Architecture
I love architecture. It’s what got me started drawing. I imagine because it’s all angles and perspective, it fit my brain when I started drawing. I’ve gotten better at being less rational, but I still love buildings. My photos from our family trip to Manhattan features about 10% my family and 60% the buildings we drove past. Some examples follow. Anything I do to release somewhere I’ll put in Photoshop and do editing, color correction, and all sorts of other bits of what I call my “Photoshop fuckery.” These are just the raw images from my gallery as I got them from the models.
Of the various models I use, Midjourney is probably the best at getting a building to be spatially believable.
Of course, if I wanted reality I could just take reference photos and draw them or treat them in Photoshop. I love the AI models because they let me cut loose with how I’d imagine architecture in other universes. As long as there’s a physical space and a conscious mind to live in it, there’s going to be some kind of theory of designed space.
The newer versions of Midjourney don’t necessarily handle the spatial parts any better but they do have a better feel for composition. I can often get pretty much exactly what I want out of a prompt I spend enough time on.
Flux (which I use in both Krea and Leonardo) is very literal. It’s really good for giving you a picture of a bird or a cup or a table. But when you scale that up to buildings and bigger spaces you see the lack of vision Flux shows. So to get something interesting out of Flux, I found I needed to add in guidance images, often of things that are not buildings. Krea styles seem to be trained much more on visual presentation than on composition, so you can trick them into some cool things.
Leonardo, on the other hand, goes the other way. It’s extremely good at doing “content reference” (meaning the structure of an image) and applying a visual style to it. it’s surprisingly clever at how it does this sometimes.
