Application of Stable Diffusion AI in Architecture Design Part 3

Do you often wonder why it is so challenging to adjust the parameters such that the drawings produced are extremely inconsistent with your expectations? Well, that's probably because the parameters aren't adjusted properly!

In this section, I will also use my examples to demonstrate a personal understanding of the three most commonly used parameters: ControlNet & Steps, ControlNet hierarchy, and Denoising & Steps.

ControlNet & Steps

0-50: Low requirements for the graphics card. You can quickly understand the outline of your drawing through this threshold, but it is not recommended as the main image producing range.

50-100: Moderate requirements for the graphics card. It's appropriate to reduce the drawing speed to improve the quality of the drawing, a reliable number range that can meet most of the image production needs

150: High requirements for graphics card. Uses the slowest image producing speed to improve the drawing quality. Though at the same time of a high quality image, there may be some unwanted noise or objects that starts to appear. At this range, it can be used reasonably according to one's own needs.

ControlNet Weight

Around 0.5: a very magical value, the probability most likely will not change the shape of the fed graph, though each decimal point may have a different result. It's worth playing around to try out the result.

Around 0.7: A value that is perfect for brainstorming. This weight will provide a variety of possibilities based on the graph fed in.

Denoising & Steps

1: The image will basically be generated according to the composition given, reducing the possibility of diversifying to adapt to some fixed composition, suitable for people with a very clear idea.

0.8: A very comfortable threshold, can provide a variety of graphic experience on a certain clear composition, highly recommended.

0.5: Suitable for students with certain composition ideas but not clear, can provide various possibilities based on your composition. Suitable for directional brainstorming in the early stage of design.

This series of the fundaments of AI Diffusion tutorials has come to an end here. I hope you can try to use AI in daily design, have fun and learn to use. I look forward to seeing you all again in my future posts!

All Images are produced by Author Himself

Translator: J