Application of Stable Diffusion AI in Architecture Design Part 2

Last article we introduced the basic concept of Ai Stable Diffusion. In this article, we will introduce several different control methods that's going to be suitable for people that have architectural models, but do not know how to render, or for those that have black and white line draft to come up with ideas quickly!

In short, drawings have two main parts to it: composition and style. People used to say that AI painting is like alchemy, but in fact, what exactly is thrown into the pot?

Stable Diffusion controls the composition and style of the picture through various "spells".

Everyone knows that these spells are annoying, it's a whole complicated and different language system, but in my opinion, it's one step further. These so-called easy to use "spells" to train AIs are actually a bit anti-human. AI's goal should have been to lower the threshold of learning and using, not raise it. With the rapid development of AI technology, these spells will eventually be eliminated and replaced in the future. Even now, there are two handy tools to solve this problem: ControlNet is used to control composition, and LoRA is used to control style.

Let's see how ControlNet controls composition!

ControlNet: MLSD linear lines detection

Control Net: MLSD model is a special detection of straight line model, based on the characteristics of MLSD model, MLSD model in the architecture, indoor direction of the processing is a good choice, can detect the original line.

ControlNet: Segmentation

Using ADE20k for semantic segmentation, defining different items or components with colors, semantic segmentation allows AI to correctly identify object types and boundaries for demand generation.

Using MLSD in defining boundaries and contours: architecture students' favorites!

Section: The boundary between the building and the outside, the shape of the wall section, the outline of the column

Exterior Perspective: outline of the boundary between the building and the outside, windows, doors, vegetation

Interior Perspective: Contouring windows, doors, lamps, vegetation profiles of indoor objects

Accurate control chart using MLSD and SEG

Use Seg to define content:

Section: Define walls, windows, doors, sky, people

Exterior Perspective: Define exterior walls, windows, doors, vegetation, people, sky

Interior Perspective: Define windows, walls, ceilings, people, greenery, doors

Thank you for your interest and look forward to more wonderful exploration of the integration of architecture and technology.

All Images are produced by Author Himself

Translator: J