This drone shot doesn't exist. There's no drone, there's no footage, and it's not a real building. It all started with a Google Earth screenshot, which is kind of crazy if you ask me. In this post I want to show you how I got from one to the other, plus two other AI workflows I've been using and really like.
I'm the type of architect who doesn't really know how to feel about AI yet. Sometimes I like it, sometimes I don't. I want to learn as many tools as possible to communicate better as an architect, but I don't want to use tools that strip away my creativity or take the fun out of my job.
When I was putting these workflows together, I asked myself three questions, and every workflow had to pass all three:
I threw stuff out. Some workflows didn't make it here because they didn't pass that filter.
For everything in this post I used Magnific, a platform that brings the current models you already use, like Nano Banana, ChatGPT and Flux, into one space. Instead of paying for a lot of different AIs that each do one specific thing well, you subscribe to one and can work with all of them at once. You can paste the prompts into whatever model you're already paying for, but in Magnific each output feeds the next one, which is why I work there. Download the free prompts I used here.
For this we need three things: a Google Earth image, our building inserted into that image, and a boost in image quality.
I go into Google Maps, search for my site, tilt the view until it fits really well, adjust it slightly and take a screenshot, like the one in Image 1 below. A high-quality source of satellite imagery would be much, much better, but I'm working with the level of quality most of us have access to.
Then I go into Magnific Spaces, create a new Space and drag my image in. Many of you are probably used to using AI in a chat format, which you can also do in Magnific, but for this I prefer a node-based workflow where I can see the prompt, which images it's referencing, what I want out of it and how to repeat it on other projects. I add an Image Generator node and paste in the prompt.
Before generating, I add some context. The prompt tells the AI what to do, but I also want it to understand the location so it generates similar images. So I create a text node with where the image is located, the architectural style, some geographic features and more details. You can copy that structure too if you want.
My first try invented a whole street that doesn't exist, and the scale of everything was completely off. What fixed it for me was making the prompt shorter. Giving it a ton of text and a ton of restrictions actually made the image worse. After adjusting the prompt, it cleaned up the image really well. You can see the node setup in Image 2 below.
I had this project modeled in D5 Render, so I pushed out a quick render from more or less the same perspective as the aerial image, with the materials and lighting adjusted for it. I dragged the render into the Space, created a new Image Generator node from the last image, connected the building image as a reference and pasted a pretty simple prompt. It did the job on the first try. If it inserts the building bigger than you want, double-click the image and go to Edit, where you can either write what you want ("can you make the building bigger?") or select an area with the retouch brush. Just writing what I wanted worked really well for me.
Before upscaling, I readjusted the composition a bit with another Image Generator node, asking for volumetric cloud shadows everywhere except on the building, so the viewer's eye goes directly to it.
The last step is to add an Image Upscaler node. You can test the Precision or the Creative upscaler, and I got completely different results with each, but I really liked the Creative upscaler optimized for photography. Image 3 below is the final result.
Going back to my three questions, this one is easy. It's not replacing a design decision at all. What it's replacing is hours of modeling a context I would never model by hand, or buying a drone, or hiring a drone photographer, which would not be a good use of my money.
For the second workflow, I converted the drone image into a day-to-night timelapse. It's also three quick steps: make a night version of the image, connect the day and night versions, and upscale. For this one I tried the new Magnific Desktop app, which has all my current projects and workflows and works pretty much exactly like the website.
This is the kind of thing you show a client when you need them to get it in five seconds. It's not replacing any meaningful creative decision. It's helping me explain an idea I already had.
The last workflow addresses a problem most architectural animations have. My D5 animation of this project is pretty good, but it still doesn't make you feel inside the project or understand what it actually does. For that, we usually need shots that some rendering programs just aren't specialized in.
Here we need three things: the media from the previous shots, a character who can thread all those shots together, and a really good prompt that acts as the director. Big caveat: this one consumes a lot of credits, so I'd leave it for the final stages of the project, when you're presenting to the client and you're sure the animations you import are the final ones.
You can see shots from the result in Image 5 and Image 6 below.
I was amazed by the result, not because I didn't think AI could do it, but because I was able to see my project differently and understand it better. The design is mine, the renders are the ones I did and the framing is the one I chose. All of that already existed, and obviously I can't film people inside a building that doesn't exist in such a realistic way. AI really helped me tell that story.
These are the three workflows I found useful to communicate ideas better without compromising an architect's creativity or design decisions. It's very easy to get carried away and start generating AI slop just for the sake of using a tool, which is not the point at all. If you're wondering whether to use AI in your own projects, my three questions can help, and be honest with the answers. Some of them might be yes, and you might want to restructure your workflow so it's not solely AI dependent.
Thanks to Magnific for making this collaboration happen. Everything you saw here was built inside Magnific, and you can sign up for Magnific here.