A few months ago, I went to the opening of a new real estate development. It was a grand affair. All the big names were there, rubbing shoulders and exchanging pleasantries.
Then came the big presentation — the inevitable slideshow of pictures and videos.
Except this one was different. They had used AI extensively to animate the architectural renderings. I suspect it cost a fraction of what conventional animation would have.
The video clips began.
Buildings and entire neighbourhoods appeared to fold in on themselves. People weren't to scale. The craziest things were happening in the background.
As heads began tilting sideways and eyebrows rose, the presenter announced, rather nervously: “Isn't it great what AI can do now!?”
The more bizarre the distortions became, the more often he said it.
The presentation wasn't supposed to be about AI. It was supposed to be about the development. Yet with every passing slide, his uneasiness about what was appearing on the screen became more apparent.
Had he not mentioned AI quite so often, many people might never have realised what was going on. Although the collapsing neighbourhood would have been difficult to miss.
The glitches were obvious to me, as they should be. But quite a lot of what we saw was remarkably close to “good enough” if you were in a hurry and working with a very limited marketing budget. Used extensively, some of those distortions could have landed a developer in trouble. For that evening, though, while it wasn't exactly setting the world on fire, it was a damn good hack job.
So is AI a flop?
Absolutely not. In fact, what that presenter was trying to do was significant. He just didn't have enough reliable information behind it.
Generative AI isn't precision 3D CAD. It's extraordinarily good at imagining what might be there — much as we humans do. That's what makes it AWESOME. It's also what makes it dangerous when the details have to be right.
Ask an image-to-video system to animate a single rendering of a development and it has a picture to work from, not necessarily a complete understanding of the buildings, streets or spaces behind the camera. As the view moves, it has to invent what it can't see. Sometimes it invents beautifully. At that launch, it was a mess.
Hollywood, of course, has always thrived on invention. AI can create breathtaking scenery, fill enormous scenes with crowds and, with the appropriate permissions, reproduce a person's likeness. The audience is there to be entertained, and the technology is becoming remarkably good at it. Actors, animators and production crews have every reason to pay attention.
But real estate is not fiction — hence the “real”. Pre-construction sales depend on representing something that has yet to be built, and accuracy is critical.
Years ago, while working on a prominent project in Sunny Isles Beach, I had a conversation with a respected, long-established property developer. He told me, in rather less polite language, that a percentage of buyers would “try to fuck you at settlement”. He built his sales process with that possibility in mind. Part of his preparation was simple: be accurate, be consistent and never mislead.
If you market a building using details that an AI model has simply imagined, you may be handing a dissatisfied buyer an argument that the product being delivered isn't the one you sold. Disclaimers have limits; they aren't a licence to make misleading representations. Why create that risk in the first place?
None of this rules out AI. It simply means understanding what you're asking it to do — and keeping a firm grip on the information that must remain true.
Four versions of the same building
Most people in real estate see AI as a cheap way to make marketing images. That's both true and false. AI can enhance imagery, and it can do it damn well. Pre-construction imagery offers room for interpretation, but the architecture being sold still needs to remain faithful to the design.
To illustrate the point, I went back to an image from Altura, a project in the Tampa Bay area. I started with the original drone photograph and the raw 3D building composited into it, with all the usual finishing work stripped away. Then I gave AI a series of deliberately simple instructions.
I couldn't help myself with that last one. There seems to be an endless supply of visualisations in muted beige tones, with so much jungle hanging off the balconies that you'd think the building had been abandoned to the forest. Are you one of those guys?
The exercise took much of the grunt work out of touch-ups and grading. But it wasn't a magic bullet. I had to start with a proper model and a convincing photograph. And I'd still check every AI-treated image against the design: using an accurate model as a starting point doesn't guarantee that AI hasn't quietly changed a window, a balcony or some other important detail.
What about turning a still rendering into a video? Certainly — but there are limits. Subtle camera movement can be convincing. The further the camera travels, the more information the AI must invent, and the greater the chance that things start to drift or fall apart.
Enhancing an animation that has already been produced in 3D is a different proposition. The original sequence provides a real camera path and continuity through the architectural model. AI can then help with atmosphere, people, lighting and other embellishments. The core of what the developer is selling has a firmer foundation, with the Hollywood effects layered on top. You still check the result, of course.
Less grunt work. More creating.
But everything I've described so far is enhancement of the finished image or video. If you're actually creating the 3D, AI goes much deeper into the workflow.
3D is hard work — far more than most end users realise. There are hundreds of small, repetitive tasks behind a polished result, and some of them can swallow absurd amounts of time.
When an interior designer gave me a fabric sample and I needed the material on a sofa to match, I could easily lose an hour getting it right. Today I can give the sample to AI and have it help generate the 3D material in a fraction of the time.
I used to spend days modelling unusual furniture pieces, many of which I'd never use again. Now I can supply reference pictures to an AI modelling service and get a model to work from — for a fee, of course. It still has to be checked and fitted to the scene, but the economics are completely different.
I'm doing less grunt work and more creating. My effort is going into the decisions that matter, and the finished work is better for it.
I'm not using AI to avoid doing the job. I'm using it to take the job further. Which raises a very valid question: are you using AI to produce something cheaper, or to produce something your competition can't match?
There's nothing wrong with reducing costs. But if you're using AI to churn out the same old product while everyone else has access to the same tools, what exactly have you gained? Buyers are getting used to more sophisticated experiences. Cutting corners and assuming you're the only one with AI is a good way to shoot yourself in the foot.
The game is evolving. AI is making work practical that previously wasn't economical. The interesting part isn't just what we can make cheaper. It's what we can now make possible.
What comes after the picture?
And things don't have to stop at the visual side we're accustomed to. One can't help wondering: what next? AI isn't going to make that decision for you. That takes a little crystal-ball gazing.
I did some of that myself, and arrived at something I'd considered science fiction for years — my personal 3D Nirvana. I started by thinking about what I would want if I were examining a new property or product.
I don't just want a rendering or a video. I want to be able to almost touch it.
I want to see it on my terms, from where I choose to stand, and in beautiful quality.
And I don't want to deal with someone else's complicated technology. It needs to work on something as convenient as my phone. No apps. No headaches.
I want it all. I suspect my clients would like much the same.
Then I thought back to my own days in property development. What's the one thing that matters most to a pre-construction project?
Sales.
You can tune every aspect of a project, but if you don't achieve enough sales quickly enough, the whole thing can come unstuck. I want the development on the market. I want the salespeople to have the best possible tools. And I want to know what the hell is going on.
Not just how many people opened a website, but what they're interested in, which residences attract attention, where enquiries come from and how the sales team is performing. A beautiful presentation is only part of the job.
A fully navigable, near-photorealistic environment connected to an intelligent sales system might once have sounded wildly ambitious for an ordinary project marketing budget. Until recently, it was.
But AI has changed the economics. By combining accurate 3D work with neural reconstruction, real-time web technology and a great deal of AI-assisted software development, I've been able to build the sort of experience I used to think belonged on the holodeck of the Enterprise. Not literally, of course — but a world you can enter and explore, rather than a camera path you're forced to watch.
It took years of accumulated 3D experience, experimentation, mistakes and a lot of code. AI didn't simply dream up the result and hand it to me. But it helped me develop and refine thousands of lines of software and solve problems that would otherwise have demanded far more time and resources. Without that assistance, I doubt I could have made it commercially practical.
The result is Spatial 3D: detailed environments that people can navigate on their own devices, without downloading an app. Behind it sits Discover, a sales platform that connects the experience to buyer engagement, inventory and the people responsible for selling the project.
Suddenly, architectural visualisation isn't just something to look at. It becomes somewhere to go — and a practical part of how a development is sold.
If you'd like to see what I mean, explore Xanadu — a 17-acre coastal resort concept brought to life as a fully navigable, interactive experience. Start from the air, then switch to walk mode and explore at ground level, right in your browser. No app required: https://archiform3d.com/projects/xanadu/ ↗
Explore the available navigable areas in aerial or walk mode, on desktop or mobile, directly in your browser.
So when someone asks, “Isn't it great what AI can do now?”, my answer is an emphatic yes. But making pictures cheaper is only the beginning. The real opportunity is doing things that weren't previously economically possible.
The game is evolving, and that freight train isn't stopping. The question is whether we're using it to produce more of the same — or to go somewhere new.
And, ideally, keeping the neighbourhood standing.
Stephen Bell | Archiform 3D
