AN AI TOOL SET FOR ARCHITECTURAL DESIGN: PART I
- Dan Edleson
- Mar 16
- 4 min read
How to Use New Technology Without Abandoning Your Process
This is not an article about how AI is making humans and existing software obsolete. The death of Revit, Grasshopper, Enscape and all the other programs you enjoy using has not arrived. Surely you have seen many articles saying this, but that narrative is not only a disservice to existing processes but to ones that can develop alongside AI. Rather than considering new AI tools as replacements for our current toolset, we would benefit to consider them augmentations – power-ups if you will, like Super Mario with the feather or the fire flower. They can allow you to rapidly advance if used properly, but if you run straight off a ledge you’re still doomed.
In this three-part tutorial I will take you through my process of leveraging AI in collaboration with my existing AI-free design process using a hypothetical project.
The tools we will look at could all be categorized as “AI”. However they are not uniform in what role they fill and the quality of results is still aligned to a coherent human-led process. These tools also, at least at this point, require patience and a willingness to discard many near-misses. For Architects this should be second nature; the architectural design process is fundamentally built on discarding near-misses in the effort for one coherent and distilled concept. After going through several design projects using these tools, I’ve found my process falls into three categories; Conceptual Design, Design Development and Visualization. Here I will discuss how AI tools fit into each. Some tools fit into multiple categories and I will explain how they are leveraged for different tasks. These are by no means the only AI tools out there for design. I get no commission and have no affiliation with any of them. However as a designer in the field of Architecture I am personally invested in their development and how they evolve the industry.
Here, starting from scratch and using a hypothetical project, I will take you through a rough outline of my process. The project will be to design a contemporary tower inspired by the Greek “Tower of the Winds”.
DESIGN DEVELOPMENT
For conceptual design I am going to use two text-to-image generating tools. The two I use are Midjourney and Dreamstudio. The reason I like them and why I use both is because they have different nuances in how they create images. With Midjournery I find there is much more variety – exciting variety at that – depending on the word prompts I use. I also like it best for blending multiple reference images together. Blending is a great way to give the computer a little nudge in the right direction. Let’s talk about blending for a moment because it is not as straightforward as just blending two images off of Google. To really get where we want to go, we need to do a bit of reverse engineering. For example if I use a word prompt in Midjourney of “A flying bird as a tower” you end up with something like this:
However what I want as a designer is less literal. I want a form that has the essence of a bird taking off, that has its structure, its energy. No feathers. I also probably want it to be structurally viable. If that is the case, perhaps I find a picture of a bird skeleton in motion and combine it with a structural steel tower. I can find these images on Google, but that can be time consuming. If I want one of the images to be of a bird skeleton flapping it’s wings how long do I have to spend on Google Images to get something that fits what I want? Instead I can use a word prompt and get this:
And then generate some towers with structural steel that evokes birds:
Once I have images of these two concepts – and I will likely have many variations I like – I can start blending them. Then I can find other images and blend them, or blend them with the images I generate, a blend with a blend if you will. After fifty or so generations, this was the concept I liked the best:
When I do get to a point where I feel good about the big idea, I tend to run into a dead end in Midjourney. Midjourney has so much variation in the images it generates that if I lock onto a concept I really like I can quickly lose it completely or get stuck trying to make refinements that don’t get me anywhere. Take this example below, where my goal was to do away with the beak of the bird still prevalent in the tower. After some thirty generations it was actually less elegant and further away from what I wanted:
This is where I usually shift to Dreamstudio. I have not begun design development yet, but I do want a little more nuance in the concept. Dreamstudio, using the Stable Diffusion engine, tends to expose the variance controls a bit more and this is great when you want to focus on refinement. It also can be nice for generating very interesting concepts off hand sketches, or even AI Generated images that have been gently sketched over. Using some really basic sketching tools (this is just markups using the Windows Snipping Tool) I can define the line I am interested in, then generate at a low threshold of variance until I get something that feels right.
So think of Midjourney as your ax, and Dreamstudio as your chisel.
Do you have other tools you use for conceptual design that I haven't mentioned here? Message me or post in the comments and let the other readers and myself know.
Comments