Ceci est une ancienne révision du document !
In the last part of this series, I mentioned I would add an example image at the beginning and end of each article with prompts to show the ease of creating images. I normally create 20 images and choose what I deem the most appropriate. All the prompt information is not necessary unless you want to get a very similar image. Using Automatic1111, the prompt is the underlined text below with additional image settings: A penguin overlooking a ledge looking at a frozen Central Park and New York City, highly detailed, very intricate, cinematic lighting Steps: 20, Sampler: Euler a, CFG scale: 14, Seed: 1609724340, Size: 1024×1024, Model hash: 31e35c80fc, Model: sd_xl_base_1.0, Style Selector Enabled: True, Style Selector Randomize: False, Style Selector Style: base, Version: v1.5.1
In part nine of An Introduction to Stable Diffusion, we will look at the sd-webui-prompt-all-in-one (PAIO) extension on the Automatic1111 interface of Stable Diffusion 1.0. It is the significant number of ever increasing extensions that makes Automatic1111 the choice for many SD users. This extension is, as the name implies, an attempt to make all types of modifiers available through a click-based format.
To install PAOI or any other extensions, select the Extensions tab in the Automatic1111 interface as shown. The current state and installed extensions are shown. You might want to Check for updates and then Apply and restart UI as seen in the interface. You will notice that I already have the PAOI extension installed. If you don’t have it installed, select Available from the interface, and then select the Load from button to show the many options. To find it faster, search for prompt-all-in-one (CTRL + f) and install using the install button on the far right in the Action column.
Once installed, the button to the left of the Prompt (red arrow below) lets you Show or Hide the Panel, and the similar button to the right (red arrow) lets you show or hide the Group Tags. The numbers to the right of the word Prompt (yellow arrow) show the number of prompts used of the maximum 75. If neither are shown, use the buttons to show the Panel and then the Group Tags.
For example, if the panel is shown, you could type in ‘early teen’ in the Please Enter New Keyword box because it’s not an option specified, or any other parameters you want to make sure are included in the prompt. Then select Person, Wings and bat_wings, then female, Wings and black_wings to produce the prompt shown. You need not type in anything but you can create a prompt using both typed input and PAOI selections (bottom right).
There are many listed options; for example, under Person view Object selecting Eyes or Ears. Many are derived from other languages or are abbreviations which may be unfamiliar, but as you become more familiar it increases the possibilities. While the results are unimpressive with a limited prompt, making 20 images helps to suggest various options when the user is unsure of what direction to go to meet the desired need. You can always type in descriptors if you know what direction you want to pursue. Individual prompts can also be modified by moving the cursor over the prompt, ‘early teen’ in this case. Various options to delete and modify Keyword and Keyword weight are then available which is likely easier than modifying individual prompts if the list already has several words.
The final image below is produced using Easy Diffusion. It's a bit late for All Hollows’ Eve but maybe for next year. Starting with a prompt of = “detailed chocolate skull, chocolate candies, [tequila bottle with two glasses], in Mexican cemetery, at sunset, high resolution, burning candle” and “seed”: 2524754007, “use_stable_diffusion_model”: “sd_xl_base_1.0”, “clip_skip”: false, “use_controlnet_model”: null, “use_vae_model”: “”, “sampler_name”: “euler_a”, “width”: 1024,“height”: 1024, “num_inference_steps”: 25, “guidance_scale”: 7.5, “use_lora_model”: null, “use_embeddings_model”: null, “tiling”: null, “use_face_correction”: null, “use_upscale”: null”,
you should be able to get a somewhat similar image.