imagination on demand: my 1st experiments in AI image generation

I was just granted access to Midjourney’s beta invite and ran a handful of quick tests to see how AI-based image generation will affect architectural lighting and experiential design.

My first impression: WOW! The raw power for ideation and early stage “mood board” imagery is incredible. It will change the way all creative designers work.

The beta test runs as a text-based service on Discord. The process is as such: Enter a text string of descriptors, and soon the service generates 4 initial options. You can then pick any of the 4 to generate new secondary options, or any of the 4 to “upscale” in refinement, or simply restart with a new “seed” to generate 4 new starting images altogether. If you’re curious, they include a guide for writing the prompts here.

Below you will find my experiments – a few typical interiors, some buzzy stuff like Gehry and Zaha Hadid style projects, some generic “designy” looking pastel drawings/moody renderings, and finally a fun, themed playground.

I titled each section with the exact text descriptor I used, plus included all the process imagery from start to end. Enjoy!

“contemporary office, workstations, wooden linear pendant lighting, open ceiling, open plan”

“frank gehry designed building, luminous walls, city location, night scene”

“biophilic design, parametric, ceiling, lighting”

The original generation of 4 ideas:

Requesting further variations of all 4 original ideas:

Choosing to Upscale one of the original 4 images:

“zaha hadid, video wall, night”

“lines of light, geometric, parametric, generative, fog”

“gothic church, dramatic warm lighting, pastel on black paper”

“gothic cathedral, dramatic facade lighting, pastel on black paper”

“rijksmuseum, dramatic facade lighting, pastel on black paper”

“wonka, factory, childrens playground, night, colorful, dramatic lighting, lanterns, sparkling lighting, color washes”

“playground, colorful, night, lanterns”

“playground, night, steampunk, colorful”