Mastering AI Prompt Writing for Architectural Applications

Staying ahead of the curve in architecture and design, means adopting AI in your design process. Software tools like Midjourney enable architects and designers to realize their creative visions with precision and effectiveness. But asking these tools to create perfectionism requires knowledge in prompt writing. Architect and AI specialist Tim Fu guides us through the process of crafting the most concise and clear AI prompts.

With a background of working for Zaha Hadid architects and now having his own studio, Tim Fu knows what he’s talking about. During his Clerkenwell Design Week Talk at our showroom, Tim guided us through the different versions of AI prompt writing. We’ve listed a couple of guidelines for good AI-prompting below.

Good Prompt Writing Starts with Understanding the Diffusion Model

To really grasp the secret of good prompt writing, it’s important to understand how AI is trained. The huge developments AI has made in the last two years is due to the diffusion model that is mostly based on pixels. Simply put, the diffusion model works by starting with a clear image that gradually becomes fuzzy (pixelated) over time. The model learns the process of this transformation, understanding the path the image takes as it changes from clear to fuzzy. Once the model is trained, it can generate its own images by doing the previous process in reverse. The more images you provide the model, the more it understands them and the wider the range of possible outcomes.

The Various Software Tools for Architecture and Design

The most known AI software tool is Midjourney but there are also other known tools such as Adobe Firefly, Dall-E, LookX, Stable Diffusion, and Runway. These tools are constantly evolving and bringing out newer versions to keep up with each other. Throughout the years, the outputs became more realistic and the tools became easier to control, which made it possible for architects and designers to play with it and learn along the way.

Do’s and Don’ts in Prompt Writing

The length of your prompt

The less words you put, the heavier the weight of those words. The key is to write a prompt that is clear and concise—neither too long nor too short—when you want to achieve a specific output. For more creative and holistic outputs, it’s better to use less text which makes the words heavier.

Sentence structure

In writing for AI tools such as Midjourney, it’s important to understand that the approach differs from writing for ChatGPT. These AI software tools prefer a mix of single words and adjectives over grammatical correct sentences. Also, the chronology of the words is less importance than with ChatGPT, it doesn’t have impact on your output.

Be clear

When writing your prompt, be sure to use precise and clear words. Don’t be vague. The software really needs to understand in which direction its output should go. Also, try to avoid using words with double meanings and instead include an adjective to provide more clarity.

Punctuation

As said above, there’s no need to write grammatically correct sentences. Random punctuation has no effect on your output. What you should avoid is leaving out the spaces between the words, because it can affect your output.

The use of parameters

In search for something really specific? Make use of the parameters. A good example that Tim showcased was the style raw parameter ( --style raw). Putting this at the end of your prompt, will make sure the AI tool outputs a raw version of your idea. Adding this kind of parameter overrides the build-in stylisation of the software.

Another example of these parameters Tim explained, is stylize ( --s or --stylize) which overrides the software’s default aesthetic style. In comparison to style raw, stylize works with a range from 0-1000. Adding this at the end of your prompt, behind your parameter can affect the style of the output. The default setting of stylize is at 100. By adding --s 0 your prompt will come out as very basic, whereas putting it at --s 500 the output becomes more unrealistic, phantasy-like. That’s why it’s crucial to find the perfect balance as to not compromise your own design style.

Balancing the weight of your words

When you want one word or adjective to stand out in the output, use hard breaks (::). This will divide your prompt into two equal weighted parts.

Tim explained this with an example of a prompt without and with a hard break. The first prompt without the hard break before the words ‘Zaha Hadid’ doesn’t give much attention to the word and sees it as just a small addition. But by putting the hard break just before ‘Zaha Hadid’, the tool sees its weight equal as all the words before the hard break, which results in an output that’s Zaha Hadid-like.

The guide above is just some highlights of Tim Fu’s Talk at Clerkenwell Design Week. Watch the full Talk below.