Ceci est une ancienne révision du document !
This is the first of several articles on Stable Diffusion, a graphics program that uses artificial intelligence to create exciting and creative graphics. At its core from the viewpoint of a newbie, Stable Diffusion (SD) is a program which can create graphics of various types from text or prompts. Understanding how and why it works is unnecessary. It requires very little knowledge to get started, although the output, while almost always interesting, is somewhat uncontrollable with first attempts.
Stable Diffusion (SD) comes in various flavors and options, some are Internet based, which generally make a transition from free as in beer, to free as in speech. A browser-based user interface appears to be the norm. The free downloadable version peaked my interest initially, and is usable with the most computer hardware. Speed of graphic generation is dependent on hardware, ideally an NVIDIA card with VRAM. Thus limited capability is reflected in generation time and size (512 x 512 px by default.)
I started with the NMKD SD (https://nmkd.itch.io/t2i-gui) on a Windows desktop, and I wondered about using my Ubuntu Studio laptop. When I saw another very simple installation, I downloaded it to give it a try. My desktop was more capable than the laptop but didn’t have a supported graphic card either.
Installation is very easy. Go to the Stable Diffusion UI v2 website (https://stable-diffusion-ui.github.io/), select the download button and follow the simple instructions. After downloading the installer file, extract it, go to the created folder (the download folder by default) and open a terminal window in that directory and run:
./start.sh (or) bash start.sh
It will begin downloading everything needed, about 10.2 GB and opens Firefox or a tab if already open. By default the generic prompt opens “a photograph of an astronaut riding a horse” which will be created by clicking on the Make Image button as noted on the opening instructions. (To uninstall, simply deleting the stable-diffusion-ui folder) Closing the application, the browser tab and restarting will automatically upgrade the installation with the newest version – currently 2.5.22, thus a reason to not run it continuously.
Now that you have it working, what can you do with SD? It is a bit daunting at first but trying out a few options and prompts is easy enough. I will expand on the user interface in another article but you can simply type in “penguin” in the Enter Prompt area and click on Visual Style below the Image Modifiers (Art Styles, tags, etc) area, select Art Nouveau (or other option) by clicking on the PLUS sign when your cursor is over the Art Nouveau option. Then click on the Make Image button. The result is interesting if not impressive. Generated images can be automatically saved via the Settings tab at the top of the Easy Diffusion window.
You soon begin to wonder exactly how difficult it is to create graphics that are more impressive, especially for those who think they have limited artistic abilities. Open art (https://openart.ai/) is a great place to start, it allows you to “Search 10M+ of prompts, and generate AI art and AI images via Stable Diffusion” and other similar software like DALL·E 2. If you need inspiration, you will find it there. They also provide information in a downloadable PDF book: (https://openart.ai/promptbook). SD has already been used to create Plasma Desktop Wallpaper which you can find using Discover and searching for Stable Diffusion. Keep in mind that a prompt not written for SD may not give you the same results as shown. A warning - you might not want to start looking at this if you have limited time.
I used the prompt “penguin on surfboard holding sign” to create a graphic I modified with the text FCM at the top and Stable Diffusion at the bottom. The penguin is a graphic partially on the surfboard although I had wanted it riding the surfboard. There were three letters initially at the top which I deleted and used Inkscape to add the text shown. I will provide a deeper understanding and further information next time, including an analysis of the user interface.
I am providing another graphic generated SD with the same intent to show how it interpreted what I was thinking “penguin standing on skateboard with sign and ocean background” in an unexpected manner. Using SD is clearly a science and an art.