Original link: https://www.dongwm.com/post/stable-diffusion-embedding/
foreword
Today I will introduce Textual Inversion. The Chinese name is text inversion. In the previous version, this function is called Embedding, which is text embedding. In layman’s terms, it is actually packaging the prompt words into a prompt word. If you want to achieve a specific character, style of painting or action stably, you usually input a lot of prompt words to define the characteristics, which is very difficult and requires experience. Textual Inversion is to solve this problem.
Use Textual Inversion published by others
You can find a lot of packaged Textual Inversions at Station C, open the filter window on the right side of it, and click Textual Inversion in Model types , so that all the packaged ones are displayed on the page:
For example, if you want to quickly create a picture of the heroine Addams in “Wednesday”, you only need to download the Embedding package https://civitai.com/models/4115/jenna-ortega-wednesday-addams-embedding (only 30.92kb), and then stored in the local embeddings directory:
mkdir ~/workspace/stable-diffusion-webui/embeddings # This directory may not exist, you need to create it
mv ~/Downloads/W3DDDN3SD4Y.pt ~/workspace/stable-diffusion-webui/embeddings
|
Then, click the “show/hide extra networks” button under the “Generate” button on the right side of the web page to display some additional options, then click “Refresh”, and you can see the new Embedding in the first Tab “Textual Inversion” package, and then you can directly use W3DDDN3SD4Y as a prompt word instead of entering all those, which is very convenient.
Custom Textual Inversion
If you do too many pictures, you will habitually use some prompt words, because they will make the quality of the picture better, such as photography, masterpiece, best quality, 8K, HDR etc., then this is just to use the Textual Inversion of this article to create some For my own use, like me, I created a positive named dongwm, and a negative named dongwm-nt.
How to create it, click on the “Train” tab on the page, the first item in the menu bar below is “Create embedding”, and then enter the name of Embedding and the relevant prompt text. Below is the content of my positive prompt word text dongwm :
Click the “Create embedding” button and it will be saved. If you want to modify and replace it in the future, you can re-prepare the prompt word, then check “Overwrite Old Embedding” and save it.
Now you can see three Embedding packages in Tab “Textual Inversion”:
This article is transferred from: https://www.dongwm.com/post/stable-diffusion-embedding/
This site is only for collection, and the copyright belongs to the original author.