Embracing AI in Architectural Renderings
In the first part of our series on AI in architecture, we talked about the advantages of integrating AI into our processes. We now turn our attention to architectural renderings and how we harness the power of AI to drive the creative visualization process.
One of the most significant use cases of AI is the ability to generate early conceptual imagery. This allows us to create renderings that evoke the spirit and emotion of a project long before detailed designs are finalized or even put to paper. There are several AI tools that can accomplish this. We’ve been using Midjourney, a powerful AI available through the online platform “Discord”. Midjourney allows us to create and collaborate through its’ dynamic digital community.
When tasked with revitalizing a 40-year-old suburban strip mall, we used Midjourney to envision options. By uploading photographs and using highly descriptive prompts, we produced realistic images that illustrated pergolas, benches, and new landscaping. The images provided visual references that helped our client understand the project’s potential. This is especially crucial for projects where clients may struggle to grasp the value of design improvements and are unable or unwilling to expend a great deal of resources to find out what is possible.
AI tools like Midjourney work best with detailed written prompts. The more information we provide, the better AI can predict and generate images that align with our vision. There is both an art and science to “prompt engineering” so the AI can come up with the best results. In this way, architects are well suited to use it effectively. Once supplied with an excellent prompt, the AI starts an iterative process, creating multiple variations. This enables us to interact with the AI to refine concepts. The first result is never the best result. A simple prompt to remove a fountain and add a pergola can generate numerous iterations, each capturing different possibilities. The ability to produce these images quickly and with little to no ‘manual’ CAD or BIM modeling helps us cut to the chase for our clients. To help us engineer the best prompts, we start by conducting what we call a “visual preference survey” with the client, allowing us to gather feedback and refine designs based on their preferences. This approach saves time and ensures the final design resonates with the client's vision.
After initial concept generation, AI tools are also valuable in projects where we’ve already done some initial modeling. SketchUp, a long-standing favorite in architectural design, and part of our toolkit for over 25 years, now boasts AI plugins like Diffusion, which enable rapid rendering production. The Diffusion plugin integrates AI directly into the program, allowing us to create various styles of renderings with minimal effort. We can produce high-quality images in seconds by selecting a view in the model and using detailed prompts.
We used this concept for one of our current projects involving a not-for-profit youth soccer club that wants to build a brand-new campus. The project includes three to four soccer fields, an indoor field, and a clubhouse with concessions, offices, film-watching areas, and a fitness center. We started by modeling the basic volumes—modern glass boxes, flat roof planes, and grandstand seating overlooking the fields. Then we used the Diffusion plug-in to generate renderings of the model. Our prompt described a youth soccer facility with green grass fields, modern minimalist textures, grandstands, and kids playing soccer on a sunny Saturday, with lots of additional details. We selected the exterior photo style to align with our photorealistic rendering goal. The Diffusion plugin features two slider bars: one for respecting model geometry and one for respecting prompt details. We can adjust these sliders to stick closely to the model or take creative liberties with the prompt. For this project, we bumped up the model and kept prompt creativity in the middle. Each generated rendering captured the essence of our model, with variations in sky brightness and other subtle details. This ability to produce high-quality images is a significant advancement from our initial experiences with AI last summer, where the images often contained glaring inaccuracies, like soccer fields with inaccurate goal placements and striping.
We experimented further by shifting the rendering style. For a planning and zoning board meeting, we chose a watercolor style to present a less definitive but equally compelling vision. In this way, we are only committing to the vision, not the specific architectural details. By adjusting the model geometry slider, we achieved a sketchier feel, ideal for discussions where budget and material compatibility are still under consideration.
Another capability of Diffusion is creating aerial master plan views. Switching to a drone view in the model, we changed our prompt to a photorealistic rendering of a youth soccer facility at sunset, increasing the prompt influence while loosening the model geometry. The results were dynamic, with AI accurately capturing the sunset and replicating the soccer fields' geometry.
A word of caution and a note on limitations, however! While these images are excellent for capturing the project's spirit and gaining early consensus from stakeholders, they are not yet suitable for detailed discussions with planning and zoning commissions or architectural review boards. They lack the precision required by most jurisdictions for final approvals, such as specific materials, parking counts, etc. Nevertheless, the early visualization opportunities they provide are invaluable for gaining initial buy-in.
Similarly, Revit has AI integration through plugins like Veras from EvolveLAB. Veras utilizes prompts to enhance basic 3D perspectives with detailed, photorealistic finishes. For a tenant improvement project for a bank, we started with a simple hidden line 3D view and used Veras to transform it into a rendering of a modern, inviting office space. By setting parameters such as material overrides and prompt specifics—like specifying Herman Miller Aeron chairs, gray woven carpets, and green plants—we generated renderings that felt warm and welcoming. The plugin also allows for consistent rendering across different views by saving the “seed” for each rendering, ensuring visual continuity.
The continuous evolution of AI tools promises even greater advancements in architectural rendering. Platforms like AutoDesk Forma, with their automated site analysis and layout capabilities, are set to further streamline the design process. As these tools become more sophisticated, we can expect to see even more accurate and inspiring visualizations that bridge the gap between concept and reality.
The integration of AI in architectural rendering represents a significant leap forward for our industry. By harnessing tools like Midjourney, Diffusion, Veras, and Forma, we can push the boundaries of creative visualization, delivering compelling images that bring design to life. We’re determined to remain on the cutting edge of this fast-moving technology. There’s so much more to come. And we’re here for it!
This article is part of a series exploring the intersection of architecture and artificial intelligence, aiming to demystify the technology and inspire innovative applications within our industry.
