of an amenity component as well, which again We designed in conjunction with the units and the cabins So just a little bit on workflow. So that was kind of the project in a whole and the concept. I think it'd be good to say like, this is the kind of modeling that we do and do day in, day out at our workplaces. And this is basically Not Easter eggs, but a bunch of trees, I think, with a mapped mesh Probably landscape, and then a bunch of units that we've thrown together quite quickly. We've put an indicative path in And we've taken a runner screen grab in render mode, and then I think with I'll jump to the Gendo board in a minute, but I think we got this in 4 images, and I think I really want to say the We didn't have to put any of these in Photoshop. It was a kind of iterative process of taking that model and just doing an image, slightly changing our prompt, slightly changing an input image, doing another one. It gets better and better. We're not losing quality. I think this is, like, something really powerful with the app that we Again, it took our word to another level. Again, I don't even think this is render mode, but really quick, we're like, hey, we need an elevation. Okay, let's make it moody. Let's put a dog in it. Done. You know, I think this one was three images. It was… this was probably one of my favorite images. I think the way that it captures the rain droplets is beautiful. We prompted that in Again, yeah, it was, it was, it was in the tight time frame, I think we did this, you know, one three days out and it was probably one of our favorite images. So for the exterior, like, I think, you know, this is an object in green space, and I think we you know, it does look beautiful. When you tie those two things together we thought, okay, let's do this exactly for interiors, but we need a little bit more, direction. They, the client specified different types of interiors they wanted. They wanted a plywood fine. So we just took this into Enscape quickly. And again, I think we even got better results just putting that up Little bit extra time I'm just checking in, Scott and putting a few materials on it. It really, really came out well. Picked up the materials of the cushions and other accessories as well, which we really enjoyed. Again, the landscape component, I'm just writing Mediterranean planting is what you see there. Under Mount Etna, it's just picking up exactly what we want. We want, prickly pears. Perfect, you know? Again, and even at a finer detail, okay, let's just let's again, this is an Enscape shot. Let's really get the touch and feel like what is the bathroom made of? What does it feel like to roll your hands across the side of the wall here? We wanted a basalt kind of volcanic stone And this is kind of the result we got. We wanted a raw kind of brassy look with the tarnishing We wanted that, we wanted the tail, we wanted the kind of small kind of ASAP bottles. We got it all. And I think this one was done very late in two images. So That's a really quick roll-through of the project, and I hope that made sense, and I hope everyone enjoyed that. Again, thanks Thanks for you guys. We're settled design. Please hit us up on Insta and have a look at our website as well We've got lots of upcoming projects which are really exciting. So Cool. I'll hand it over to the questions. Yeah, I mean, awesome. Thanks, Ed. I mean, beautiful, stunning, delicious, all the additives you could choose I mean, you sort of touched upon some of your Workflow bits there. Maybe it's quite clear. I mean, I guess I'm interested while some other questions come in to know what the difference of using Enscape rather than sort of Rhino Raw did for you Yeah, I think Definitely mapping materials and landscape. Obviously gives a bit more direction to Gendo when you definitely want a you know, plywood interior, you'll get a plywood interior. And I think what we loved was that we were even getting down to iterating grains of plywood interior in Gendo. I think

The one thing that that did was just jumped a few steps. We didn't have to deal with, like, looking at passport interiors. It just went to… went to a plywood interior, and we could really focus in on the detail I think the one thing that we found and I can share, if you want me to share my board, I can Yeah, please, yeah Sure, Chuck, if you're going to do it in Okay, there you go. So that was that first image I showed. I mean, this is a rendered rhino mode. We got this one and we thought, hey, let's just make it moody And then we said, hey, let's just delete that guy He's still there And then we've got the dog. So I just want to reiterate on what I was saying before. I think, yeah, we got that in four images. That was great. On the Enscape question, which we use for actually I'll just go to go to the other project board for Interiors, bear with me We We did that just to guide the materiality a little bit better. I think what we found was Previously, we've always put the line weights on full to read the geometry and had no problem, which was great. As you can see, this is the first pass. It was a bit confused with the fireplace. We needed that as an iron Yeah, it got there in the end. We got some little additional features which we thought were cool as well. But I think, yeah If you've got a bit more of a solid brief Yeah, we kind of Needed to just give a bit more direction, but yeah Yeah. Yeah, nice. And we've had a little flood of questions, which is awesome. Some of which I'll just skip because we'll touch upon them later. But maybe one is about the Depth of prompting and describing the environment of landscape, like, how do you think about prompting? What do you say when you're asking things, how do you phrase it? Yeah, absolutely. I mean And for full transparency here, by the way, guys, I mean, there's our prompt. It was agritourism cabins. I mean, by the way, we put these prompts through court or something like that. We actually thought about them. And we were as descriptive as possible without being over descriptive. It sounds weird to say that, but You know, I think there's a level of and Gendo team, you would definitely know this from the inputs that you get. There's definitely a level of detail that's enough and too much and not enough I think, and I think the beauty of And this world that we're jumping is, is how much control the architect, or how much authorship the architect has versus how much authorship the AI has, and I think there's a fine line between really getting it right, and I think it's not just the prompt, I think it's definitely your… your reference images need to be on point. You really need to know what you want to produce I mean if you need to know, understand the environment that this takes place in. I think we were very surprised how ultra specific as well Gendo was at understanding that this was under Mount Etna and that it kept us putting that little smoking volcano in the background, which we loved So I guess, yeah, I would I would just play with the tool. I mean, it took us a few goes to really understand, you know, what sort of, like, constant prompts we needed to put in just about we put these in our Gendo team may be able to answer this. I don't know if it does read this, but we always put them in We do this at work too. Never change the camera, never change the geometry. Yeah, I call this prompt anxiety. Yeah, I have it as well. Yeah Yeah, exactly. It probably does nothing, but we do it anyway. But I mean, yeah, you can be as specific as you want, but I think something like this is I guess it was enough for us. Yeah. Again, it took us a few goes. We went too light at the start and too heavy at the end and then we kind of found a middle ground. We saved our problems off And then use them more and more So yeah Yeah. I hope that answers that one. Yeah, yeah, yeah, clear. I mean, there's a couple of questions about achieving sort of details. I guess there's a couple of references to like architectural details, but I guess you've also mentioned some materiality and some of the like finer sort of layer of things. How did you Yeah Yeah So in terms of details, I mean For example, I'm just going to zoom in on this one, the hardware on the frame is something that we did not prompt and has come through Again, hats off to the Gendo team. This adds a layer of realism to these images that is very appreciated by panels and clients that don't completely know what happens in the AI world. So it's you know, it's adding door handles and things. Again, we're not prompting that. The lighting We had a light, and it gave us some other pendant ones that are actually floating, but it's fine. Again, these things were… were happy accidents, I guess. And things that we rolled with in the end, so… Yeah, again, if you want them in prompt them. We never did, and it just read our geometry as a door, which was great.

I mean, sure, there's a level of modelling. I think this opens up another question, is like the level of modelling and also the the scale of modelling. I assume that Gendo is picking up an image here and knows that a door frame is about that big, you know, 2x4, whatever, and it's looking at my image, and it says, okay, that looks about 2.4 proportionally in that view And I'll add a bunch of hardware to it. So I don't think there is jumping in modeling if you want these kind of results, I think you need to model them well to get these kind of things To work in terms of details. Yeah Yeah Yeah, yeah, yeah, yeah, awesome. There was another question about reference images. I mean, you mentioned finding the right one. Maybe there's a couple of examples, maybe some thoughts about what helped, like what kind of image did you use and for what reason? Yeah, I mean, a lot of the reference images are, I mean, first off, we try and find an image close to what we're doing. This one I can see is a kind of timber cavern raised off the ground in a forest, late afternoon sun. Like, as I'm talking, they're the things that we kind of wanted. We wanted a timber cabin race off the ground in afternoon sun with a kind of pine tree, kind of dancing around, and we kind of got the result. So I think again, be specific with what you're looking for and make sure that the reference image expresses that We tried as much as possible to use real-life references rather than other renders. I mean, sure, you could use other… there's an amazing rendering companies out there that really bring them to life as an art form. And if that's something you want, sure But, I mean, we were going for quite a real life output at the end of the day, like all our images are basically So yeah, we search long and hard and built up quite a library of reference images. Oh yeah, there's that one there, for example. So Yeah, I think just This is actually a render, but I think a lot of ours were based on as well, lighting, atmosphere and flora and fauna and the nature that the building sits in. But yeah Yeah, yeah. Yeah, clear. I mean, I guess the sort of killer question and theme is about consistency, right, between sort of using across the project. And I don't think anyone could say that these aren't very consistent and obviously the same project. But how is your sort of experience in maintaining that across Multiple views Yeah, it's a really good question. I think it came down to the prompting Definitely, there was a consistency in what we were prompting. So it was always… there was always a few words, like utilities lee, Mount Etna, Mediterranean So that… that was always used. We had a little kind of library of prompts, and every image, we kind of put that into that to just keep that style. You know, you put yourself in a client or a competition panelist shoes. They want to see that That it's the same project. Which, again. cuts out that middleman of Photoshop, right? Putting the same filters and masks and etc, etc. on all the images to make them look the same. It did it for us, and I think it was only a sentence of text that really drove that home. So Yeah. accurate, consistent, discipline prompting, I would say. Yeah, yeah. Yeah. Yeah, that makes sense. And then one more from me is, what would your advice be to someone who's touching Gendo for the first time today? What would you say is the sort of most important thing to know? I mean, honestly, just dive in, iterate as quickly as possible. Don't be too careful of, you know, and cautious at the beginning stages Set up your models as detailed as you can in the time you have. For example, that… something like that And just go for it. Really experiment with the prompting, build up a library of reference images, and experiment with… with atmosphere and lighting and everything that you want in an image Do it quickly and as fast as possible, because that's the only way… I mean, we never use Gendo 3 months ago either, and we just had to dive in. We've kind of forced ourselves to do it, so I think the answer is don't be scared and iterate as fast and rapidly as possible because you'll find patterns

Which will then get you consistency and control in your results, and I think that's when it becomes really powerful, and, you know, you can… you can produce these images in you know, 10 min. It's pretty amazing. So yeah, just go for it. And enjoy it. It's a lot of fun. Awesome. Thank you. There's been a bunch of questions in there, but some of them relate to what we're talking about later and a bit more sort of product-y. So unless there's anything else from the team before we Release your head back into real work. No? Okay. Hold on, Jake Maybe… maybe just one, sorry, just going through, because, yeah, maybe just to mention we'll get back to everyone if we've missed any of your questions, for sure afterwards as well. But Ed, I mean, it's a hard one, but I know in obviously the case study, we've got you kind of put an estimate to time, but is there any way you can put a rough estimate in terms of the time saving towards the client presentation kind of using Gendo Yeah, I mean, look, we kind of said it was about 80% in saving. I think Yeah, to be honest, I mean, it was a four to five week project, but I think really the last week and a half was… was pivotal in getting that image generation out, and I would say the last week… I mean, even the night before, or the day before the presentation, we were still tweaking lighting and things. So I think I would say, look, it was a five week project I think if we didn't have gender, it would have been another Yeah two, three weeks of rounds of images you know, getting lighting, we would have taken it to Unreal, and we would have done a kind of Twinmotion pass, and then The lighting alone in that takes for one image is a day to get it really, really right, you know, we might spread it over two days or something. So, yeah, I would say On image time, we're saving about 80% Yeah I see. Yeah, amazing. I mean, maybe a sort of a question to expand on that. Do you think your design process was different for being able to iterate languages that quickly, or were you still sort of designing and then sort of producing an image at the end in a sort of more traditional sense? No, I think it was it was I think it did change our design So a lot because what we did is, instead of just designing and front-loading the design and then front, you know, end-loading the image we did design image design image design, image design image. And then we've got a bunch of results. And what would that that would do would just like save time down the line like, oh, we know that already doesn't look great. Like, can it, let's go on to let's run this horse, you know, run that design. Okay, great, let's have a look at that. Image, image, image. Do you know what I mean? So, I think it… it made us produce more accurate, final images far earlier in the process to give us feedback on, you know, what these designs will look like. And For us, it's just quicker feedback. We can get to the solution far quicker, we can actually, you know, in a short timeframe, we don't have to just grit our teeth and go for one option, and then produce, and then do the image at the end, which might be the complete wrong option. We can And on quite a lot of design options quite early on and test them and see how they really look and feel. I think that's the big big jump here is, like, how they really feel, like, what their… what their atmosphere is… is like, and, you know, at the end of the day, as architects, we're trying to, you know simulate what it is like to be in these buildings, right? And if the closer we get to that at the earlier stage, the more successful we are I think We yeah really changed our workflow and it gives us more time to think about the nuts and bolts of the design and get that right and the planning and the other things because like you know Sure, we love spending hours and hours on images, but it has freed up the designer to… and actually, to be honest, is going to free up the discipline To be to empower ourselves elsewhere. So Yeah, absolutely. Yeah, well, amazing. Yeah, agree, agree entirely. I think it's a lot of pointless technical work saved and then much more room for design decision making, you know, which is actually what matters. Yeah Exactly. Yeah. Awesome. Well, yeah, thanks again so much for taking us through. It's a beautiful project to see and thanks for taking us through in so much detail what you're up to on Gendo. No worries. Cool. So maybe Cheers. Bye-bye Thanks, everyone. Thanks, Gendo. All right Thanks, Ed. Okay, I'm just take over the screen again There we go Perfect. Thanks, George. Yeah, just to, yeah, thanks to Ed. And we've just got a very quick poll, please, if you could just quickly fill it in, it won't take too long. Mainly just covering kind of industry and then if you've used Gendo on a real project as well

And then where you struggle most with. So maybe just as that's happening, obviously, thank you so much for all the questions, George, Ismael and Will going to cover quite a lot of those in the next section of this as well. So yeah, please do keep answering through the poll. That will help steer some of the conversation as well And again, thank you for everyone who's put in so many questions as well, which is brilliant. So we'll get to that We'll just give it another few seconds Yeah. Yeah, please do keep the questions coming. We are taking notes. We will have tactically left some because we know we'll touch on them in a minute, so please don't worry if your question hasn't been answered yet. Cool, perfect. Cool. I'm going to wrap up that poll now and yeah, I'm back to George. Thanks. Thank you. Cool. So we're constantly working to improve the platform. We've got a small but mighty team making changes, some of which more obvious than others, but we also do constant calls with users and studios to make sure that gathering feedback. So we thought it might be worth just showing you some of the things we've done in the last few weeks, so that you're up to speed, which again will answer a couple of the questions we've had The first is that many of you have noticed that over iterations of images, this sort of quality can take a bit of a hit. So we've been working, or Ismael rather has been working hard to find a solution there So, in this video is an example of the same prompt run against Gendo and NanoBanana, so that you can see the difference. Basically, this is sort of a fundamental tech problem, so you'd experience a similar thing in ChatGPT or Grok or whatever, that sort of Use of the same image can rather destroy the quality. So what we've shipped in the last few days is a change which means that you should no longer see the images degraded anywhere near the speed that you were before, and that again we've restricted the chance of it making changes outside of the area that intending to. There's a couple of questions around that. We've sort of mentioned in previous calls, AI is inherently random, and depending on how you've prompted, what your input is, etc, you can open the door more or less to the random, but it is always there. But we do a number of things to try and control that and keep it architecturally specific. So this is a really exciting one because it means you should be able to iterate much more regularly and rapidly and maintain image quality while you do so. So please do have a go and see what you make of it What I'll do is just pan across to some boards, and we can maybe talk about a couple of other things. I've got another example here of the degradation. So again, same prompts, nana banana and Gendo in this case, but we can see that even, I think, by the third image This is the Nano Banana one. There's pretty mucky vegetation in the background here around the changes, whereas with our New update, it is kept much more tight, much more consistent, and not messing around with anything outside of the intended changes. And you can see that it starts to change the overall image quality and tone, not just in sort of resolution terms So, yeah, we're very excited to have Got this one solved. So the eagle-eyed of you will have noticed a couple of other changes that will have happened this morning. But this may be sort of first one maybe goes back to what some of what Ed was alluding to in terms of reusing prompts and references that you've used previously, a regular request was being able to go back to previous generation details and to extract that information to then reuse in a new one. So what you should see now inside of the history panel, I've generation in this space, just for example, is a new details button. And when you click on this, what you'll see is the feature used. So render design in this case My original input image, the image reference used, and the prompt used here. So I can click this to copy and paste it into another image. I can drag out these references again to reuse them. So again, you should never lose any of this information And if you want to go back and reuse a prompt that you found useful, you can find it here, and then you can also then just paste it into a sticky note. So that you can come back and find that later too. But that's in the project history now against every generation. We have also done a slight refresh of the UI. So all works the same, just more of a graphical thing. So if I select one of our images here and click on render design, for example.

What you'll see is just a slight change in layout where we've now got the sort of subsections with little indicators as to whether or not you've filled this in. So I've got render design here. I can drag just this image as a reference, and that little icon is now pink And I can say I'm not seeing just being a bit loose here, but you can now see that that is ticked, ready to go, and we can generate. But that is across all of the features. So it should be much clearer to new users and odd users alike what it is they need to put in the state of the tool as you're using it So that you can make sure that you're not missing any information that you would otherwise want to include, or why you may not be able to press the generate button. This is something we tested a few months ago and have only just now found the time to implement. So again, the older users will recognize this from the original version of the product. We have also then added a new preview mode. So again, what you'll recognize is that we have the little preview panel. But when I click to look at the outputs and compare them to inputs rather than Holding them or proving them above the original input image, it now moves to the right, which is where the image would end up when you press the plus icon to add it to the canvas. We found some people were getting a bit lost as to which version of the image was actually on the canvas, so this now hopefully is a little more explicit and clear. But again, just a small graphical change, I can press the plus on these and get them on the canvas As before, but hopefully a little clearer and tidier. And then, oh, yes, a personal favorite. We had the button down here, fit all in view, which sort of depending on how big your canvas was, when you clicked it would only zoom to get most of what was on the canvas in view We have sort of unrestricted this a little more so that when you press it, it does actually zoom out and hopefully should capture everything. So if you get lost in the board, can't find something, I find it helpful to press fit all in view to then zoom out, see everything from a bird's eye view And then you can select the asset you want to work with. And then there's this handy little refocus button here, which then brings you back in onto that asset directly. So if you find yourself at a lost in That's a nice little quality of life change that should help you navigate the board. Cool. So a few of the things behind the scenes and we'll touch upon a bit more and some things we're working on at the end. But before handing over to Ismail to talk about How to get some consistency, some best practice, and sort of run through some examples. We thought it might be helpful to just touch on prompting and some best practices here. I mean, hopefully, Ed has teed up most of these general tips for me already Which is awesome. So yeah, again, Ed said this himself, that when prompting, you want to be specific about what you want to see. And I mean, maybe when we're talking prompting, Ismael will take us through some input images as well. But Ed said that You want to model the things you want to see and have that level of specificity alluded to the fact that having some indication of materiality, even if that's just sort of colors in places, is helpful. But then being specific about what you're describing, what you're expecting to see in the scene Because if you don't say it, you shouldn't expect to see it generated. So, as he said, there is sort of a happy middle ground of, you know, enough information, but not too much or too little, which will come from a little bit of practice more than anything. But what also helps when prompting is to not try to do too much at once. So we saw in Ed's examples that he iterated on a couple of images. He wasn't trying to get everything all at once and didn't have a prompt that sort of alluded to every possible aspect of the image But described, you know, the general direction, had the right reference image to push it in the right direction to, but then went into detail to be like, okay, well, this is 80% correct, I'm now going to fix the material on the stove, I'm going to remove the man and add the dog and Ed's examples But those are specific changes, and then that means your prompt doesn't have to be overly complicated because you're not trying to fit everything in. You can be direct and clear in a way that both a person and an AI model would understand Which maybe takes us to this next point that There was a sort of phase of prompt engineering where you really had to be quite structured about what you were saying, the order you were saying in this funny sort of word salad sort of grammatical thing. This is less and less relevant. So the sort of prompt anxiety that Ismael sites, I think, are sort of born of that early phase of the tech, but everyone who's training these models to understand conversational instructions in simple, clear instructions better so that you don't have to have some secret prompt skill

You should just be able to say what it is you're wanting to do And we also look at the models that we're using because we use a number behind the scenes, most of which are our own fine tuned models And we can see what you're asking for and make sure that the model understands that based on your instruction. So one of the questions was about some of these small changes that might happen. As I said, we're improving that and trying to control the models But adding in things like don't change the camera angle, like Ed did. I mean, we assume that that's not necessary. The end game is that you don't have to say what you don't want it to do. You just tell it what you want it to do. And I think sometimes saying what you don't want it to do can be confusing, because you're throwing in keywords for things you don' which can muddy the water a little. So I would always be careful with the words that I'm using and the sort of context implied with that. I think Ed was saying about adding in the little details of the hardware when talking about the door or material finishes or the sink. Some of that is just sort of in the language as well as in the input image. So if you say window frame, it will understand or frame that that is what these things look like and the little details are part of that. So this goes back to the specificity. You want to include terms like that to make sure that it understands those things in detail. And then, yeah, preference images are part of the prompts. Critical parts and always quite helpful because as Ed said, they contain a lot of context. There's atmosphere, detail, lighting, materiality camera angles, all of that stuff is in there, and we're looking at how we can sort of break apart the referencings, and so you could be more nuanced. But at the moment, it's sort of everything all at once. So you want to pick the right image carefully, as Ed said, to make sure that it is pointing in the same direction as your prompt, critically. but also has all of the things that you're looking for inside of your image. And as I said, they spend a long time looking for the right references, but what we see people do is using the describe tool to generate a reference that they like, that they might quickly generate a bunch of descriptions. So, the, what was it, the Agri cabin was… I think was that the phrase that Ed used on his prompt? Something like that I'm just going to generate that. You might be able to put in these keywords, describe what you're imagining in general terms, generate from just the prompt, and then you can see, okay, well, is this the right aesthetic? Is this the right lighting, etc. And then use that as your reference image to then carry into more design specific assets. So that's a very important part of prompting and getting it right. And then the last thing to say about getting what you're expecting out of it is maybe to make sure that you choose the right tool. Ismael will go through this in a little more detail. Greg Kapan has given us some tractors, which is maybe not the reference image we wanted, but hopefully the point is clear still I should have used A more specific prompt, agricabon is too short and not enough context, so illustrating some poor prompting for you there. But choosing the right tool is critical, and we'll go through them in detail, but all of the ones that we've sort of got on the toolbar here They're Trained to be architecturally specific and for certain steps in the process. So we find that people who jump between tools, between chat, between render, generally get their outcomes faster and better where they've worked out, you know, the appropriate place to leverage each. But, as I say, we'll get into that With Ismail So I haven't been watching the chat, but unless there's any specific questions on sort of prompting, I can maybe hand over to Ismail Okay, cool. What filters on as we go? Well, I'll stop sharing This one can take over Awesome Thank you, George. I mean, now we will try to jump into a more practical demo, product demo sort of workshopy session where we focus on the design consistency, but just like adding also to the prompting and what George was just discussing right now. I would also leverage the Gendo Assistant for that like, if If you are still a little bit earlier in the, like the design phase and you still have you want to have some ideas about like the design and you're not like very specific about I want to do 1, 2, 3, the Gendo Assistance is really good at that and it can like nudge ideas back and forth With you, which is something that I hopefully would be able to showcase right now. So I have a SketchUp model of a project that is quite detailed, I would say. So this is a SketchUp model where

a hint of, like, materiality is added with some with some colors as well as like differentiation between what's transparent, what's solid, some trees Some references to like the street, and the point right now for us is to do it live with you. I mean, George and Will and Joe hopefully will be joining me also On the on the board and helping me is to create like consistent views of this project, and I want to take you step by step on how we would achieve that There are a lot of different inputs that Gendo is able to deal with. I will show you in a bit all of these different inputs But for now, let's say we're going to start like this is the model and we're going to start by capturing One of the views and sort like what Ed was discussing also in his showcase, you should just like take an image and start and see where It goes from there. So I'm going to like it's as easy as grabbing a screenshot. So I mean, I can do it now live with you, or you can export an image So Gendo deals with any sort of image input, so it needs to be a PNG, a JPEG, WebP, any sort of image input that you can then just drag and drop inside the canvas. Maybe I could show you Really quickly how this works So I'm going to take another screenshot. Sorry And then I'm going to open Gendo and you see here, this is my screenshot. You can just drag and drop it. You can copy and paste as well. And you can hit the upload image to add any image in. So this is the screenshot that I just took And I mean, I'm just going to start. Yeah, I mean, I have this, again, quite detailed model, not all the way where it's like rendered into a realistic image Sort of like an landscape or alumin. But it's quite detailed. And I just want to start by rendering this into a realistic image, something super simple. So I'm going to select the image and hit the render design. Let me try to Minimize this. Yep And hit the render design, which is like the first feature that you would see related to the image And I have prepared this reference image, which is an AI generated image. You can see here from the faces of the people, but I don't care about the details of it. I just like the atmosphere of it. So I like the way the glass is represented in this, and I like the soft shadows that come with it, and I want to apply them basically to my design here. So what I'm going to do is I will simply select my image Okay, let me, sorry. Because this is pre-done. I'm just going to copy and paste this. And I'm selecting the image and I'm hitting the render design And then here I'm like welcoming us as the new sort of sequential panels that George was just mentioning a minute ago, where you collapse These inputs and whenever you fill up One of them, you get this nice indication that it's done. And here in the description, basically, I'll just describe what I want. Yeah, I want to, first of all, I want to render this design using the lighting photographic style And atmosphere of the reference image Just add that the things that I just told you that I like, which is basically overcast sky, overcast Sky Soft shadows, yeah And I'm just going to hit generate. While it's generating, this is one of the inputs that you can have with Gendo. I've prepared here a board with all the different types of inputs. We have also other ones that I can walk you through, but we can go through them really quickly. So The render design, for example, or like any sort of feature that we have can handle something from a very conceptual sketch like this one, where you're just, like, thinking about the ideas to a more defined one where you are already setting the lines correctly. A massing model. So this could be something that you modeled in Revit Right, no, it doesn't matter. It's software agnostic. You can model it anywhere. SketchUp, whatever

A more detailed but still white model. So here you can see we have a little bit more detail on the facade elements here, as well as some a little bit more detail on the context around it. You can even have a layered model where each material, for example, is on a different layer that has just a color. And basically you can use that to reference what each color means. So if, for example, this is the prompt, basically, where cyan is the glass, beige is terracotta, brown is wood, and it maps it for you Just need to communicate it quite well Yeah. Yeah, maybe just on that one, Ismael, I think what's helpful to say is there's obviously you've used the sort of color ID effectively quite literally in the prompt. I think I'd also add that you Exactly. Don't necessarily have to be so explicit in the prompt, depending on the kind of asset you've got, because while it's not sort of explicit IDs in another way, then it sort of understands that those are the same object, or might have the same finish, you know, the blue is glass, it sort of can now understand across different parts of the facade that it is, you know, the same sort of finish. So it can be helpful just to do that not full of materialing, which I think Ed had done in a couple of his examples, which definitely helps with the specificity and control. More information to lean into But doing these sort of block colors, even if arbitrary, can help distinguish between different sort of material finishes, which is, yeah, helpful. Yeah. Yeah, yeah, exactly, exactly. And also, I mean, again, I would definitely use the Gendo Assistant for that as well. Like you can just put this image and communicate to it what you want or what each layer is, and it will craft a prompt for you correctly as well. Yeah, and maybe there's a throwing back to one of the questions about sort of consistency in some of the more creative things that happen. Chat can be one of the more creative features, and that's because there's sort of one more layer of AI between you and the sort of generation Because while you're telling it things, it is sort of writing the prompt for you, which is then its interpretation of the instruction you've given it, whereas when you use a feature directly, you write the prompt much more explicitly and can have more control, and therefore it can be more reliable. So, as Ismael said, when you're in the earlier stages, chat's great because you can fly through it really quickly and do lots of iterations. and variations, but when you start needing more specificity, more reliability, it can be helpful to jump to the tools directly yourself to keep that Yeah, absolutely. I totally agree with that. And I will show you an example also with our board that we are preparing now. Just to walk through the other different inputs that you could use and you all need just to be images. Yeah, I mean, it's similar to what we just did with SketchUp. So this is similar to Our SketchUp model, where there are some sort of colors to the materials applied. The materials are not necessarily textured or realistic. But there is a little bit of a curation to the, at least the colors of the materials. And this is like a little bit more detailed, let's say, all the way to CGIs. Yeah. So you can have already a CGI like what also Ed showed with his endscapes And it works really well also with that. And at this stage, the model follows much more your input because it understands that the designs have been made and a lot of decisions has been have been set So it sticks your like colors and inputs a little bit more which is quite handy other types of inputs. Yeah, you could have like a a Google Earth with your design in it collage sort of collage, and you can start like rendering with that Also, you could also sketch on this is an example that we did earlier on, you can sketch on it So it's sort of like a sketch where you can then apply like some sort of materials and then render it Yeah. In place. Let's see Howard. I always find this. Yeah, I need to move this. Let's see our Generation or first generation, how it went So it did two options, and this is the new preview panel that George was just showing. So you're able to compare not on the… this is the actual asset, and you're comparing next to it, yeah? So we have two outputs I think I prefer this one maybe I would just place it On Canvas and You can see that it's… I really like that it sort of captured The reflection of the glass that I liked about the reference image, as well as it captured really the colors of the reference image quite well. So this is a really good 1st step, and

whenever I do this step from moving from excuse me, something that is like a sketchup model screenshot into a realistic image. It's sort of like you You find right away what you don't like about the image. So, for example, what I can see here is that there's so much Of the space down here that is occupied by like this road That does not make it quite friendly or pedestrian friendly, something that you see when the image becomes real. So this is why maybe starting like doing this leap, the step of Just realism makes you make better design decisions as well. So right now, I mean, I'm going to use because it's still early and I still don't have no idea about like what sort of material I want to apply here other than like to replace the asphalt. It's a very good chance for me to use Gendo assistant to like, come up with some ideas. So I'm just going to drag and drop this image here And tell it, I want to replace The Svelte Road material With something That is a bit more inviting. And This trend Friendly I can see already like down here below me, someone is already cracking on and creating other like images. I feel like I will be behind these guys. They design super quickly. I think it's Will or maybe George But let's see what chat does. So basically right now is going to look at the image. So it's inspecting it and trying to understand what I mean by what I'm saying. And then it will start like giving me some suggestions about what to do with this I felt So that's a great idea, replacing it. To get started, I need you to select the area. So it's asking me to select the area of the road and it's giving me some options here. Right now, also, this is like a good tip. I mean, I will cover the edit function where you can select areas. But because the aspect is sort of occupying a big portion of the image. So it's not like a specific area inside the image. I would be touching it a bit more later on this. But because it's like taking occupying a big portion of the image from side to side, so I'm going to use the entire image And then it's asking me which material we should use Yeah, I would Maybe do that With earthy tones, yeah, maybe it would give it like a better A better look, inviting look, and I'm going to use the entire image. Again, because I'm not like manipulating a portion of the image So this is a good example of just like editing everything. And it's just giving me here the confirmation of what it's going to do and if I'm happy with it to start rendering. Also because right now I'm like exploring ideas I'm not really picky about like what it's doing. So I just want to see How it looks and how it feels with this design right now While it's generating, maybe we can start taking a look at what The guys we're doing right now I was sure it was you, Will. You want to jump in maybe Yeah, I mean, just to say that On the theme of consistency, obviously something that works quite well is Isma kind of established an initial render obviously to be edited and changes to be made to his image, but that was already enough for me to take and use Ismail's output as one of my references. So you can see, yeah, what Ismail has on screen there is It's a slightly different view where I've taken one of Ismael's outputs, which has the right materiality, the right lighting that we're after, and then using that to apply it to a new, slightly different view of the same At the same building, which works really nicely. And I'm just cooking a video in the background while we're Oh, we're doing the rest Amazing. I think the render is done now. I mean, it definitely looks better, I would say. What do you guys think? For the sake of time, also I will not like be super picky about what it does, but it for sure makes it much more friendly.

pedestrian friendly and more inviting. So maybe we just need to add a little bit more life to this road, maybe by adding a couple of cars. So this is a very good occasion for us to use the edit tool. So I I would use the edit tool if I want to edit a portion of the image like I like what what's going on, and I just want to maybe add something, remove something. I want to edit the a portion of the image, and it works really well when we select an area So let's say we want to add some cores maybe here in the driveway, maybe add a core here So I'm going to, because right now I'm editing a portion of the image. I'm going to it's asking me for an area. I'm not going to use the whole image. I'm going to select the area that I want to manipulate basically in the image. And really good advice. And this is something also related to how the models were trained and the types of images that they saw while they were being trained. Always try to, when you're selecting the area, first of all, this is not a precision selection, so I'm not going to like try to select like this is my core and I will try to, it's a bounding box of the core. No What I would try to do is first of all give it like enough context so I'm going to, for example. Select a larger area where, like, I want to add the core. So I'm giving the model some context. And this is. This does not need to be precise at all. It's not like the the edges of this box need to fall exactly on my design. This is the first thing I would say. And then the second thing I would always try to make it more towards squarish selection rather than, like, being too Wide or too tall. And this is because the probability of the model making something wrong when it's that like a tall selection is much higher than when it's a square selection, and this is basically because most of the images that it has seen, like, while it's being trained, are more towards the square or maximum, like this, a landscape sort of image, while, like, when it was being trained, it was never, like, fed with these sort of images, right? So or it maybe was fed with these sorts of images, but they were like proportionally much, much, much less than these sort of images. So always think about that while you're selecting. Try not to go extremely wide or extremely tall So I'm going to like roughly make something like that. And then I'm going to save selection and Here, we have an optional reference image that we can add. Maybe if I have, like, a specific card that I want to add But here I would just go add contemporary I'm prompting like an architect now car to the driveway Okay, and I'm going to hit generate. So I'm hoping Right now, it will only edit the portion of the image that I have selected, so everything outside this Books will not be edited and to hopefully stitch, add a core to the driveway here Maybe a couple of questions about masking have sort of come in specifically about the edit one, but I think, as Melissa said, it's more like a focus area rather than like an explicit thing Yeah. That being said, we are doing some work to allow for masking across tools more consistently, but I think there's another question was sort of how do we decide when to be more creative or more controlled I think this is one of those tricky areas where we sort of want to give the AI enough information and context around what you're doing so that you might want shadows from this car, for example. So if you're tightly cropped A mask of a car, you might not then see the shadows, etc. So there's a fine balance, but we are making that hopefully a more user-friendly. Yeah, yeah, exactly. I mean It's a fine balance between. Yeah, if you want your like If your selection is exactly what you want the model to manipulate Even if it's that this case, it always relates to something outside the selection as well. So it Yeah, as you can see here, like if I selected exact area of a core and added it to this image

It wouldn't have blended that nicely into the image because then you lose all the shadows that it adds and everything that comes around it, yeah? So this is why We made like the edit selection the way we did it right now. So we allow for these sort of edits to fall perfectly inside your image If that makes sense. But that said, like, what George is saying, we are working on improving the selection experience in a way that still allows you to do this kind of stuff. Yeah So we've added our card to this image, and I think I'm pretty happy. Maybe we could like populate it with some people, but I've reached a point where I'm quite happy. Maybe we can also enhance it. Let's go to the populate and here you have some options about like what type of people you want to add to the image. I'm just going to make the density to sum And in the description, I will just because I like the look of it, I will add like motion blur people. Or just motion blur it and see what happens. And yeah, I will leave like the mix Open. And I'm basically I'm adding some people, adding some life to it. And like probably maybe after that, I'm, you know, like the results. I really, really like the how it's captured the Yeah. the glass material here, I don't know, what do you think about this, George? I mean, you have an eye for this stuff Yeah, no, it's looking good, it's looking very convincing. Maybe to field another question that's come in in the chat, which is when you choose to use An edit and the balance of those things. I mean, maybe I can give my answer first, Ismael and see what you think. But I think some of our assumption around those features is that with render that All that's really true in that image is the geometry is your design, and that we will then sort of infer from that and add materiality, lighting, context, etc. So render in the traditional sense, whereas other tools like edit and swap material will assume to varying degrees that everything else is also true, that the lighting and the material is true and therefore to be respected and consistent across iteration So our assumption is that if you're coming in with a fresh sketch, a fresh screenshot, whatever it is, that the first tool you might use is render because you can get 80% of it done the architecture geometry stays the same, the composition should stay the same, etc. And then you go into iterating details with edit, with swap material to make those small controlled changes Swamp material will keep geometry consistent as well, but change the materiality like swapping a material in SketchUp or 3disc or whatever. Whereas edit will allow bigger geometry changes. So we've seen Ismail added a car with edit what material a reader wouldn't have done that, because they would have wanted to keep the driveway and the garage and that sort of geometry clean and consistent across iterations, whereas Rendo will do the sort of will deviate from that. So more flexible. Yeah, yeah, absolutely. Maybe we can do it like a swap material over here. an iteration of it. So Like, as George just mentioned, swap material is very specific to like changing materials, right? So maybe let's change the material of the Garage door here So it allows you to select perfectly, let's say, because you're sort of changing a surface, right? So this gives you like the lasso tool. Again, we are working on unifying the sort of selection experience and like the way you communicate with the models is a little bit of a complex problem that we are actively solving right now so that you as users, we will be able to, like, choose whichever way you want to select with And then here I'm let's replace The material Door with Like, TVL I'm just going to hit generate Yeah, I've got a couple While it's doing that, I don't know if Georgian really want to mention something or I can jump into another image. I can see Will absolutely smashing it. Yeah, sorry.

Huh. Yeah. Yeah, well, I've got a couple beneath here. I mean, I was just going to drag mine up next to Will so that we can sort of compare our consistency and maybe copy one of yours as well Because I think it looks like we sort of done a similar thing right with using prompts that sort of align with the material palette, but using the same sort of references across those things. So the first key reference, the first image that Ismael has made. And I mean, you know, again, we're being quite quick here I would say that those are pretty clear and pretty consistent, but the architecture is consistent, as is the material palette. Obviously, as male is added different roads textured to his, but we could repeat the same process But hopefully it's consistent enough, you know, across this sort of set. I feel like I'd be happy to put those ones out in front of people and say that it is, you know, the same project, at least I just wanted to call out as well a really nice feature from or videos, if people haven't already seen videos are great, and I've got an example there that was generating the background. But also there's a really nice feature which allows you to save a frame from a video out into the canvas Which is what I've done on mine as well, just to kind of present an array of different views. It's a very… if you've done something like, in my case, I did a cinematic edit which part of that preset is to create different views and different movements around the the the building. And what I've done in the board is then extract some of those views and use them as references to other things, or just to have them as a slightly different option, different materiality, etc. So yeah, really nice little feature that's maybe hidden away is using videos and extracting frames just to Nice To add to your board Yeah, I like really like this shot. Yeah. I also did something with, again, just for consistency's sake took Ismail, some… one of your generations where you've added a car. I've taken that image and used that again as a reference in an edit to to have some consistency. So if you did what obviously mine's a slightly different view. But if you did want to have similar cars or similar things You can obviously cut those out and use those as references in edits as well. It doesn't have to just be a reference for a render. It can be something that applies when you're just doing an edit. So you can see there, I've got the car The same car parked on a different spot Sue. Yeah, nice, there was a question about scale, which maybe also relates to that image too and about getting things the right sort of size What typically helps is that there is context in there where you might have, like, a low poly image, or not loads of detail. It then leaves it open to interpretation, like how big things are, whereas I think in Will's picture, the scale is… looks pretty decent, because the garage door is there, so it could sort of scale against that Relatively. So I think we populate, sometimes people can be a bit of a funny height too, but where you have, like, door frames or furniture or something that is, you know, human scale, then it does a better job of placing those people because it can sort of relate that to the rest of the scene. Yeah, I would say, yeah, one of my generations, the scale was off and definitely I would something about like how you prepare your models. So I would definitely Here in this model that I have, what was missing is a reference to scale, and just one person Here would make my shots just placing a person here, even if it's like a low poly, just an indication of a person height Would make this shot much, much better and would make the model understand the scale much, much better. Yeah, maybe And not do these type of mistakes. Yeah. Yeah. Yeah, maybe another point to make quickly is that, you know, we've touched upon the inherent randomness of AI. Gendo is a creditless platform, so there's nothing stopping you from rolling the dice again and seeing if there's a better outcome. Of course, we'd want to get it right first time and work towards that, but not having credits means that you can iterate as many times as you like to get an image that you're happy with I think that's an important detailed highlight. Yeah, and reflecting back to what Ed was saying, which I completely agree with, like about when he was describing the process of being model image model, design, image design image. like, what we're doing now is part of the design process, yeah, and you shouldn't be while you're designing, you shouldn't be thinking about, should I make this decision or it's too expensive? It's a bit strange, like it makes you stop Your design process. So I think it's fundamental that the image bit Because images are so essential in the design process that there is no blocker there. You're just like trying things when you have an idea, you try it, and even if it doesn't work, you have to try it and know that it doesn't work. And since we were talking about how you prepare your models before you take them into Gendo, this is a very interesting example that something that happened to George while he was

working on an ARCHICAD model. As you can see here, this is the input image, yeah, and there is a very tiny little detail of the like the crosshair of the camera when he was taking the screenshot That got implemented into the design somehow, like Gendo is like Our models were able to capture it, so yeah, I mean you need to prepare your screenshots the way you want to communicate things to the model or less. Yeah, this was a funny one, Georgia. Yeah, yeah, I think it's also things like ISOCAVs in Rhino, if that's what you're working with, or edges in SketchUp, like you can adjust the viewport sort of settings and styles because going back to some previous points and questions, like we assume the design is intentional and that everything about that is intentional So hence sort of seeing the crosshair and thinking, okay, well, that's part of the design, therefore it has been reflected in the render. So please make sure that when you have those extra lines that it is what you want to see, because we'll assume And run with it to some funny outcomes I want to touch maybe just because as of right now, we were using render design all the time. But I mean, this is a little bit of a tricky sort of view where you have like a lot of detail that is so far away from the camera, let's say So in that case, I'm just going to try it. In that case, I would add an extra step Again, I'm trying things so it could work and it could not work. I don't know we will see, but I want to take the leap from like it being a screenshot to realism using the enhanced tool, which basically this is what it does. The enhance It takes your image exactly as it is, and it just tries to make it photorealistic. So it's much more precise Then the render design, which has a lot of different inputs where you can, like, add a reference, add a prompt, say stuff, so there's a lot to handle there the enhanced image is much more simpler in the way it's just focused on enhancing photorealism Which I think would be the right start for this image, even if we don't get the color consistency of the rest of the project, maybe Yeah. We could then reference it right? Yeah. Yeah. Yeah, yeah, so maybe to contrast this against what we were saying about render, where we assume it is only the geometry, which is the bit that matters and maybe like the blocking of colors for segmentation, enhance will assume that the textures are also true. So we should hope to see that grass is in the right place, the wood is in the right place, that those material palette sort of comes through, but in a photoreal way There was a question in the chat about Enhance, maybe changing the lighting and some of the details. We're introducing a slider so that you can say more or less creative, because sometimes you want it to stick really critically closely to what you've got, and sometimes you want it to fill in the details a bit more where you've maybe not I've specified everything. So that's something that will be out shortly against that feature. And no reference image for the enhanced because we assume that that is photorealism, the last sort of 5-10% just to elevate everything that's already within the image. Yeah, just to add on to that as well, I mentioned in the chat earlier on that that can be quite a good way as well just to harmonize things in the image a little bit if you've done Tons and tons of changes, and maybe things are slightly, you know, a few seams or a few lighting things that are a little bit inconsistent sometimes enhances is a nice way of making everything sort of reharmonized in the same image again. And actually we're we're going to be releasing something soon that allows you to decide how strongly you want that enhancement to be, because that will help in that situation where you can just do a tiny bit of enhancing and that will just knit everything together nicely. If you do have slight differences in parts of the image Yeah, exactly. I mean, you definitely have these situations where your images almost finish, like it's almost done and you just want to take it like a step further into realism. So you would have like the control for that and other situations where like the one that I was doing right now, where you want actually the model to do the leap from realism without losing any detail. So yeah, we will give this slider that makes you do These different modes Let's say So yeah, I mean, I think this is good. I mean, I will try to use now render design To apply the lighting and see how it does. I will. I will use actually my OG reference image for that And I would say Only apply the lighting And

Photographic style of the reference image Let's see what it does. It's a tricky shot, this one, to be honest. I mean, we are also working on Capturing these tricky shots automatically and being able to deal with them Yeah. A little bit better Yeah I was maybe going to highlight something that Ed referenced too, which is that in his prompt, he sort of says, where in the world this is, like, whether that's literally like in the shadow of Mount Etna, or in Sicily or in London, whatever it is, this sort of goes back to what we were saying about the words you use, because Yes When we say that we're in London, it will know what London is, and it will know sort of brick houses, grey skies, all of those things that people associate, whereas with Sicily, it knew the right kind of vegetation and landscaping and context and vista to maybe sort of show So I like to include, like, where I am in the world in a prompt, because it will then, you know, fill in those gaps nicely, put the right kind of vegetation is in, whereas when that is a bit absent, it then just sort of, you know, puts something in, but it's a little less specific. Absolutely, yeah, yeah, I totally agree with that. I mean, this is the first example that came to mind when we were rendering this image Just by saying it's London, you can feel like something in the lighting and like the houses here that just feels London, right? Perfect, grey and flat, yeah. That was about saying that's the grey sky. I was trying to be, politically correct I'm muted Yeah, nice is a bit warmer Yeah, exactly. It applied a little bit without changing anything about the design. I don't know, it feels like the same project. I mean, of course, also when you're presenting a project, you don't want, like we saw it with Ed's images, you don't want all the images to be exactly the same lighting, right? Like you can definitely start playing a bit maybe with this image to make it a little bit more towards the evening. Again, adding the context of the location would make an evening look Correct to where these buildings are placed. But at the same time still keep it within the same like design language. So Yeah, I mean I think I've covered everything other… if you want me to cover something else, or if there are some specific questions about things we Yeah, maybe there's one question. This has come up in a couple of different ways, but about aspen ratio and image resolution So we work hard to keep the aspect ratio the same, whatever's sort of happening. But what happens with the image resolution is, as I said, we use lots of different models behind the scenes, and depending on which model we're using who you're using and how you're using it, the resolution may change, so when you click on an image and zoom in, on the bottom left corner, you should see what the resolution of that currently selected image is, so you should be able to track this as you're going. But the change is basically due to the models, like, generating things at a higher resolution takes exponentially longer And again, we assume an iterative workflow so that you sort of work through sort of between 1 and 2K images, and then when you've got something you're happy with, then we sort of upscale to higher resolution things. But in that sort of swapping between models and processes, we do upscaled and downscale a bit Which then introduces rounding areas with the aspect ratio. So what you may see over a course of lots of different features is that one or two pixels out most on the aspect ratio, but the resolution may deviate a little, depending on which step So we try to keep it high where we can. We try to composite where we can, but there is a sort of natural limit to that. Yeah, I have a couple of things also to add to that. So first of all, about our choice of iterating around between one and two K The first choice is, of course, it's an iterative process, and we want, like, to serve you the images much faster than having, like, a full resolution 4K image, but also because models tend to be much better around this resolution, and I'm guessing it's also because of the way they were trained This is the vast majority of the images. They stand between one K and 2k. And the model seems to perform much better when it's working with these sort of resolutions. So it's also a model choice, not just like speed But thing to add also is when you're using the edit, so for example, here I use the edit to the core. This will keep the resolution exactly the same, so because it doesn't touch anything outside Of that, so if you have an image where you want to edit a portion of the have a very fixed resolution aspect ratio using the edit

function would mean that this is fixed. Yeah, even if this was like a 4K image of a much higher resolution image, it will not be downscaled. It will only add or edit the portion of the image that you have selected there Yeah. Yeah, I mean, this may be also an answer to another question someone asked about sort of AI faces of people and depending on how big they are in the image, they can be a bit squished. What it is good for as well is that we sort of zoom in on a detail. So you can make faces look better if they're front and center to improve the quality and have them sort of stitch back in. And there was another question about sort of density of populate. The tool has a sort of population slider which is relative to the image, so you can see it goes from low to packed A packed living room would be 10 people, but a packed football stadium would be 10,000, so it sort of understands contextually within the image what packed means. So this, again, is sort of relative. So someone asking about football events, you should be able to get busy scenes, and you should be able to describe in the additional keywords, like that they might be doing football or whatever, or watching sport or cheering, you know, and get some very specific activation in depending on, you know, what your project actually is. Yeah, also just to jump in as well, we had a question about being restricted to photorealistic images. Obviously, the focus on today has been Consistency and that tends to be photorealistic consistency, because you're trying to represent the same project, but I just did a generation while we're tracking there, just to show off the style features So yeah, you can, of course, you can put any of the images you've got into, as Ismail showing there and a number of styles that we've added to the platform which are very consistent themselves. If you use the same preset across multiple images, you'll you'll get a lot of the same They're looking like the same style, effectively, which is really nice, and that's… that's one of the ways to take it out of the photorealistic realm and into something that's more stylized. We will be adding very, very soon the opportunity to add your own custom style. So if you do have your own in-house Graphic style, or maybe just an aesthetic style or something that is something that you always tend to produce images Yeah. Using, you can upload a reference image and make it a custom style for this feature, so it doesn't have to be one of those presets but biased. I do think those are some great presets to get you going and cover a few different types of aesthetic, which is nice, and they are consistent. So we could put another one of those photo realistic images in, then we could get a very, very consistent art style out to match the one that I've already done. Yeah, maybe to add to that a little, the style tool, like Enhance, assumes that the materiality lighting architecture is consistent and is only working to change the aesthetic without changing the content, the information in there. So you can do this sort of style transfer right at the end Yeah. But it is also true that you can just prompt from the beginning a hand sketch a watercolor, you know, whatever it is you're after. You don't have to do photographic all the way through and then stylize. You can just work within your style. But as Will says, we've done photorealism for the consistency benchmarking But you can, of course, just do that the whole time. That's up to you. Yeah, maybe, maybe just because I can see a question from Sophie there that's just come in asking about using a reference image from previous work. So yeah, it depends on the image. If you've got… if you're doing a render, as Ismael showed, you can… you can upload one of those images from previous work as a reference image, if that's what you want. Like I said, we'll be adding the style, custom style soon, but you can still kind of do that, as George is alluding to as well, if you do a full image, maybe as well, you can show the edit panel. There's the option to just do a full image edit. And this is quite a nice way of doing that. So if you do have a style you've done previously You could do edit tool up and then add I mean, in this case, maybe we can add the one I've already done, but you can add an image reference and edit the whole image, and that's a nice way to get the style to change to something that you've done before. Maybe your own practice has its own style. Yeah, exactly that. Yeah. And also gender assistant helps with that, I think We… I'm very excited, super excited about, like, the custom style, because it's, yeah, it's another level of, like, custom styling, where it actually the model tries to actually extract the style of the image, which is like not as simple as it sounds. So styles are very specific, and like you need to understand like the way the, for example here the pen is like leaving the ink. They are very specific, and you can see how well it transforms The sorry, I'm getting used to the new preview. How well it transforms the blurry even vegetation here into a style. So yeah, I'm I'm quite excited for the custom one where you can literally just like, give it an image of the style that you want to replicate, and it studies the image, and it tries to understand the style, and then applies it to the

Your input image. Okay, well, conscious of time, so maybe we can show you some of what we're working on next. So is my luck, maybe I can take the screen back and Will can talk us through Yeah Yeah, maybe just say before I do that, that we will think Joe will email a link to the board, the public board at the end of the session as well, so that you guys can actually explore the canvas, zoom in, see the details, because obviously it doesn't always track over Zoom in the same way. So you can actually go and explore the So then we'll follow up with a link to that after the session. And watch the video. Yeah. And watch the video also. It looks so cool, yeah, yeah. Nice. Yeah, amazing. Yeah, great. Well, yeah, to talk through some upcoming features then, I've alluded to some in the chat if you've been reading the chat, but Maybe, yeah. So, we had a question about annotation in the chat about is it just for annotating for your own benefit, or is it actually something that can change generations? Currently today it's purely for markup, but very, very, very shortly we're going to be releasing a change that allows you to use annotations as part of the input itself. So George is flicking through a few examples. You can see in this one here We've drawn, I don't know who's drawn that, a rough stool outline. Looking good, George. Of where you want this change to be, and a reference image in the top corner, and then in the panel, I don't know if we've got an image of that, but in the panel there's an option to select, yes, I want the annotations that I've drawn to be part of the generation, but you don't have to choose that. So even if you have annotated it You can disable it so that they're not part of the generation, but it's a nice way here's another example. And actually, this maybe came up earlier as well, just in relation to scale and scale of people. This is a nice way of doing that as well. So Here's an example George has done with annotating where he wants a person to be generated. And due to the size of the annotation, it also gives the model an idea of how big that person should be. So it's not just positional in terms of where they get generated into the image, it's also a good indicator of scale, which is super useful. And consistent with style as well, so not a photographic. Yeah And consist… yeah. Exactly. Yeah, consistent with style two. So the nice thing is the annotation is it can be interpreted in a realistic way if your image is realistic or stylized, as George has just shown there. So really powerful. This is a nice one Maybe hard spot initially, but George has added in some interesting lighting. I would say it's to my taste, but I'm not sure it is. Some hanging lighting drawn with different colours as well, which is super nice. So that… that didn't mention that, but there's different pen colors too for the annotation tool for that pencil tool And so this is a really nice way of designing in Canvas. If you do want a quick test of a variation or you want to try something out and you don't want to go and model it, I mean, you can get a super nice indication of how that's going to look by annotating, choosing that annotation when you generate and then Using that in the output, which is really nice, so that's going to be very soon. You can see some messages in the channel saying they're already excited for that, which is… which is awesome. Nice and Justin. So that's the new annotation. Yeah, go on, George. Yeah, and this would just be part of edit, right? Because, as we sort of said, edit does the bigger changes, deviates from the architecture, like here, where I've added new things in new geometry. It will then sort of live as part of the edit feature and not the others. Cool Amazing. And then, yes, on the topic of we did a swap material example earlier, but This is… we've added a feature now to allow for even more control of how materials are generated. There's obviously Scale is one aspect, so sometimes you'll have a material sample, and that could be a very small sample that doesn't maybe reflect the scale of material would look in situ. So we've added some additional controls, which are optional. You don't need to do this, but you can get even better results by actually changing the scale of your material sample. So in this case here We've got tiles. Obviously, there's different tiles of different sizes. This gives you a rough indicator of, okay, this is roughly how big my tiles should appear in the image. And you have that slider there to control that And you have the same scale. So if there's a certain orientation of material, obviously rather than generating multiple times until it happens to be the right orientation, you can actually just set that rotation as part of the controls. So these are two new fields that basically give you More control over that generation. I can see someone's asked for timeline. This is very soon, so in the next few weeks, for sure. Yes. Yeah, and works nicely with our material bank catalogue on the platform too, where we have a huge, I mean, 16,000 different material assets of real world products that you can specify on your projects and you can click through to Material Bank to order those samples for free But for our purposes, it's super helpful just to quickly grab the right reference image and then take it into the swamp material tool.

Yeah, great. And maybe let Ismail talk about this one a little bit because this is something that he's been cooking up in the background, but we've been working on some improvements to chat, full stop. So the Gendo assistant that many of you will have used before. We're trying to improve, make smarter, make more dynamic in the things that it can do Maybe I'll hand over to Ismail to talk about someone I think actually asked in the chat about is there a way to have project memory and project references that kind of persist? So maybe I'll pass it to you, Ismael, to talk about the upcoming changes in the next couple of months towards this point. Yeah, so it's because basically I'm a huge fan of the gender assistant. I think it's the way we should be designing. As you saw, like with the demo that we were doing, it's so nice when you were just trying ideas and it just like bounces These ideas with you and gives you some really nice suggestions. So I've been working a lot on improving it. And one of the biggest it's a tiny thing, but it does a huge improvement is basically giving it some project context. So basically a memory Of what's going on with the project. Like what Ed was describing when he was going through his project, they were saving these prompt templates and reusing them. You wouldn't need that because Chat, basically, if it has this context, it will be able to already automatically do it for you. So one of the things that we are working on is that it remembers things and you have the sort of control on when to add this memory in and when you wouldn't want to use it. But other really important things that it can do also is the way it references images, and it references styles. So as you can, as you saw in the little snippet video here where inside the prompt. Okay, so this is maybe we can go through them one by one like these little ones. Where there is a light version of it inside the canvas that is able to execute things for you on the fly. This is a very simple example of it like reordering things. But also inside the chat itself You're able to reference memories and reference styles as well that you have saved as an organization. So this is another example where you're doing this on the fly sort of rendering thing because it has all the information it needs, it's able to do it on the fly for you And this is another example of it, because it understands how Gendo works. It understands how each image was produced, and you're able to replicate the steps that were made to do this image To be able to remember stuff on the fly or like create different views for you on the fly. So there are these two versions of it where you have this assistant that you're able to call in and do Quick tasks for you, but you have this very information rich, I would say, agent that is able to that understands a project context and understands all the nuances that come with it. So it's able to assist you in a better way, let's say. So yeah, I'm I'm really excited for this one. It's going to take a little bit more time, I would say, but it's looking to be quite useful Yeah, I mean, maybe to add to that, I mean, we said right at the top about the ambition of Gendo and wanting for it to be like accessible, easy to use, architecturally relevant. I think this, the material bank integration, those sort of things, I think are great indicators of the direction of travel Where it is operating within an architectural context with real-world references, all grounded in professional context, not sort of AI gimmicks, which is maybe sort of an answer to one of the other questions in chat about like what differentiates Gendo specifically against other platforms. And I think it is that point really that there are lots of places you can go if you want to try 50 different AI models and you want to play with all the knobs and whistles and depth that comes with that. And that's great if you want to do that But if you want to worry about delivering your architecture project, making use of the best tech, but not having to worry about all of the sort of AI nonsense, that's what Gendo is for. It should be the case that you can come in only think about design, but get what you need quickly and then move on and carry on with the other important bits of work, not to be connecting nodes and workflows and worrying about data pipelines and which of these models is good for what step, blah blah blah blah That's what Ismail's for. So you can leverage all of our collective knowledge that's wrapped up inside the product so that you just don't have to stress about all that detail So I think hopefully an answer in that there.

So Yes Just flashing this up quickly, I mean, one of our more recent webinars was to launch the video tour, which will briefly demoed. You can find that recording on YouTube if you'd like to go back and watch it in more detail, but one of the immediate bits of feedback we got during that webinar and the sort of following weeks was having unlimited video generations in the same way that we have unlimited images on the platform now, which is something we believe in, that you should just be able to iterate and edit Generating videos, however, is quite a different ball game from generating images, so we are sort of forced to move that into a different kind of subscription But I think what also opens an opportunity for is to service more of the mid-large size studios to have more organizational control, standardization, repeatability across design teams and projects So a lot of what Studioscale is about is that sort of larger organisational thing, but part of it is also the unlimited video generation People who are on a paid plan can use the video tool, but it currently just requires you buy seconds as separate one-off banks to add to your account and use at your leisure. So you can get to the video tool. It is just priced slightly differently outside of the typical subscription So if you're interested in this, please do get in touch. We're talking to some early access studios at the moment to run some of these new features and things against. So please do make yourself known if you would like to be among them But just to flash that that is also a sort of upcoming in the next few weeks and months. Nice. Maybe Joe, for one more poll. Yeah, very short one, just to launch. Let me just pop this one up now, so hopefully all on your screen. So yeah, maybe just to cover a couple of bits as George was talking, you know. Thanks so much obviously for going through and hopefully giving kind of a variety of different bits of advice and follow through. So it'd be great to understand how much more confident you may feel now using Gendo than you did at the start. You know, we appreciate all feedback around this as well. So please do And then, of course, George has just touched on something which is the studio scale, which we're looking in. So it'd be great as well just to, if you know which one you're most interested in, I suppose, to help gauge some of those delivery timelines as well. And then, of course, if you're interested to hear a bit more about that and what that entails, just simply hit yes and we'll be in touch in the next few days There's also a free form box as well, if you've got chance. Maybe just as that's happening, as we'll go through, we've had a lot of questions. We're going to get back to everyone, don't worry. If your questions hasn't been answered. We've had a lot directly, and we've had a lot in the chat as well But George, maybe while this is up, I don't know if there's any other ones you want to tackle Yes, there was maybe one I think Amit has asked a couple of times, apologies for missing that in the flow. A question about generating drawings. An important question, I think there are, again, lots of platforms you can go to to help with sort of that side of things We operate under the assumption that you, the user, are the architect, are the designer, are in control of those things, and the AI is supporting you in that, which maybe also goes back to what Ed was doing at the start. Like he was putting in detail into his model, specifying Lots of nice little things that was architectural design, and then Gendo was adhering to that. So we try hard to stick to your design as you've specified it, rather than do the design for you. So we don't really do technical drawings. You can bring in technical drawings to turn them into a waterc For example, or maybe to texture them, but we again will stick to the drawing as it is in the image. So We don't sort of generate those for you. And about changing camera views, you can ask chat to do it, and it does an okay job sometimes. But again, this is something that I'm always a bit hesitant about, because the AI only sees what is inside of your image. So if you ask the camera to do a 180, it will make up What is it seeing the other way entirely, which is not what us as designers should be expecting. But when you might want to make a small camera adjustment or turn it into an exodiagram, as long as all the information is within that image to begin with, it could do an okay job at making those changes. But again, we assume that the composition and consistency is intentional, so some of that stuff is not super reliable, and if done, is maybe not more out of luck than it is out of intention Awesome. If you want to hit the next slide, George? Perfect, and then, yeah, we've covered that in one next slide. Yeah. Yeah, I mean, massive thank you to everyone. I mean, the chat's been incredible. Thank you so much for everyone kind of asking questions and as I say, we'll get back to those. I know Ed had to jump at the start, but like in the middle of it, just massive thanks to him as well. And of course, massive thanks to Will

Ismael and George for going through everything in a lot of detail. As George mentioned at the start of the call, we always do offer onboarding calls as well, so if you're interested as well, just get in touch. Feel free to contact any of us or just email support@gendo.ai and we'll always happily go through things in more detail Thanks for dropping in all the questions and feedback as well, and this is one thing to mention, of course, if you are looking at the annual plan, we are happy to offer a 50% off, plus the two months free with Control 15 as the code as well. So I'll be in touch on email on that as well, but maybe just pass back to George Will and Ismael to say, say thanks Well, yeah, I mean, thank you to Ed and his absence, again, for running us through the project. Thank you to you all for taking the time this afternoon to listen to us, and for all of your amazing questions. Hopefully, we covered most of them where we haven't. We'll try and get back to you But otherwise, yeah, thanks to Will Ismail and to Joe for making sure this all plays out nice and smoothly. So yeah, appreciate it, everyone. Thank you. Yeah, thanks so much for tuning in, guys. Thanks. We'll be in touch with the next one, which will be in the next couple of weeks as well. So keep an eye on your emails. Thank you so much. Thank you. Bye-bye.