PROMPT GUIDE
How to Use Embeddings (Textual Inversion)
Those name-like words you often see in example negative prompts, which are not ordinary English tags, are usually embeddings. This beginner guide covers what an embedding is, where to put it and how to write it, how to use negative embeddings, and how they differ from LoRA.
Last updated: approx. 2 min read
The short answer
An embedding (Textual Inversion) is a small file that packs the meaning of several words into one name. Put it in the designated folder and write the file name in the prompt or negative prompt. It does nothing if it does not match your model type (SD1.5 or SDXL), so check the distributor notes first.
What is an embedding (Textual Inversion)?
It is a small add-on file that teaches the model to read one name as a bundle of meanings.
An embedding is a file made with a method called Textual Inversion. It does not change the part of the model that draws; instead it adds a small dictionary entry that says "read this name as meaning this". That is why the files are tiny, and why many kinds are shared for different purposes, such as style, quality, and negative ones that avoid common flaws.
How do you install and write an embedding?
Put it in the designated folder, then write the file name (without the extension) in the prompt.
Steps to use an embedding
- Check in the distributor notes which model it is for (SD1.5, SDXL, and so on)
- Put the file in the embedding folder of the interface you use (often named embeddings)
- Reload the interface and check that it appears in the embedding list
- Write the file name (without the extension) in the prompt or negative prompt
- If it is too strong, lower the weight, as in (name:0.8)
The folder location and how the list is shown differ by interface (WebUI, ComfyUI and others) and version. ComfyUI may use a different syntax, so check the documentation of the interface you use.
How do you use a negative embedding?
Writing its name in the negative prompt avoids common broken shapes and low quality all at once.
A negative embedding bundles things you want to avoid, such as worst quality or bad hands, into one name. It is handy, but too many can change the whole mood of the image. Add them one at a time and compare before and after with the same seed.
Rinka's tips
- Shared embeddings may come with usage terms (such as commercial use). Read the distributor notes before using them
- Compare before and after with the same seed and settings to see whether it really has an effect
- To compare many settings at once, lay them out with an XYZ plot
For how to use LoRA and choose its strength: How to Use LoRA (Syntax, Strength and Trigger Words) →
For a basic negative prompt template: Recommended Negative Prompts and a Copy-Paste Template →
For comparing settings side by side with an XYZ plot: How to Use the XYZ Plot (Compare Settings Side by Side) →
Copy-and-paste templates
Replace the parts in [ ] with the embedding name you use (the file name without the extension).
1girl, solo, long hair, smile, upper body, masterpiece, best quality (negative: [negative embedding name], worst quality, low quality)1girl, solo, ([style embedding name]:0.8), long hair, school uniform, classroom, upper bodyFAQ
Frequently Asked Questions
What is an embedding?
It is a small add-on file made with a method called Textual Inversion that bundles the meaning of several words into one name. You use it just by writing the name in the prompt.
What is the difference between an embedding and a LoRA?
An embedding only adds what words mean, while a LoRA overrides part of the model. Embeddings are small and suit quality and negatives; LoRA suits reproducing characters and styles.
My embedding does not work.
First check that it matches your model type (SD1.5, SDXL, and so on). Then check the folder location and the spelling of the name.
How many negative embeddings should I use?
Start with one. Several can change the whole mood of the image, so add them while comparing with the same seed.
RINKA WORLD
Play in Rinka World
MORE
