S5E12: Zo zet je AI nu al in als interieurprofessional met AI expert Stijn Käller

[Music] Nice that you're listening to a new episode of the Interior Club Podcast. My name is Marc Thimo and this is episode 12 of season 5. In this episode we're going to talk about AI. Where are we now and what is there possible as an interior professional? My guest in this episode is AI expert Stein Keller. And did you know that we have various workshops at the Interior Club about AI? Among other things, a workshop with an introduction about AI, where you see different AI tools and learn how they work, the online workshop Meet Journey to generate beautiful images and videos and an AI workshop to make an AI assistant. If you want more information about these workshops, then look at our website www.theintreorclub.com A lot of pleasure to listen to. Welcome Stein Luxizer Band. Could you tell us about yourself? Sure. I'm Stein Keller. At my company stein.ai I help other companies use AI practice. That can be about creating content on scale with AI. So you really have about image materials such as images and videos for marketing, for example. To really use AI within business processes. So then it's really about drawing from a business process and within a business, of course, more processes. And how you then inject AI into it, if it were. That's what I'm doing now combined with a job at ABN Ambro, so I'm a business consultant at ABN. I'm in a process of automation and in the digitalization field, so we build all kinds of systems for the bank itself. And I then accompany all kinds of automation projects for the bank. But since 2026, after six or seven years at ABN, I'm going to start by myself, so fulltimestein.ai and all the projects and all the things that I really want to focus on, I'm going to work out. Exciting, what fun, a nice step. A lot happens in AI, in the AI event. Can you briefly describe what has changed in 2025? We often made podcasts about AI here and then it was still a bit vague. But in the meantime, it's becoming more and more concrete, can you tell us about it? Yes, good, yes. I compare it a bit when I look at where it started for me with AI. That's really the gen AI, the genitive AI we're talking about now, of course. That started for me three and a half years ago when the Chatchapati came out and with Journey. So the first picture you make of it. And then it was already great that you could make a word, an image with it. That was already a kind of, well, breaking the track, of course. And what I see is that the tools of course are getting better and better, so the technique that's behind it. And what you notice is that in the past years and in 2025, not otherwise, you notice that those big companies actually compete with each other. So you see that each company wants to make a better model than the other with better results than the other models. And you see that in 2025. What I see is that there are now certain models on the market that can make such realistic, well, mainly image material, such as images and videos. Such realistic voices can also be made. So for example, I have 11Labs, which is a tool where you can make voiceovers with, also, a joke or something like that. So that's super interesting and also the making of texts and so on, of course, is getting better and better. And I also think that it's getting brighter and brighter, that AI really goes from a hype to a kind of settled one. And that the real changes now only come up in companies and that the hype is a bit off, maybe. And the real changes are going to start now. And if we're talking specifically about designers, architects who have lists, what is it that they can do with it now? Because I get that question very often, from Marc Leuk, the AI, but what can I really do with it in my company? Well, I think you have to imagine that there are actually two pairs, I think, that are understood within the Gen.A.I. world where you can focus as a company. So if you look at it very globally, it's either what I said, improving business processes, making it more efficient, fitting AI into that. Or really generating image material, image or videos. And if you then look at things like, of course, well, interior, branch, architect, that kind of thing. Yes, that is of course very much also visibly focused. So if you have to generate buildings, or a concept for a building based on a description, or an interior based on a mood board in a second description with an AI tool, or a concept development, so 10 images of the interior are generated, which you have stopped in a kind of mood board, so you can get very quick inspiration, or very specifically, I have once harvested a chair from my grandma, and I wanted to choose another material for that, and then you go to the Internet and then you're all sitting there, scrolling down. But you don't really get a picture of how the chair will actually look like. And if you throw that in certain tools, the chair and the fabric, you can actually see within a second how such a fabric looks like on a furniture. And I think that's the most valuable thing in how you're going to use it on a scale. So not like I did with a chair for a month, that's nice. But how are you going to really prove that with 100 chairs in one go on your website? Those are the things that I see. How is AI being used the most now? I think it's kind of the same for people or companies, but I always see it as four layers of using AI, and how deep it is, how complex it is. If you look at the surface, you have layer one, and that's how your AI now adjusts your AI tools. So for example, chat-gibberish or mid-journey. Those are the kind of surface functionalities of making a text or making an image. That's of course already great. If you look further, you have layer two, and that's really about how AI is integrated into software, so existing software. So for example, you have Copilot in Microsoft or Gemini with Google. There are all kinds of ways in which you can talk to your emails in emails with your documents. That of course goes a bit further. Then you have layer three, and then you see that AI is really being adjusted in automation. Then you have a business process that is described in steps as it was, and where you then adjust to AI. So for example, think of a client conversation, if you had someone on the phone. Normally, we would work with a client service, then maybe have a conversation for half an hour, work together, and then put it in a CRM system. Well, if you do that automatically and then look at the scale, then you look at AI automation. And layer four, then you really have more about the agents, which is also called VAG. An AI agent is as fast ... I see that more than, say, but in the Volksmond, an AI agent is called as fast as a kind of task that runs autonomously. I sometimes see that as an AI automation, but if you're really going to look at, for example, a supervisor agent, who, for example, puts different agents on the job, and therefore, in a way, a department runs autonomously. I think that's more of a future music, but that's the layer four that eventually goes even further. So it's a bit more complex, but then you really give some tasks out of hand. And on which layer you also look at, I think it's still at this point, because there's always a cooperation. What kind of cooperation is that? But it could be very good that you don't need ten cooperation, but two that control it and continue to do so. Yes, and if we look specifically at an interior professional, if they do a project, then it starts with a creative phase, concept development. Yes, how can you use AI well? What are the tools that you could recommend? Yes, that's a good one. I think that such professionals or companies have a database or have an idea of the images they find beautiful. So they really go to the visual side, so imagine you find a certain design direction very beautiful, and you have a database of hundreds of images of it, and you throw that in a mood board, for example, in mid-journey. Then you can, based on the mood board and a very good instruction of what you want to see, make a new direction in that style. So you can work on scale in a tool as a mid-journey, very quick concepts. So then you look at layer one, so that's really the use of an AI tool. And if you look at layer three, then that's really about AI automation. Then I would look at tools like weevee.ai, or there are also other tools that I can do, but if you focus on weevee, what I really like about it is that you can actually make a visible workflow where you can connect different models to each other. So from the beginning, a concept, so for example a prompt or an image, then you can combine it with an AI model, for example nano-banana, nano-banana, so that you put that concept into the real world. Then you can make a moving image, so you don't go from an AI model in one tool to another, but you can do that in weevee, and you could also look at make.com or NEDN, so that's really automated platforms where you can take out such a process and connect it to each other. But then you can, based on a trigger, generate a lot of images on the scale, or ideas, so there are a lot of possibilities. - Yes, and how can you do that? I also notice that Prof. Nostroms also takes off because they are not happy with the images they get or with the scheme they get. How can you ensure that you get really good images and that you also keep your own handwritten? - Yes, that's right. What I see is that, in what kind of specialism you are also in, so if you are a photographer, then you know exactly what camera settings, what lens, what kind of camera, what kind of lighting, what kind of angle you want to use for a certain scene, to get a certain image you have in front of you, and in that way you also have to look at AI models. If you describe what you want to see in a certain way in a very detailed way, then AI models can also generate what you describe, and that can be done in chat GPT or in mid-journey, but they have a certain way to prompt such a model, in chat GPT that is different than in mid-journey. And what I often do is, for example, in chat GPT and GPT use, so that's a sort of assistant, so to speak, where you give an assignment to make a certain format of prompt, and then you don't have to, because sometimes prompts can be, I don't know, 500 words, then you don't have to write that out and go out to the doctor, but then you can, in such an assistant, you can write that prompt, and in that way you think that prompt is all, but that's because it's not a sort of a mix of random things, because then you still stay behind with a ugly photo, so I think that's a bit of the corner where you have to look at. Yes, so you actually use other tools to use the tool correctly, and if we go to, for example, renders, flat grounds, what are the handy tools in there? There are a lot, for example, in a SketchUp, there are a lot of AI tools, but what would you recommend there? What can you do there? Yes, good, it's funny, I think a week ago I applied such an online SketchUp environment, but I also worked with Blender, but what I noticed in Blender, for example, when AI wasn't there yet, is that you have a scene, and then you go to a computer, for example, for half an hour or a quarter of a render, in order to get an image, and then you see that there is a fault, and then that has to be renewed. And with AI you can not render the image, you already have that, of course that costs a second to make, because there is no material in there, no depth. If you put that image in a tool like the NanoBanana, this is just a very good tool, it's an AI model, actually. So if you put that in there, then you can, with the description of the materials, of the lighting, of the moment of the day, you can actually fill everything in, and then you don't get a result within a quarter, but within a minute, or within a half minute. So I think that's a very good thing, and NanoBanana is actually a model that is in different platforms, and of course it's easy to use it via Google, but there are also other platforms. From the platform, for example, if I was given that from a Funda platform to a render world, so I have a Funda platform, where I actually copied the images, and then used that in a model that can be described very well, so you need a model that actually reads an image and then describes what it looks like. I did that with, I believe, with JetGPT. So with a GPT model, I tried to describe that, and with that description, I make the part of the house, so you fill in a flat background from 2D to 3D, that works up to a certain height very well, but you notice that such an AI model often has trouble with, for example, whether there is a window to the ground or not, because it doesn't really know what a flat background is, so you have to describe that too. If you have an error, which we see a lot in Amsterdam, then it's difficult for such an AI to get that out, so there are still specific things that a AI, let's say, from a description to a generated image, can really visualize it well. So also the place where a door is on your wall, in a house, it has to be more, say, left and right is not necessarily a logical understanding of an AI model, that's just a kind of description of, okay, in that corner or in ... So it works up to a certain height, but I'm sure that if you spend a lot of time there, and I'm going to do that, then you can actually generate all the rooms at once from a single flat background in a certain way within five minutes, so that's a bit where I'm going. Professionals of course also have a lot of clients, they have to give presentations. What are the good tools that developers could be able to use to present to their clients? Yes, so you're really talking about a deck, say, a PowerPoint deck or something like that? Yes, I think that now a lot of people, or a lot of professionals are now, for example, in Canva or just in PowerPoint, to make a presentation, what would you ... Is there something that is easier and faster? Yes, that's a good one. I also thought about it today, because I actually want to take business assessments at companies, so that they have a certain input, right away with an AI agent who, if it were, is my copy, get certain advice about how to use the AI, and I also want to do that in a kind of deck, but I do want to do that with a certain effort, so that it looks professional. And what you see now, I have, for example, the Gamma.ai, that is a certain tool, software, with which presentations can be based on a prompt. Only there you see that if you put one prompt in that tool, then you get a template with a certain visual style back. I'm not really convinced yet, to be honest, so what you should do then, that works too. Then you're going to make a presentation, it can also be done with a Google slide, you can put variables in it, and you can actually generate those variables with an AI automation, so that's possible for each branch. And then you're going to fill your template, and that way you can generate automated decks. Yes, so that would be a bit of an A4 feature at this point, I think. Yes, I also notice that it is sometimes a bit of a search for what works well, and are you going to take the time to set up such a template, or is there going to be a new model next week that goes faster? Yes, well, yes, and that's nice that you say that, indeed. What I said at the beginning is that a lot of companies, of course, compete, so they fight each other to create the best model and the best tool. And I also notice that a lot of people are maybe crying about it, or just waiting for something to come, or something to come. This can be fixed, so I'm not going to focus on it now. But what you do have to think about, the tools that are already there, and the AI models, and how you can use it all, and how you can do it in a certain way, is already so far, that even if there is now a lot of GBD6, that doesn't change a lot of how we can do things now, so to speak. Specifically for the making of decks, I can imagine that there will still be a better tool, but for other directions, I think it's already very far. Yes, and for example, having a presentation, writing a project story, a good page that is visually appealing, how can you help with that? So it's really visual or textual? Yes, textual as well. Yes, I think that the writing of texts with AI always starts with your own tone or voice, but I think that's already said a lot. So how are you going to put your own tone or voice very firmly in such an AI model? And the easiest way, of course, is to use the majority of people's GBD, if you have a paid version to make a GBD that can be used to feed with your own tone or voice, that can be with documentation or a description of the way you are speaking, or your brand, marketing, and that kind of thing, and that you then write texts based on that, by controlling that assistant, and then you can get a lot of textual, I think. Yes, and I also see that a lot of people don't really use chat GBD well, like in all of their new chats, never working with those GBDs. Yes, that's true. The whole background can't be filled in. Everything else is still in the new chat, so it all comes together. So I'm also trying to explain to a lot of people that that's not the way you have to deal with chat, because then it doesn't go well with what you're going to get. Yes, indeed, you can not really write a whole book about that, but then you can really use a kind of cursory approach to how you use chat GBD specifically. Yes, there are so many functionalities that either have an effect on the output of what you put in it, or indeed how you can make it even better output, but that's also with the making of images in chat GBD. In principle, with a very simple prompt, you just get AI-like images. But if you know exactly how to get a prompt, you just get super realistic images that are not really different, with a prompt, which you can also see in a camera, but you have to describe that prompt. That's precisely the authenticity of what you see in the world. You have to describe that and then you get the right output, because otherwise you get a very flat image, or a very flat text, or with all those stripes, or something like that. Yes, it's also very much in the use of AI tools and where it often doesn't go well. Yes, it certainly starts there. Okay, project stories, so chat type, I think that everyone already uses that. And if it's further about workflow, about the business side, removing tasks that you have, making offers, planning, doing emails, what could that help? Yes, so I think that when you have a company, I work in the process automation, and then it's a bit the same. There are parts to our part, with a question, and that's actually an challenge. So that's, for example, a process that doesn't go well, or a task that it costs 10 hours every week. And they ask us, how can we automate this in a certain way? And that's what you have to think about your own company. How can I control these 10 offers every week in an automated way to build a sort of system, that in any case I had 100% in the first place, it still doesn't get per se, but if it's 80% or 50%, then that's already a lot of profit. And then you're going to look at, okay, how can I put that into a fairly simple way, so to speak. It doesn't have to be very complex, for example, if you use a tool as a make point, and I understand that if you get in there, it's a fairly complex environment, but you can even describe a certain automation with AI, so you don't have to connect that to each other, but you have to configure that platform before you do. So you just have to configure the kind of different parts. But then you can look at, okay, if I put a certain input in a certain channel, let it be WhatsApp, or I send an email to a certain email address, or I send whatever to a formula or to a different place. That data, so the information from that piece, or from that message, you actually want to catch that in such an automation. And with that automation, or with that data, you're going to do things in that automation, what you want to do. That can be, for example, that you want to make the name, the date, and the address of some of the offers you want to make, that you, for example, send that to a formula, the formula catches that data, sends that through to the automation flow, then it goes into a certain template, which is automatically generated. For example, it is generated to a PDF file, for example, that is sent somewhere, or it is automatically sent to the person who thinks, okay, that's where it has to go, but you can also make Excel, and then just put everything on a list to everyone who has to send an offer. And then it is also automatically picked up. Or you can also, based on an email you get from a potential customer, of course, it goes a little further, but that you then let that be transferred by EI. Okay, this email is about a paid application, there must be an offer for that, that's what I'm going to do now. Well, then it will also go through an automation flow, and in that way you will automatically do that kind of process. I think it's important to look at what is really going to save me a lot of time. Because I do have the idea that, of course, now online, with a lot of people, the idea is that, okay, I can just set up an automation flow here, or with this system, with all these EI agents, and then you really have a sort of, well, you see a dashboard with proofs of 100 notes, components, and they are all switched on, and that's what your whole company is going to do, that's just nonsense, really. It starts small, and with how much it does, and that's where the real value is, really. Yeah, and if we talk about an EI, just about emails, is that all that emails are automatically answered, or a sort of semi-automatic? Yeah, and you can set that up, or if you use Outlook, then you can use Copilot, for example. I would always recommend not to answer an email automatically, but in any case, to set a draft, so that you get an email that, well, let's call it an assistant. The EI assistant, he gets the email, he reads the email, he knows based on how you talk, how it has to be answered, also based on other emails, he sets that up as a concept, and you have to only speak in the morning or in the evening, when you also check your emails. Just reading and sending, that depends. That can also be pretty simple, so it's in order, yeah. Tell me, you're a designer, what would be a good step, what would be the first step? To start with EI, yeah. Yeah, then I would always start with, say, the demand within your company, or in your daily life, within such a company, okay, what do I do a lot, where I don't get the energy of, what do I have to do often, what does it take me a lot of time, that's all the progress points where you can think, okay, is there not just a system that can do this too? Well, that's where I'd start. And then you're going to look at something that provides the most value, so that's literally also how we calculate something within ABN. If there are ten of us, but we can only build two things for departments, then we have to prioritize it, and then it's literally asking those departments, okay, how much time will this save for how many employees, and can we then build a business case for that? And if, for example, well, I say something 100 hours per year is saved for an employee who has 80,000 employees, I say something, then you can calculate how much value that adds to the bank in this case. And so you have to think, look, I'm in my own company, if my time is so much value, let's express it like this, and I spend 10 hours a week here, then you can calculate that pretty quickly, okay, if I don't have to do this anymore, or for 50%, then that's quickly won, so to speak. So that's where I'd start. And then you have to see how, indeed, whether the whole process that you have to go through to, for example, send that offer or that invoice, whether you can automate the whole thing, or for a large part, and that you can actually count on yourself, because I think that's also a bit of a problem for people. It's not really visible for people to understand how much value it really provides, but if that really, literally, expels you with a small amount of money, and it's also not a lot of investment to automate it, then you won that very quickly. And that's how I think about AI, and that's how you always learn, like, okay, maybe I can take this on, oh, maybe I can do that, too. Yes, it's just a lot of use and do, and the addition, to be honest. And if we talk about, of course, a creative branch, inspiration, but also imitation, how does that look with Build of Aqua Design? In how far is something inspired, and in how far is it really being copied? Yes, that's a good question. I also did research there, of course. At the moment, it's still a grey area, but if you, let's say, generate a Mickey Mouse, because you've thrown all sorts of Mickey Mouse images in there, then it just remains a Mickey Mouse, so you don't have the right to do that. And as you can see with, for example, a design, furniture or something, if that really is a specific design of someone who has a certain right, and that's being copied into an AI image, then that's your right in my eyes. I'm not an advocate, but that's how I see it. Oh, well, an inspiration or, for example, a cover of a music, something that stays, that can, so it can, that direction can do. But if we talk about, for example, indeed furniture, if you have your own piece of furniture, which is really a design piece, but you want to show it in hundreds of different places, in hundreds of different interiors, but you don't have that possibility. Then you can do that yourself, if you are the owner of that piece of furniture yourself, then you can actually copy it in that way. And then it's copying your own work. And I'm actually very curious where there are people from the interior branch. Is there a lot of freedom or is there... Well, I'm not sure if there's a lot of freedom. I do notice that people, for example, are going to put AI images as a portfolio on my website, if you then think, well, I don't have what you made. You have this with AI, so you have it made, but it's not real. If someone can see that it has been made with AI, then I think it does its job on your own reputation. If you don't include that, I think you just have to include it. And then there is also not really a shelf. I have, for example, a location, that is an event location, those are empty rooms, they want to rent, but they don't have the money to make a conference room the whole day, and then another room, the shoot room, whatever. So you can do that with AI very beautifully, very realistically. No one will see that it doesn't exist. Then it's fine, then it's just a win. So that's the way you put it, the output. The output should just be great. How do you see AI over two years, the creative studios, the designers? How do you see the role of AI there in over two years? Very big, because I think the visual side of what we are now talking about, and that kind of thing, is that it's just going to get better and no more of real differences. You just can't do that. So very honestly, if you don't spend time on that now, then I think you're going to lose people who always spend time on that. And that's just the reality. You can see all that platform, like what you say, Blender or Sketchup, that implement or Photoshop, by the way, that implement all AI models so that you can verify something and then put something down with one prompt. Whether it is your own model, or another one, or whatever. And that's just going to get better. Until the moment when it's not better, but then it's not really going to be really different, so that's just how it is. That's almost how it is, and that's where it goes. And then I think it's also going to look at how available it is for everyone. And that's now reasonable, but you still need very complex prompts. But I also expect that, and you can see that with the GPT models. That started with the fact that you really had to write very complex prompts to get a top result. And now GPT can think of itself, oh, this question or this prompt needs a deep thinking from me. So it's going to take the first step and think 30 seconds before it gives an answer. And of course that's going to be in such a design branch. There are just models, where you have a certain image upload or a certain idea upload. And that's the model of self-knowledge. Ah, this is necessary. And then you get the perfect output. Or you can upload it to a certain field and that it then... That it is perfect. Yes, absolutely, yes. OK, and let's conclude. What do you think is the biggest misunderstanding about AI? I think the biggest misunderstanding is that AI is going to take over all of us. I think it's going to have a big impact. I also think that the misunderstanding is that people say AI is a nep and what I do is real. Yes, what is real and what is nep? I think that a big misunderstanding is that I'm not going to use AI, because I really want to stay, or I want to stay myself, or that kind of thing. Yes, then you're going to lose it in all aspects of the people who do that. Yes, if you look at how it all used to go with, for example, the Internet, people who haven't armed it. I don't know where, yes, they haven't won it per se by the people who have armed it. And that is now also, you see that many people are still excited. They don't want to use it. And then ultimately the people who survive, where a job is for them, as far as I'm concerned, is very black and white, of course, who use AI and other people who don't want to use it. They don't use AI, but they will lose it quickly. The conclusion, do you have one more practical tip for professionals to take part in? Yes, certainly, yes. I have noticed myself that through the years, the use of AI in what way, whether it is a tool or in another software, that it starts there. So start using it well in the use of a tool, know the features of a ChatGPT, know how to get everything out of it. Keep yourself a little bit at the top of the latest AI news. Not per se all the latest models, that's not so boring, but be at the top of the new AI feature in another. And then focus on a number of tools and be there just to prove the best in it. So if you take a journey, just know how to get the absolute best out of it. ChatGPT knows how to get the best out of it. Then you come up with a combined set of tools, as long as there are only three or four, you will get a very far end. And then you know what about AI in any case, and then you definitely have a advantage on a lot of people. Yes. Nice conclusion, thank you Stein Keller. No thanks. That was another of the Interior Club Podcast of this week. If you also want to apply with AI, go to our website and check our workshops. Thanks for listening and see you next week.