AI Garden Design?

​By Felicia Fredd

AI generated pincushion landscape

A few weeks ago, Google invited me to try out Gemini. I have no idea how this AI program differs from any others out there, but it’s free, and I wanted to see whether it could help me visualize some design ideas that have been rolling around in my head for a while.

Without reading any instructional preliminaries, I began asking Gemini to create and modify images. I ended up with dozens of iterations of a rusted metal wall with silver raindrops, which eventually morphed into a canyon slot garden with golden bubbles (images below). I gave Gemini the elements, the context (desert Southwest), and a few additional descriptive parameters—and it gave me configurations of those things.

The only frustrating part was that when I asked for just one specific, isolated change, Gemini would completely rearrange the image. I have no idea why that is, but I had very little control in terms of fine-tuning anything. On the other hand, I realized that this very lack of control was one of the program’s strengths as a creative tool. Not only did it keep me working, but many of the chance results were really intriguing. I was down a rabbit hole in no time, playing out one idea after another—each rendered within seconds.

I also found that I could ask Gemini to apply moods and styles—which will teach you something about moods and styles. If, for example, you apply a Baroque style to an ikebana arrangement of yucca leaves, grass, and a banana, the banana will be half-peeled. Apply a Gothic style to the same arrangement, and the banana will be half-rotted. Ikebanana? Gemini was able to apply cultural visual language to novel ‘problems’, and I thought that was pretty amazing.

Sadly, Gemini’s plant designs for landscapes were really boring, as were most color compositions using plants. I think these inputs are just too nuanced and complex in terms of layered juxtapositions. The plants it generated were pretty stiff, isolated, and cartoonish - okay for broad strokes maybe, but it didn’t even seem able to identify and depict plants by scientific name, at all.

​As to the question of using AI to design a real garden, I think the answer is: sort of, maybe, to some extent. It’s very good for exploring ‘big picture’ spatial layouts and concepts, but from my very limited experience, it doesn’t yet have the power to anticipate the many unique real-life conditions that inform functional design. In the end, you’ll need as much background knowledge using AI to design as you would without it. It doesn’t tell you the questions you’ll need to ask. Maybe it will in the future. Goodness.