I Finally Found an AI Workflow That Actually Fits Archviz

So I've been avoiding this video for a while, not because I didn't know about the tools out there, I did, but every time I look at what other people were doing with AI in Archviz, it felt like just chasing hype, you know, framing it as this AI is going to replace everything kind of thing and I just don't believe in that. So I waited until I found something that actually fits the way I work, where the tool assists the process, where there's actual control and I think I finally found it. One tool that is going to assist us in pretty much everything we need in terms of visual communication for architecture. I found Magnific. You might have known it as this powerful upscaling tool in the past, but it has evolved a lot since then. It's now a platform that gives you access to the best AI models out there for image video and of course still upscaling with just one single subscription. And the cool thing is that it wraps all of that up into one unified workflow, especially with spaces, this infinite canvas where you build your own multi-step workflows. This is very, very powerful. And yes, this video is sponsored by Magnific. I had been hearing about it for a while online, but also people were coming and asking directly to me if I was using it. And coincidentally, as I was entering this architecture competition recently, which I'll definitely make a video about it soon, so be sure to subscribe to the channel, Magnific reached out. So the timing was perfect. I had a real project with a real deadline and competitions are even tougher and then actual deliverables. It's just the right conditions to actually test it. And what I found out is that this is genuinely useful. Definitely not as this one click result sort of machine, but something that fills specific gaps, speed things up where it makes sense to do it. And most importantly, keeps you in control. So that's what this video is essentially a super tutorial with a bunch of practical stuff, things that I've been testing and doing in the past few weeks. I'm going to place timestamps down below so that you can jump in and go back into whatever's most relevant to you. No slop, I promise. Let's get into it. So Magnific is a web-based platform, so there's nothing to install. You're just going to create your account, choose a plan that fits your needs, and everything runs here directly in the browser. We're going to start with the image section. Here we have many tools to choose from, and they can all be very interesting and more might pop in here in the future. But most of the things we'll do happen here under the image generator. Add people to images. So say I have a finished rendering here, a finished image, and I'm just doing some lightweight post-production in Photoshop, adjusting colors and so on. And I want to add a person here. Instead of just placing the entire render into Magnific, I would say that it's best just to add the area you want to edit. So I'm going to grab an area here with the selection marquee tool. I'm going to duplicate it and make it a smart object. Then go inside, Ctrl-A to select all, then Ctrl-C to copy everything. And then with your mouse hovering here on add, you're going to hit Ctrl-V. Nice. So it's now added here to Magnific, and doing this way, you didn't have to spend much time saving this as an image, which might be little, you might think, but I guarantee you it's a time saver once you get into the rhythm or just Ctrl-C and Ctrl-V between the tools. Now here in the prompt area, you can be as specific as you want. And I also don't believe into selling you this magical prompt. It really varies on the demand you have. But what I can tell you is that you can use GPT or Cloud to help you out with this, or then with the assistant here, which we'll get to see later in spaces. Essentially describe what you need, how many people, the action, clothing. And you can be as specific as you need or let the model handle that. But what I do recommend is that you use key phrases like don't change the camera angle, maintain image proportions, keep architecture intact, don't change materials. It doesn't have to be this specific way of writing, but this idea. For these crops that I use, I do auto as the aspect ratio, 2K on the quality. This is more than enough. And then depending on the plan you have, for now, at least Magnific, they don't cost in generations, which is quite amazing. And right now, NanoBanana 2 is the strongest model for image generation, but I wouldn't get too fixated on that. The whole premise of Magnific is that when a newer model drops, you're going to already have access to it. So I mean, maybe in a few months, it's going to be NanoBanana 3 or NanoBanana Pro 2 or something like that. That's one of the biggest advantages of using Magnific, just one subscription, as I said. Now hit generate and it's softened fairly quick. But the good thing is that you can start working on another generation right away while this one processes, or even just do a slightly different version so that you have more options to choose from, or even with another model like the NanoBanana Pro or the GPT model or something like that. And this is very efficient when compared to the chat version of these models. Okay, we have it here. This looks really good. And instead of now downloading this as an image file, and then again, cluttering your folder with the many generations that gets often made, I like just coming in here and then just copy image. Now we can go to the smart object in Photoshop, Ctrl V to paste it, align the borders and make sure that that matches somewhat perfectly. And then I like to go then here one step further and just mask the people out. And then you could even then hide the base layer. And this workflow here means that the base render will always remain the same or most of it at least. You will avoid making your images go through the many iterations and generations inside an AI model and start losing the details, which is quite common. And so you can use this smart object workflow to grab each part of the image you want to enhance or tweak and be very precise and specific about it. And this also means that in the future, if part of the design changes, you know, a client requests something different, you can always then re-render from your software, then just replace the base render here in Photoshop. And you're going to be pretty much done. Another way I like to use this workflow of adding people is coming in here with 3D people already in place, you know, from the 3D software, that way I can guarantee scale and positioning and only adjust then obviously realism, which is much needed texture characteristics and make sure that I have, I treat this as something already existing. I'm just improving and not creating from scratch. Or then even treating this as a typical cutout and even adding motion blur here in Photoshop. I mean, obviously you can do that already with the AI model, but here in Photoshop, you have more control and you can do this non-destructively, you know, more blur, less blur, you can do everything here. Imperfections and enhancements. Now with the same thinking of smart objects and being very precise about how you change, how you control the change with your images, we can use the image generator to add imperfections to certain parts of the image. For example, the side table, or let's do together this concrete column here. And a tip I can give you here is to be very careful with the intensity in the prompting. So I recommend using words like subtle or soft and others so that it has a bit of wear, but it's not something destroyed. Just a little bit of imperfections, that's all we want. These are things that you could obviously do with decals or advanced rendering engines with material mixers and all of that. But using AI specifically here allows you to move faster during rendering and modeling and only attack specific parts of the image when you need it. Same thing for vegetation. You can enhance existing vegetation with a quick prompting here and only enhance the realism or even modify it completely. Just change the style of the landscape altogether. Though I find it better just to use it as a simpler enhancer. For me, that's the way to go with the architectural visuals. And most of the rendering engines nowadays from the simplest to the more advanced are providing AI enhancers, but they are often limited. So having the ability to do here with all of the customization you need feels a little more like the red movie, you know? And by the way, I'm Oliver. I run Upstairs, an educational platform focused on architectural visualization and representation. This is literally what I do. So I wouldn't be here if I didn't think this was something genuinely worth showing you. And if you want to start using Magnific today and try everything that I'm talking about in this video and a lot more, you can check out the first link in the video description. So now let's keep going. Variations. This is interestingly useful and very applicable to architectural visualization. Think of it as expanding the range of possible results. It's essentially an image generator with specific built-in prompts and templates. It's great for quick testing, but I find it even better to just run something like this over spaces, which I'm going to show you later in this video. But I just wanted to mention this here because it does exist and it's fairly quick to use and it does give us some interesting results. And of course, one of the most used ways of taking advantage of AI is as an upscaler. Here, you can use it in two ways. One, in a more creative way, though I don't think this is really the way that Archviz benefits too much. Or then second, with the precision mode, where you can just enlarge an image maintaining its fidelity. This is really where the gold is. And then even with the precision mode, I see two ways that we can explore this. One, with the intention of speeding up the rendering process. So for example, if you're in a rush, you can just get an HD image, just like, you know, 2K resolution out from the rendering engine and then quickly pass it through the upscaler, like a good upscaler here with Magnific, to get it to 4K or higher. And then two, if you already have a 4K resolution, something like a final image, and then you want to push it to 8K maybe, or sometimes even more, to do specific crops of your image as sort of like detail images, you know. This is great for mood boards, Instagram posts, or just sharing specific details about your image with clients, competitions, portfolios. I use this a lot. Now, you can definitely explore more of these tools inside the image section. There are so many options to choose from. But let's move to video right now to make things move. Video. Now, we're very used to just showcasing and communicating architecture, especially unbuilt spaces, with still images. Videos are often very demanding to make, even though recently it's been getting easier and easier with real-time renderings. Still, it's kind of consuming. But with AI for video, things start to get a little bit more interesting. So here under the video section, go to the video generator. But, and very importantly, we're not generating your video from scratch, from a prompt, pretty much ever, I'd say. We always start from a frame, from an image. This will guarantee more consistency and control in your generations. In terms of models, I know that at the moment, Seedense 2.0 is getting a lot of spotlight. But I found out that I had more success with Kling 3.0. It does what it tells you. It has a little bit more precision, I felt, while Seedense, or Seedense, well, does what I think it looks good. So it's a little bit more creative. I mean, that also can change in the upcoming months, for sure. For Archviz, I need precision, plus Kling 3.0, it's a cheaper model at the moment. But I mean, feel free to test all of these out. As you can see, so many models, and each one does something different. Simple clip. I like using this for a simple, static camera with people and vegetation moving, so the scene feels alive. You add a start image, your render, and then the trick with the prompting is, again, to use subtle words, like instead of windy or wind, use breeze. Use words that are subtle and natural, especially for people movement. It makes a huge difference, as you're going to see. And you can give it a twist and even ask for some camera movement, too. But I would say keep it cinematic, in a sense, like keep it subtle. And the biggest pro tip I can give you here is just study real architectural showreels for reference, like the Local Project, EST Living, OpenSpace. That's the kind of movement you're going for. For the prompting itself, I always recommend Assistant inside Magnific or just going directly to Cloud or GPT to build something that's a little bit more detailed. But I would say that still keep it simple. Sometimes if you add too long prompts, it confuses the model. So after generating, this looks amazing, right? It always blows my mind how good these generations are. It keeps architecture and just only moves what we need. Six months, like one year ago, this wasn't really that possible. And another pro tip I would give you here is that credits run really fast with video. I mean, as you can expect, because it's a bunch of frames, a bunch of still images put together so that it creates a video. So I would say that do generations with 720 or 1080p and small in duration, just for testing, make sure that everything works and then go longer if you need it or higher quality. And then you can also use the video upscaler to get to 4K. But often from what I've tested, 1080p, sometimes like social media and like specific deliveries, sometimes it's enough, you know? And then remember, obviously, that you can be very specific with the prompting and ask for certain movements on people. And I would say go with no audio for most of the things, at least most of the generations. And then you can always add that later in the video editing process if you're going to be composing a showreel. Time-lapses. Now, time-lapses are one of the coolest ways to utilize video generation. It's just static cameras or just even like subtle zoom-ins or zoom-outs. But the idea is that we focus on the building being used instead of focusing on specific actions or specific individuals. It's about showing scale and atmosphere and not really a movie set, right? I think time-lapses are one of the best ways to really communicate architecture visually. Start and end image. Speaking of time-lapses, something very powerful is utilizing the start and end image. There are honestly so many creative ways to use it for architecture. You just need to understand the concept and then you go crazy. I mean, obviously, it's very simple, but start image is the first frame you're doing video and then end image is where the video is going to end. Simple as that. And then AI is going to fill that gap. So for example, you can get a day scene render, then use the image generator here and then ask it to make it a night version, or then you can even render a night version. Then place both as start and end image or vice versa. You can change this and then prompt it so that you have a time-lapse transition from one to the other. This is very, very useful, very cinematic, and can definitely fit into an architectural show reel if you used right. This is a really good storytelling tool. And you can do the same with, let's say, an empty lot to build architecture, but this already touches a little bit on the slop area. Same thing here for an empty room to a furnished room. A bit of slop, if you ask me, is this really necessary to the story? What's the purpose of this? But I wanted to show you, there's so many possibilities. For example, just the fact that you can be very specific with things like, let's say, opening a solar shading system, this start and end gives us a lot of control over the output of the video. A little less margin for hallucinations, especially with the correct prompting. Spaces. Now, everything we saw so far in terms of tools is also here in spaces. Spaces is the biggest strength of Magnific, I would say. It really completely changes how powerful this platform is. I'm going to show you a few examples of how this works, but the reality is that we could even make a full course on the mechanics of this tool. And if a course would interest you, actually, let me know in the comments below and we might even do that. Now, let's start with lighting scenarios. In this first one, I'm going to show you what kind of step-by-step. So we start by coming here to spaces and let's create a new space. This is an infinite canvas that works with nodes. I mean, sounds complex, but very simple to understand. And you can start from nodes or just drop in an image directly in here. That's what I'm going to do. So let's say I wanted to test a few lighting scenarios based on this render here. Something I do a lot and to give me ideas or directions I can go, I would do this manually in the past, but now it's very fast with this tool. Instead of just using a text prompt, like something ready, let's right-click and add an assistant and type something like, give me 12 exterior archviz lighting setups across main daylight scenarios. And then obviously more details here. And then I'm going to ask it to export as a list, which then gives me the 12 individual prompts that I asked. This is pretty cool, right? And so now we can right-click and add a image generator and feed the list into the text input and the base render into the image input. Like this is kind of the logic you've got to understand. And so it's already set here to 12 on the outputs automatically. And this image generator, it's the same one that we were using in the image section. I'm going to do a nanobanana2, auto aspect ratio, 2k, and then run. Wait a bit, 12 images are being generated, but overall this is very fast for the output we're going to have. Very effective. And after a little bit, look how amazing this is. It always blows my mind. And with this in mind, obviously I then go back to the rendering software to recreate a similar atmosphere. I use this more as a reference and ideation tool and not so much as a, as a final result, though you could use it if you want it and then just run this to like through an upscaler. And the best part for me is that since this space is already set up, I can just swap this base image, hit run again. And kind of instantly, I mean, after a little bit, we have a new set of scenarios for a completely different render. That's the power of building these workflows once and reusing them. You see where I'm going with this furniture replacement test. Now this one is also kind of mind blowing. I added three images here, like these three different lounge chairs in white background. And I wanted to test them in this rendering here, fed them into a list object. So it's organized and then added a crop of the render I was working on. And again, crops are just easier to specify things as prompt and just avoid like bigger areas for hallucinations, but you can test it with a full render as well. Then I created a text prompt, like replace the, the selected chairs in the interior scene, and then make sure to use the at symbol to link the chairs from the reference list and the base, the base render as well. So that your prompt is even more specific. And then you can go as detailed as you want with the as well. Like use assistant, use cloud, use GPT, whatever works for you. But sometimes like simpler prompts also get the job done, especially with the at symbol connecting everything. And then you feed that text and the image list into an image generator, then a banana to auto aspect ratio. One K for this one is enough as well. Just to, just because it's just quick testing and the output comes back as a list. Amazing, right? Quick iterations, very consistent lighting, correct color material throughout. I mean, this is, this is really nice, right? Obviously you don't have as much control as in 3d, like specific rotation of the chair and so on and changing like specific materials. I mean, you can do that with prompting, but I mean, I think it's better for just quick testing. This is really incredible. Of course, if you want it to use from here, you can just upscale this, just paste it back in Photoshop and take it from there as a typical post-production use masks and kind of stitch everything together. And I mean, the level of detail on these furnitures coming from an image, very hard to find assets like this. They're both very realistic and not that heavy to work with, especially for a typical architecture firm. And well, from this workflow, you can, of course, then upscale, download everything, do video generations, iterate even more. You get the idea. Reframe. As I said, the options are endless. So we can take that reframe workflow that I showed you before with the image section and then do it here manually in spaces with a little bit more customization. From what I've tested, to be honest, this is one of those things that sometimes works, sometimes doesn't. Are you going to need a few tries? I mean, that's, that's for sure. And you're going to need to be tweaking the prompting a lot. But for example, the ultra closeup material study of the chair, I mean, this looks really good. Now the medium shots, maybe not so much. I mean, hallucinates a bit. The top down view is very interesting too, but whatever you can't see is going to just invent. And I added even like a closeup screenshot of the lamp specifically because it kept getting wrong before. I mean, you get the idea, right? Base image, prompting, reframing, and be very specific. If used from the image section, it comes with pre-built templates. From here, you can use the assistant or cloud to kind of help you imagine possible directions you could take here. And you start connecting the nodes. And I know that this kind of looks complex, but once you get the hang of it, it's really just playing with it and just kind of building this as you go. Character placement, nice example here. And I think this one is the one that makes sense to end this because it's about taking control over the result and kind of using this as more of a complete workflow. I'm using two characters here, a woman and a dog, and then two images here, which is one exterior and one interior, just to show you that you can do everything here together in the same space. And then with a simple prompt, I ask it to place both on the render. And then I reference them. And look, they're already in the image, very consistent with the original characters. The scale sometimes is hard to adjust, if I can be honest. Sometimes I find it better to just do it with the 3D people and just kind of just replacing them. But just to show you that this works. And then from here, you can fit this straight into a video generator, ask for a cinematic work with some natural movement. And there you have it. I mean, pretty great, right? You can do this directly with the individual video generation, for sure. But spaces give you this overview of the entire process, especially when you're working with consistent characters from beginning to end, or especially when you're working with multiple scenes and you want to treat this as a combined workflow. And honestly, another strength of spaces is that you never have to save anything to your desktop. You can copy and paste from Photoshop directly in here, grab stuff from the internet, Google, and then go back and forth. The clipboard keeps everything moving fast and everything you generate, everything that you're using gets saved to your assets automatically. All versions, the full history, it's always backed up. And then you download the final versions you need or just the ones that you want to keep. All right. I know that was a lot of information and we could go even further here, but I suggest now going to check out Magnific yourself with the link in the video description. And let me know if you'd like to see more videos on this in the You know, I've been pessimistic about where things were going or rather are going with AI and still am to be honest, but I truly believe this tool doesn't replace the thinking. That part still on you. There's a lot of hyped up stuff online saying that AI is going to replace the traditional workflow, that the core skills of visual community architecture are not relevant anymore. And that's just not the case. You still need consistency. You still need control. And Magnific comes in as this creative assistant that helps you move faster, iterate quicker, handle specific generations or adjustments. A bit of post-production help for sure. And for video generations, especially, I think it's amazing. Now, let me know in the comments what you think about this. Have you used it already, plan on using or not using it? I really want to hear from you guys and I'll be looking at the comment section. I'll catch you guys in the next one.