Freelancing Spotlight: Using AI Art for Architecture & Interior Design

I’ve recently been doing freelance work for a real estate developer who wanted to learn how to use Stable Diffusion and ControlNet. I assisted them with installing the tools and learning the basics, as I have done for other clients. But throughout this project, we both discovered exciting new use-cases for AI art in architecture and interior design, which I will be showcasing in this post.

ControlNet for Interior Design Sketches

The first use-case we found is the most straight-forward: using ControlNet to finalize interior design sketches.

We start by finding an interior design sketch online, and then cropping it into a square shape:

Then we put this sketch through Stable Diffusion using ControlNet’s “lineart” model:

Midjourney for Architecture Ideas

Midjourney can be used for generating high-quality images of buildings, which can then be passed along to architects to design. Although this use-case could also work with Stable Diffusion, Midjourney generally provides higher-quality results.

Here are some tips for getting the most out of this technique:

  1. Use the names of famous architects in the prompts, such as “designed by Zaha Hadid” or “inspired by Frank Gehry”.
  2. Use the /remix command on generations you like to get better control over variations.
  3. Use the /blend command on both uploaded images and generated images to combine multiple visual styles.

ControlNet for Architecture “Massing”

Massing, according to David Balber, is “how the building looks in terms of basic masses, or chunks”. Below is an example of massing found online:

Take #8 from the above image and crop it into a square shape, so Stable Diffusion will have an easier time with it:

Then use ControlNet’s “lineart” model once again and experiment with several different settings and prompts:

We can also take one of the generations that we like and convert it back into lineart like the original massing image. Here is a cherry-picked example:

Finally we take the above image, put it into ControlNet’s lineart pre-processor, and then invert the colors of the preview image:

Conclusion

Real estate developers, architects, interior designers, and others in similar roles can benefit from using AI art like my client currently is. And there are still countless other applications for AI art that haven’t been realized by other industries yet.

Stable Diffusion and ControlNet have a bit of a learning curve and may require assistance to get set up, but other tools like Midjourney are designed for convenience and ease of use (although Midjourney isn’t free and lacks a few advanced features).

Leave a Reply