Ceci est une ancienne révision du document !
In part eight of An Introduction to Stable Diffusion, we will look at the new release of Stable Diffusion 1.0 (SDXL). The article starts and concludes with two images generated with the new version of Stable Diffusion via Automatic1111. Generating 20 random seeded images, 1024 x 1024 px, with the three-word prompt of Full Circle Magazine, the images highlight the creative possibilities of this software. Text remains an issue with the free version although text was not specified as part of the prompt.
I upgraded to a more capable graphic card which also required an additional upgrade for the power supply. Thinking that it might be a good idea to start with a fresh install of the OS (Kubuntu) - that is when the fun started. I almost always use Ventoy (https://ventoy.net/en/index.html) to install any of a variety of distros to machines to either try or install Linux. Inevitably, I now saw a “Bluetooth and malformed MSFT vendor problem” error. The OS would not start.
Upon searching the Internet, I found it to be a very common problem. There were some quirky workarounds, but obviously they were not the answer. Even when using the workaround, I would find that it would never come out of sleep mode, just a blank screen that would require a reboot.
Noticing an error referring to not having the appropriate Nvidia driver, I noticed the graphics driver was an Open Source version. After installing the suggested proprietary driver and rebooting, everything seemed to be working fine. I did experience a few glitches, but it was nevertheless much more stable. An Internet search found negative comments about Nvidia drivers and suggested use of AMD cards although after an update which included Nvidia drivers, it seems even more stable.
I was expecting the Easy Diffusion (ED) install, now at V 3.0.6, to be the one-click experience as before (https://github.com/easydiffusion/easydiffusion#installation). However, it required CURL to be installed on my now Kubuntu OS. Otherwise, it installed everything as needed. The safe sensors models are considered safer and SD-V1-5 gets installed in the models subdirectory under stable-diffusion. Although version 3.X will run the SDXL models, it needs to be downloaded and added to that same subdirectory. In my mind, the new Interface is a bit spartan. Users need to know to select the image settings button to choose appropriate options, and to select the +Image Modifiers button to select from those options.
The Automatic1111 interface version installed easily as before (https://github.com/AUTOMATIC1111/stable-diffusion-webui). The SDXL 1.0 models must also be separately downloaded, or copied from the ED install (https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features). You can check for updates by selecting Extensions at the right end of the tabs of the interface. It opens with the Installed tab selected showing a variety of options preselected. The Check for Updates button will do that, and you can update any option to the current version using the Apply and restart UI button.
I mentioned previously that Automatic1111 is a very popular interface for SD. One reason for that is that many additional options are available. These may be accessed via the same extensions tab, then selecting the Available tab option and the Load from: button. That will load the available options from the default Extension index URL. More than one hundred options will appear – which is a bit overwhelming given that an SD novice may not be familiar with the vocabulary. (Helpful for anyone is the Definitive Stable Diffusion Glossary: https://theally.notion.site/The-Definitive-Stable-Diffusion-Glossary-1d1e6d15059c41e6a6b4306b4ecd9df9.) A portion of the options are shown and I installed the Style Selector for SDXL 1.0. Selecting the Install button adds that option and changes the button to Installed as shown.
Installing the Style Selector for SDXL 1.0 extension option results in a choice of button prompt selections. It can be enabled in the interface by selecting the small triangular button at the far right of the SDXL Styles tab near the bottom of the Automatic1111 interface as shown. You can enable the style selector as shown. The styles can then be selected individually by selecting the checkbox for each. You can also generate images of all the options by selecting Generate all Styles In Order, which uses the prompt you have input.
The image below is another of the Full Circle Magazine images generated as noted at the beginning. In both cases they look like magazine covers albeit the text language would not be recognized. The ninth part of this introduction will look at other possibilities.