In this post, I will briefly walk through some of the auxiliary functions provided in stable diffusion web ui for earlier idea testing and experience enhancement.

Script

At the beginning of generation, it is often the case that you need to test different prompts and parameter settings. Besides manually trying all combinations, webui has built-in scripts that automates this process.
This feature could be enabled by selecting the corresponding scripts. The script could be selected at the bottom-left corner of the webui.

Prompt form file or textbox

Used for testing different prompts using files or handwritten prompts.
It's pretty easy and straightforward to use so I would leave it for you to play with.

prompt matrix

Overview

Used for testing different tags' (in prompt) influence on the generation result.
This feature could be further enhanced by the extension dynamic-prompts
Below is a screen shot of the available settings.
For those interested in reading the source code, check here

Usage

To properly invoke this plugin, you need to append the tags you want to experiment with | symbol, follow the format below:

```txt basic_prompt, | tag1 | tag2 | ...

- Put a `|` symbol before each tag you want to try solely, and separate other tags using `|`.
- I used `tag` here for simplicity, but you could also try prompt snippets.
- In this way, prompt matrix would try the below combinations
	- ```
	  basic_prompt, 
	  basic_prompt, tag1
	  basic_prompt, tag2
	  basic_prompt, tag1, tag2

Example

You could find the tags used is displayed in bold, while the tags not used are ~~underscored~~
One common thing you want to experiment with is probably the style of the output.
- Here is an example, I tried different medium of generating a girl's portrait with the following prompt
  - 1girl, masterpiece, | pencil sketch | oil painting | photo | watercolor
Another thing you may want to try is that whether certain tags are taking effects.
- Here is another example, I tried different quality tags
  - 1girl, | masterpiece | best quality | extremely detailed
Or you want to try different lighting options
- 1girl, RAW photo, masterpiece, best quality, extremely detailed, | spotlight | rim lighting | sunlight | natural lighting
I would strongly recommend this website - promptmania if you are struggling with finding suitable tags for a specific aspect of the generation (e.g. medium, lighting, camera setting, etc.). This website providers wonderful visual cue for different tags.

XYZ plot

Overview

Used for testing different parameter settings
Below is a screen shot of the available settings.
For those interested in reading the source code, check here

Usage

The interaction logic is straightforward, select the parameter you would like to play with, fill in the values by hand or use the 📒 icon to automatically import all available options (for checkpoints, sampler, etc.)
In my previous blog, I used this script for doing experiments with testing different checkpoints' performance and parameter setting's impact.
Example usage
- Test different models performance on a given prompt
- Test different parameter settings' impact (like sampling steps, CFG scale)
  - X type = CFG scale, X values = 3,5,7,9
  - X type = Steps, X values = 10,20,30,40
- Test different parameter combinations.
  - By the way, it is often best practice to use recommended sampling method.
  - X type = Checkpoint name, X values = filled
  - Y type = sampler, Y values = filled

Auxiliary tabs

Skip the training related part for a later post (like checkpoinst merger, training, etc.)

Extras

This step is dedicated for upscaling images. It plays a similar role as txt2img tab's hires fix, or img2img tab's SD upscale script
Upload a image and choose a upscaler for the task.
- RealESRGAN_x4plus_anime_6B is good for animation characters
- realesr-general-wdn-x4v3 is good for realistic characters or scenery.
When an upscaler is used for the first time, it may take some time to download.
- If failed to download for some reason, check this website to download upscaler manually (and place them in stable-diffusion-webui/models/ESRGAN/ - not sure about all other upscaler, but worked for the above two)

PNG info

A useful tab for extracting generation data stored in images.
Enable setting to store the generation data
- In the Settings tab, check the following path
- saving images/grids > Save text information about generation parameters as chunks to png files
- You would then be able to retrieve generation data.

Extensions

Used to install and manage extensions.
Check the official extension list
Could also be downloaded locally and manually add to path/to/your/webui/root/dir/extensions folder
Recommended extensions
- Prompting
  - tagcomplete
    - when writing prompt, give you tag recommendations
  - tagger
    - Provided an image, use extra models to deduct the possible
  - dynamic-prompts
    - Mentioned in previous section, for quickly create different prompts
- Better Control
  - ⭐️Controlnet
    - One of the most helpful and important extension for creating aligned images.
    - Next blog post would be on this topic, discuss the detail usage of each one and share my workflow.
  - lora-block-weight
    - Used to control lora's performance
  - openpose-editor
    - Used as an auxiliary for character generation with fixed pose.
  - depthmap
    - used as an auxiliary for character generation with depth information (usually used for hand recreation)
- UI enhancement
  - Aspect ratio
    - Add buttons for common aspect ratios.

Tutorial

#Artificial Intelligence #Diffusion

SD - 3. How to use diffusion web ui? - Scripts, Extras, Extensions

https://delusion4013.github.io/2023/05/21/How-to-use-diffusion-web-ui-Scripts-Extras-Extensions/

Author

Chenkai

Posted on

May 21, 2023

Licensed under

Recommender Series 1 - Overview Previous

SD - 2.How to use diffusion web ui? - img2img Next