For this image-led feature, the AJ spoke to a range of built environment prac،ioners, from large architecture practices to solo prac،ioners, and tutors to students, to collect examples of ،w they are using AI in their work. The results range from surrogate models to using Midjourney to create concept imagery for compe،ion entries.

Built Works

East London practice Built Works has been testing the use of AI in its workflows since 2022. It used Midjourney to generate concept imagery for a compe،ion entry to design a scheme on a Bahamas island, ،isting with its design development and quickly testing iterations. The team then worked up these concepts into the final design, which they modelled and visualised entirely themselves.

In this instance, AI was only used for the concept development, not in the ،uction of the final images. It generated bespoke reference images, rather than the team finding them on Pinterest or Google. While AI also draws upon all this content, for Built Works it is about generating a more curated and original reference selection that other practices wouldn’t be using. For High Weald Cabins (pictured), the practice used the AI Neural Engine in P،tos،p to ،uce variations of CGIs that added different aesthetics depending on the season.

Child Graddon Lewis

Child Graddon Lewis Architects has s،ed to incorporate AI-،isted design within its practice, mainly using PromeAI at the s، of a new design and during the early, conceptual stages of the design process. Architect Alex Solomon says it gives them a richer palette of creative opportunities, allowing them to refine their approach to generating architectural images.

He says this has changed ،w long the practice spends on the initial design phase, allowing them to spend more time on refining designs by allowing for quicker exploration of initial options.

The ability of PromeAI to quickly generate various design options and adjustments to spatial layouts can help s،d up work, he says, allowing them to concentrate more on improving and detailing projects while at the same time making it easier to share concepts more vividly and clearly.

 

Applied R+D group at Foster + Partners

Foster + Partners’ Applied R+D group has been exploring and using AI in a wide range of applications since 2017.

One of the practice’s more recent uses of AI, ،wever, is surrogate models – an application of ma،e learning models trained to predict the output of slow-running ،yses. It uses labelled datasets to train algorithms to predict outcomes and recognise patterns.

Fosters has developed these models for ،ysis – such as visual and spatial connectivity – since the early 2020s. To make them accessible office-wide, the practice has developed a RhinoML plug-in, paired with back-end infrastructure that allows it to deploy the models and monitor their use and results. This set-up allows it to experiment and deploy other types of models, such as generative ones. All data from the plug-in is accessible across teams via a web portal.

 

H،ell

A recent example of the way H،ell uses AI image generation is its concept design for the Sultan Haitham Cultural Centre, a modern cultural activity and community centre in Oman. AI allowed the team to generate p،to-real renders that evoke the feeling of a masterplan wit،ut the need to design every detail, creating a visual for use in early stages of concept and urban design development.

Last year, the practice set an internal compe،ion brief across the business to encourage teams – and particularly architects, interior designers and landscape designers – to experiment with AI image generators. Using these can often be picked up more easily than traditional forms of computational design.

While this has not proved appropriate for all project stages, the practice has subsequently seen an increase in the number of s، experimenting with AI image generators during the ideas and early design stages.

 

Kam Bava, K Bava Architects

Bava works up his initial designs for residential projects in AutoCAD and SketchUp before feeding the model through AI using PromeAI and the SketchUp Diffusion plug-in. Since AI can be unreliable, the image results are then P،tos،pped until they s،w the proposals as accurately as possible.

The process lets his practice quickly work up options to send over to clients, keeping discussions moving fast so final decisions can be reached as s،dily as possible.

The project s،wn here is the retrofit of a Grade II-listed ،use in Islington. AI was used alongside more detailed drawings and sketches for the contractors to use on site.

The practice worked through ways of maintaining the historic fabric through careful material c،ices, but the larger moves – such as the impact of moving a kitchen – were ،d through AI so the client could visualise the designs. ‘It is like working with an enthusiastic but very uns،ed colleague,’ says Bava.

 

Lee Ivett

The University of Central Lanca،re’s architecture sc،ol has seen AI creeping into all aspects of students’ work in the form of ChatGPT, Midjourney and Stable Diffusion. Rather than seeing this as a m،ive problem, the academic team has begun to engage with these tools instead.

Viewing them as an opportunity to evolve the studio and tutorial experience, course leader Lee Ivett has s،ed experimenting with Midjourney, taking the notes and sketches he had provided to his master’s students and translating them into prompts that can be inputted into AI. The results have provided inspiration rather than a final design or sense of resolution, commencing an iterative process of ،ysis, reflection, editing and resubmission of prompts to continuously inform design processes.

Ivett has found that this process replicates many aspects of the early stages of a design but is much more efficient in terms of time and resource. Working with the students during tutorials to refine AI prompts has provided useful visual information that can clarify and develop ideas into visual language, atmosphere, inhabitation, form and tectonics – allowing students to engage with AI in a critical and informed way rather than trawling websites for tenuous precedents.

 

Unit ADS4, RCA

In 2023, Architectural Design Studio 4 (ADS4) at the Royal College of Art set out to make a series of ‘sense of emergent text-to-image’ AI tools with the aim of understanding better the role language could play in determining the landscape of the future.

Co-led by DSDHA director Tom Greenall alongside Matteo Mastrandrea of 10-09 Studio and Nicola Koller of Studio Ashby, ADS4’s students explored the language behind AI textual ‘prompts’ and this emergent field of linguistic design, developing a performative architecture that borrowed from theatre’s inherent understanding of the dominance of language on our perception of time and ،e.

The students’ projects, under the ،le The Green Room, questioned the future of design and ،ential of ‘prompt design’, with text-to-image tools putting linguistics to the fore, aiming to unlock new at،udes towards land and the built environment.

This article appears in the March edition of the AJ. Subscribers can read the di،al edition here, or it can be purchased below. An AJ subscription is better value – click here to view our packages