Ceci est une ancienne révision du document !
In part three of this introduction to using Stable Diffusion, the emphasis is on guiding the user on the Stable Diffusion (Easy Diffusion) interface using the 2.5.35 (Beta) version, with a focus on writing effective prompts. The previously noted issues with GPU settings have apparently been resolved with the more recent updates.
While it’s easy to write a prompt for something you might imagine, getting it correct is obviously the big issue. You can think of it as an English sentence describing the image with more important words or phrases at the beginning and less important ones at the end. You can also emphasize text with parentheses and, to de-emphasize, use square brackets or add text to the negative prompt area. A general approach is to then create 10 or more images from which you can select the most applicable to then modify. You can do that in the Image settings area by selecting random for the seed. Then change the number of images to 10, or 4 in this case. Choosing a larger number of images allows those with less capable graphics to start making images and walk away.
Having a general idea is important because, with that as a framework, there are many options built into the interface. They are grouped into types of Image Modifiers which account for an additional 207 options. This currently includes 8 Drawing Styles, 44 Visual Styles, 6 Pen Styles, 7 Carving and Etching Styles, 22 Camera styles, 10 Color styles, 15 Emotion Styles, 74 Artist Styles, 8 CGI Software Styles, and 13 CGI Rendering Styles.
You can add any of these by expanding the Modifier styles if not open by selecting it with the mouse, Drawing Style for example. Then when hovering above a particular style thumbnail, it causes a Plus sign to appear. Once selected, that same thumbnail is added above the make image button as opposed to written in the Prompt area. Hovering above either location brings up the Negative sign which allows you to remove it. Thus a few clicks allow you to add styles faster than typing the same information into the Prompt area – which is helpful when you are first learning those options. That number of options might seem like a lot but wait, we are just getting started!
Over three hundred prompts were noted at Stable Diffusion V1 Modifier Studies, expanding the imaging creation possibilities, and were added before November 2022. While there is some overlap, most of these styles were not available prior to the update in the Image Modifiers area. To test this, I simply tried typing in a style, in this case Atompunk, and made 10 images. (Atompunk is a colorful view of the future from the perspective of the 1950s and 60s.) The below is an example of what was created. You can also simply type in the name of an artist such as Adi Granov, a Bosnian comic book artist, and get an image reflecting his style below. The ethical and legal question of incorporating artists' work style in the created database used to generate images is currently being tested in court.
Another option to consider depends on your purpose of the graphic that you wish to reflect. If your image is more landscape oriented, it’s helpful to change that option. Changing the type image can be done under Image Modifiers (art styles, tags, etc) by clicking on the Image Style selector down-arrow. Each of the modifier example thumbnails then changes accordingly. Next time we will look at modifying some of the images created.