AI in Architecture: Top AI Tools Transforming the AEC Industry

Problems in the AEC Industry

We’ve all seen it. Rows and rows of drab uninspiring buildings, interior designers turning everything a mix of gray and beige (a.k.a. greige), and hulking modernist cubes that pretend a few square pop ins and pop outs make them “fun”. Even their designers curse these buildings when told they need to work overtime on the client’s umpteenth redesign request. So what might save us from this architectural inhumanity? Surprisingly…. generative AI tools.

From the cost, to the impermanence, to the inefficiency, to the overall drabness, the architecture and the AEC industry over the last 50 years or so has its problems. However longtime readers of this blog may be aware we may be on the verge of a new era in architectural design… an explosion in efficiency and architectural creativity fueled by generative AI tools in the AEC industry.

So how is AI going to tackle the problems so many people have with architecture and the AEC industry today? What is the case for why there should be no more boring buildings? Let’s find out.

Key AI Tools and Trends in Architecture

After researching and writing about AI in architecture for over a year now, here are a few of the key trends I’ve found that will impact the AEC industry:

Conceptual Design:

It should be no surprise that AI is making conceptual design much more accessible, as we explore in other posts. With AI tools like Midjourney and ControlNet on Stable Diffusion, you can imagine countless concepts on a given site with a few brushstrokes and the click of a button, as demonstrated by architect and AI educator, Design.Input. Thus, in the future when creative design choices are just a click away, there will be almost no excuse for boring architecture.

Note: Accepting cookies may be required for embedded videos to work on your browser.

It will also allow for much more abstract concepts. While text to image is the way most people prompt, the truly creative among us rely on image-to-image prompting. This lets them integrate disparate concepts like a breathtaking landscape, the foam on a latte, or a crumpled-up piece of paper to create truly novel design concepts.

There are countless image generators available now that can help with this phase, but for comparison of how some of the more established AI image generators perform, click here.

Making Program and Performance Decisions:

Outside the realm of concept renders, AI-powered tools like Maket and PlanFinder can generate an endless array of floorplan options to make design decisions [1] [2]. Another platform, Finch3D, provides not only layout options, but also performance optimization, enabling designers to generate and refine plans optimize their efficiency in terms of useable space, daylighting, etc. [3].

Developing Design Options:

AI can retexture and render designs in varied styles, materials, and color palettes, empowering designers to explore thousands of options quickly. With tools like Krea realtime generation, Midjourney Editor, and Stable Diffusion with ControlNet, designers can swiftly visualize variations, making it easier to handle redesigns without the time-consuming process of designing and rendering several more design options via traditional methods.

Until recently this was reserved for 2D images in the conceptual design phase. However, with parametric modeling alongside real time AI rendering like Krea, AI Revit plugins like Veras, and AI video-to-video platforms, some innovative designers are starting to bring AI into the 3rd dimension in later design stages. One example from Studio Tim Fu (a firm that focuses on using AI in architecture design) studio shows the full evolution of a tower concept from boring cylinder to pinecone-inspired, balcony-studded statement piece building while in another study his studio experiments with using text-3D AI to generate massing concepts, renderings, and physical models.

Note: Accepting cookies may be required for embedded videos to work on your browser.

Furthermore, AI can even identify potential design flaws by attempting to fill in the blanks of a concept sketch lacking in detail. This helps designers catch issues like missing windows or illogical roof transitions early in the design process.

Understanding Client Preferences:

Struggling to capture a client’s vision, despite hours spent on meetings, mood boards, and options? Tools like Midjourney’s personalization feature can simplify this process by adapting to a client’s aesthetic preferences in a user-friendly, algorithmic way.

In Midjourney, clients can indicate preferences by liking images or choosing between options, much like how YouTube or TikTok algorithms learn user tastes. This creates a personalization code that can automatically adjusts the lighting, colors, and styles of your renderings to match the client’s unique preferences, capturing subtle tastes often hard to communicate in a design brief. This tailored approach can help clients get excited and approve designs faster with fewer revisions.

Dark and Brooding Usonian Inspired Home Concepts by AcanthusAlchemist

Code Compliance:

Large Language Models (LLMs) like ChatGPT excel at scanning and explaining dense text, which could one day help architects review code requirements with much less effort—if they can overcome their biggest flaw: hallucinations.

A “hallucination” is when an AI confidently presents incorrect information, such as counting errors or fabricated sources. While AI can answer simple questions, you must always verify its answers to avoid errors. One way AI users are reducing these issues is by training custom models on more curated training data. Custom GPTs or LLMs based on vetted sources offer more reliable responses than general models, which may includes unverified data and might just respond to your serious code compliance question with a joke answer it found on Reddit.

One company, Perplexity AI is addressing this problem by training their model trained to consistently prioritize citation backed responses. It is more targeted at researchers and assisting technical professionals. This makes it more reliable, though as always backchecking is critical.

For comparison, we asked three free LLMs to answer a question about roof insulation requirements in California and looked up true answer separately in the 2022 California Energy Code. Here’s what they came back with.

This question was intentionally vague, but as you can see Perplexity did fairly well with the only citation-backed response and gave a technically correct answer for nonresidential roofs, though it missed many key details and said this applied specifically to wood framed buildings. It did calculate the correct effective R-value from the code’s mandatory U-factors though.

ChatGPT was also sort of right from a practical standpoint by suggesting two common nominal R-values for insulation products used to meet code requirements, but its response was full of unhelpful generalizations, it was missing citations, and it did not correctly report the exact language in the code. Lastly Gemini grabbed a number from the residential requirements, but didn’t report climate zones correctly and did not cite the source directly.

However, the most important finding among all three is that LLMs are trained to give you an answer, not ask the right questions, and there are many questions the AI should have asked to determine what type of building it is, what climate it’s in, etc. to determine which code sections apply.

3D Scanning and Field Measurements:

Measuring and drafting plans for existing buildings can be a tedious process, with designers often missing key details. Fortunately, several AI-powered 3D scan-to-floorplan platforms, such as Polycam, Scanniverse, Matterport, and LumaLabs AI Interactive Scenes, are making this easier [2][4][5]. These tools use advanced technologies like LIDAR, SLAM, point clouds, NeRFs, and Gaussian splatting to create digital twins of your project.

Though integrating these systems into your practice may seem costly, some are as simple as installing an app and taking photos. We’ll review the pros and cons of each in a future post.

Cost Estimating AI:

How often have you come up with a design you love, only to find out after pricing that it’s way over budget and will be hacked to pieces by value engineering? Well, that may become a thing of the past now that AI cost estimating is starting to hit the market as well.

While this technology is still new and not yet accurate enough to replace estimators, it can assist them by speeding up their workflow [6]. Tools like Togal.AI [7] automate tedious takeoffs, and the industry is already exploring how AI can support 4D BIM (accounting for time and sequencing) and 5D BIM (accounting for costs).

With faster, more reliable cost estimates, project teams could soon run cost-benefit analyses on numerous design iterations in record time (under the guidance of an experienced estimator of course), allowing architects to discover cost saving design strategies they might have never discovered based on intuition alone. Many assume that stark modern boxes with a few pop-outs to represent “motion” and “intersecting masses” are the most budget-friendly option, but AI tools may help project teams discover innovative ways of making more expressive and affordable facades. Thus, discovering savings this way may play a key role in solving the world’s housing affordability crisis.

The Future of AI Tools in Architecture

The world of AI is rapidly evolving. Platforms come and go and sometimes major breakthroughs in AI tools seem to come on a near daily basis. That being said, the AEC industry is notoriously slow at adopting new technologies so it is up to project stakeholders to experiment with these tools, influence their development, and conceive new ways of integrating them into projects.

If you would like to learn more about these types of tools, comment below about what you are most interested in and consider subscribing to our free Pixels to Plans newsletter below to be notified whenever we have new posts on the intersection of architecture and AI.

AcanthusAlchemist

Designer and engineer exploring the intersection of AI, architecture, and urbanism.

email: acanthus@pixelstoplans.com

Interested in exploring other AI generated architecture concepts and design workflows? Sign up for our free mailing list to be notified whenever we post new content here on Pixels to Plans.