Perhaps I’m not qualified to comment on this as to do so would require access to the brains of many different professionals, all at different stages on a journey, some working together and some thinking alone.

However, from what I’ve seen, it is in architecture rather than in planning, urban design or heritage that the tools under the genAI banner – and particularly the image-making “diffusion models” such as Midjourney – have found their first adoption.

At least, this is true within the placemaking nexus. Look outside it to adjacent fields and it’s perhaps the digital artists who have been most violently shaken up. Some, in response, have thrown themselves at these tools, bringing highly advanced 3D and virtual production skills to bear, whereas others have avowed never to touch genAI as it’s corrupt, robs people of the results of their labour and undermines the whole industry. For anyone interested in analysing the impact of genAI on an industry, digital art would surely be a good place to start. And why? Because the loss of work was immediate.

Take a step sideways back towards our own little camp and we arrive at the architectural visualisers. How are they doing? Have they lost their work? What will they do?

An ecosystem of tools is rapidly springing up, from the already old-school-seeming Midjourney and Stable Diffusion to newer software such as Krea, Magnific and Runway. These constitute part of a long list of tools that can create images from text, from other images, can upscale and introduce new details, create short moving clips and so on, and this is to say nothing of ChatGPT and other large language models that are primarily text-based, from Claude to Mistral7B. Inherently, there is a certain imprecision or impressionistic quality to these image generators. They are not like rendering a hand-crafted 3D model, where every detail can be controlled.

Does that matter? Perhaps not. If you need to sell a proposal or provide an illustration of the kind of result you’re going for, these tools are already good enough that responsibility for creating the images is shifting.

What we don’t have yet are integrated platforms capable of taking in all relevant site, regulatory and client information and “solving” it like a big equation, spitting out all possible solutions that satisfy the constraints while the architect, planner, conservation officer and urban designer sit around a table clicking through until they find one they all like.

However, the kind of browsing for ideas one might once have done using Pinterest or Google has been made more powerful by genAI, more specific to one’s use case, and the speed with which an emerging concept can be visualised has also increased. The model needed to constrain image generation only needs to be quite simple. Indeed, it doesn’t need to be a 3D model at all. A well set up image-to-image workflow can provide a convincing-looking picture of something, but still, that’s the catch: it needs to be well set up. And it doesn’t yet help much beyond the imagination stage.

One thing is clear. Within the placemaking nexus, there is very uneven engagement. This, I think, is fine for the time being, but, much like a farmer casting glances at the clouds, we need to keep one eye turned skywards as we go about our business.