Research

AI Image Competition: Dorsa Hosseinkhani Shares Her Insights on Round 2
I am Dorsa Hosseinkhani, currently working at PLP as a Part II Architectural Assistant. I’ve come to the practice after completing my master’s degree at the Bartlett School of Architecture UCL, where my research focused on the use of AI and specifically “How to Write Correct Prompts”.
Exploring briefs in different geographical contexts
In this competition, I’ve been exploring the capabilities of AI tools, particularly Midjourney, to generate realistic and innovative architectural designs. Whilst doing this, I’ve found an interest in the challenges and opportunities of applying the competition briefs to different cultural contexts, like my home country of Iran.
Round 2 asked us to create images that integrated new housing into a historic European city. As I dug into the brief, my thoughts immediately went away from Europe and instead to Iran.
Iran is a country deeply rooted in history and culture, with cities like Isfahan that host ancient sites such as the Masjed-Jame of Isfahan, which was built between 1644 and 1656. These cities are prime examples of how architecture can preserve and reflect cultural heritage. I wanted to see if AI could handle this complexity and produce designs that respected and highlighted Iran’s unique architectural legacy. Testing the brief in Iran felt like a natural choice because our historical cities are perfect candidates for this kind of challenge.
As a result, my competition submission for this round includes separate images from both a European and Iranian context.
Developer X is being tasked with rapidly increasing the number of homes available within a historic European city.
Considering the heritage value of the area, demolition isn’t an option. New housing stock will need to be innovatively added into the area in a way that maintains its historic value, allows for quick addition of sustainable homes, and regenerates the area into a place that not only attracts the summer-time tourists but creates a year-long community.
Your job is to create an image that will showcase the innovative housing solution to the local authorities, in order to convince them to approve it.
Create an image that is as realistic and true-to-life as possible.
Your test site measures 80m by 40m and already features a diverse range of historic buildings, including shops and homes.


Generating the Image
To capture the essence of Persian architecture, I focused on three main elements: materials, arch shapes, and open spaces. Persian architecture is renowned for its use of specific materials like bricks and tiles, its distinctive arches and domes, and the incorporation of courtyards and gardens which provide open, communal spaces. These features are not just aesthetic but also functional and deeply cultural. By specifying these elements in my prompts, I aimed to ensure that the generated designs would stay true to the Iranian style.
My final prompt was:
The scene should include the area is full of Isfahan heritage architectural buildings and also there are some new and modern tall buildings which keep the languages of heritage Isfahan architecture without dome. we need to see both types of buildings next to each other and combined in the that area, all building and houses are built with Islamic Persian tiles or at least part of them have them, the area has a river in the middle and people and tourists seat next to the river and singing next to the river, A highly detailed aerial view, super realistic, detailed, super beautiful –ar 4:5 –v 6.0
The results
Surprisingly, generating images that accurately reflected Iranian architecture was quite straightforward. Persian and Islamic architectural styles have distinctive features that are easier to replicate compared to the diverse architectural styles found in European cities. European architecture is widespread and varied, making it more challenging for the AI to capture accurately without extensive prompting and adjustments.


Are there biases in AI image generation?
I didn’t notice any significant biases in Midjourney when generating designs in the Iranian architectural style. The tool seemed well-equipped to handle diverse cultural contexts and produced results that genuinely reflected Iranian architectural vernaculars without resorting to clichés or overly generalised portrayals. This was a pleasant surprise, especially considering that just a year ago, during my master’s research, I noticed that AI tools struggled to give realistic results. Seeing the progress in how Midjourney now handles these diverse contexts is impressive.
While I didn’t encounter significant biases, if they exist, they should be addressed by expanding the AI’s training datasets to include a broader range of cultural references. Prompt engineering can help to some extent but ensuring that the AI has diverse and comprehensive training data is crucial for it to represent different architectural styles and contexts accurately.
The impact of using AI across geographies
I can see immediate benefits from being able to use AI to translate designs into different regional styles and settings. It allows us to create culturally resonant solutions and adapt our concepts to various environments efficiently. This capability is especially beneficial for international projects where understanding and respecting local architectural traditions is essential.
Generative Studies on European Vernacular


