Ceci est une ancienne révision du document !
In part five of An Introduction to Stable Diffusion, we will learn to install Automatic1111 and be introduced to a different interface. We will also use this new interface to generate and modify images.
The current version of Easy Diffusion is 2.5.39 beta version and includes Seamless tiling as the added feature. I started Easy Diffusion from the command line and the below information appeared while loading. It indicates Easy Diffusion would automatically open in a browser window. On my Ubuntu Studio installation, that did not happen, possibly because I am running Brave as my default browser.
Having had some issues with Easy Diffusion, I installed the Automatic1111 version which is apparently more popular. As the name implies, Easy Diffusion is indeed easier to use for a new user and the interface more obvious. Automatic1111 is more capable if less user friendly. I will therefore start using both in comparison and when possible present both interface screens for comparison purposes.
First, we must get the Automatic1111 version installed. Go to the install page: https://github.com/AUTOMATIC1111/stable-diffusion-webui and go to mid-page where installation information can be found. There you will find procedures for Windows, Apple and Linux operating systems. It seemed more difficult to install on Linux in the past but now just install the Python dependencies for Debian, Red Hat or Arch based OSes then run the provided command in the directory desired and it will be installed. Run Automatic1111 by typing in:
webui.sh
in your terminal and opening a browser tab and going to the indicated URL:
It should then bring up a considerably different Stable Diffusion interface:
You will notice a series of tabs near the top with txt2img highlighted and img2img to the right. We will use these options to first make an image and then modify it. With txt2img selected create an image by typing in an appropriate description. In the following example, I used: penguins dancing in Antarctica, sun going down, rough seas, photo-realistic. You can use the default 512×512 size and Sampling steps of 20. I also used a batch count of 20 to create a variety of images and -1 as the seed produces a random number each time as shown by placing the cursor over the option (shown above).
Also indicated at the bottom of the page is the current version of Automatic1111 (v1.2.1) and versions of dependencies. Use the single graphic generated or pick out one of the batch generated. If you have batch produced several then one of the graphics is a grid of all generated as shown, twenty in my case. You can scroll through the batch and find one of interest (bottom left).
When the large version of the desired graphic is shown instead of the grid of all, select the Send to img2img button (shown above).
The graphic should then be seen in the img2img interface. You can then make the graphic wider by moving the slider or typing in the desired width, doubling it to 1024. Batch count should be changed to one and you can press the Generate button.
Look carefully at the newly generated graphic. It is not the same and seems very distorted. Making it more accurately reflect the original is done by maximizing the CFG scale to 30 and reducing the Denoising strength to zero in the img2img interface (top right).
After Generating the modified version the result is a very similar graphic but everything seems to have been stretched. There are ways to modify the generated graphic but it may still seem a bit of an art to get the desired result. One of the new updates is for image tiling which may also be utilized to make a wider graphic. We will address some of the numerous options as we go forward (bottom right).