03/22/2026
This is an early version of the house we finished in PKS a year or so ago. In this version, the right wing was elevated, but not enough to create the understory that we envisioned at the outset. I would have liked to have explored this version a bit more before it hit an early budget crunch, but it naturally evolved into something cost effective and practical. Here we are in 2026 with all of these new tools - and I am very eager to advance my understanding of how to most effectively communicate with AI models to produce desired outcomes. So, I thought that the early study of the Ocain house presented a unique opportunity to study input/output synergy. At the beginning, I had envisioned a rather austere, monochrome structure. I learned through the process that the owner had notably different ideas about color and materiality. But for this study, I stuck to my roots and decided to treat this as a wood-formed, site-cast concrete structure. The original program did not call for a pool, but for this exercise I thought it made sense. Also important to this vis study are elements of gardening. The actual built house includes a lot of natural pines, oak and yaupon - blended with the owner's preference for turf with strategic ornamental camelias, magnolia and a single featured Japanese red maple. But for this exercise, I took a different garden approach... an organic blend of Southern Coastal maritime forest with Asian and Polynesian influences. One this note, I thought the AI was fairly responsive. Due to the inherent 1-off nature of iterative AI image creation, it is difficult to retrieve similar result from one scene setup to another. So, of critical importance for this study was to figure out how to use "character" images as interchangeable reference content between image files - literally asking the AI to grab content from one file and apply it to another file with different camera parameters. It was a little tedious... but it worked well enough to establish general continuity between different images. The ultimate goal of this study as a proof of concept was to push the limit of AI capability to achieve meaningful results using the most basic starter model possible. It's not perfect and the limitations are probably easily noticeable even from the most casual observer. But on the personal learning curve, I'd have to rate this as a successful endeavor.