Artificial intelligence in interior architecture

When ChatGPT came out, the interest in artificial intelligence (AI) quickly gained momentum. Since then, AI image generators like Midjourney, Adobe Firefly and DALL-E have also become available to the public. And now the text-to-video model Sora is heralding the next era.

At Dobas I’ve been exploring the impact that AI could have on creative processes in interior architecture: To what extent can text-to-image tools help us in designing interior spaces? And where are our limits? I’ve spent the past several weeks immersing myself in the world of Midjourney and I can now say: to use this tool is to learn a new skill, which requires time and patience.

From positioning concept to prompt

The first step of design processes for interiors is to define the positioning concept. Once this is done, we put together images to create a mood. With these images, I give the customer my first ideas for the interior. Typical mood sources include projects from architecture magazines, interior design websites, Pinterest or my own sketches. The risk of getting inundated by images is especially high on Pinterest.

Text-to-image tools present a new, huge and constantly expanding source of inspiration. In addition to creating my own images, I can do a keyword search for images that were created by other users in Midjourney. Every image it generates is public and can be seen and used by all users.

Thanks to the ability to create my own images, continue working on existing images with revised prompts, and generate images with my own sketches, it opens up a variety of exciting possibilities to achieve my objective more efficiently and present my design ideas in a short period of time. Specific spatial terms like «bar» or «restaurant» yield an image of a fictitious space; terms like «mood board» inspirational images unrelated to physical spaces.

Prompt design or prompt engineering

Prompts are commands that AI uses to create images. Depending on the order of the keywords and the description in the prompts, Midjourney weights the information differently. For me as a user that means test, test, test: generate an image, change the prompts and repeat. Creating targeted prompts is undoubtedly a skill that requires practice, but that also produces incredible results incredibly fast.

In Midjourney, a prompt yields four proposals. You then choose the most suitable option which can, with targeted instructions, be further developed into the desired result. During this iterative process, every AI-generated image remains in the feed, so you can always go back to previous results.

Unlike images from sources like Pinterest or iStock, which often have a similar «look and feel», AI-generated images can – with little effort and no technical image-editing skills – be tailored and adapted to meet project-specific requirements. And you always need to bear in mind that every image is public and other users can see all prompts.

More time for strategy and storytelling

AI proves itself especially valuable at the beginning of a design project. The somewhat lengthy search for mood images and illustrative examples can be noticeably shortened if you are proficient in prompt creation. In some cases, the programme even adds independent elements to the images in order to reinforce the atmosphere of the moodboards. The tool also allows you to choose different lighting situations, for example, «evening light» or «morning light», which improves the quality of the output.

The time saved can be dedicated to developing the positioning concept and brand-specific storytelling, which is quite welcome given the often limited budget allotted to the design part. My own experience has shown that projects often need to move as quickly as possible into the planning of time-consuming tasks such as building applications, and that sometimes less importance is given to the truly important foundational process.

The interface between AI’s abilities and humans

AI can undoubtedly enhance the creative process, save time and bring about unexpected solutions. But the impressive skills of Midjourney and other such programmes are only useful in combination with a human being’s real-life design expertise.

Materials are one area where the importance of an interior architect’s experience is clear. An AI-generated, visually appealing mood board alone is not enough if specific knowledge of materials is lacking, especially for projects in the commercial sector. What I envision as an interior architect must be something that can actually be implemented later in real life and, above all, be suitable for the specific use.

Similarly, a collection of tactile samples will never be replaced by an AI-generated material board, because the emotional reaction to a material cannot be achieved without feeling it. What’s more, material samples should be looked at in different lighting situations and in the actual project environment. I remember one time we had selected a material and, on the project site, it generated static electricity and every time somebody touched it they got a shock. And, of course, we immediately discarded that material.

My conclusion on AI in the design process

Artificial intelligence will be able to help us in many interior architecture disciplines in the future. However, I am convinced that hands-on experience, a feel for space, and storytelling creativity are irreplaceable in order to create sustainable, high-quality designs.