PROMPT GUIDE

AI art glossary (beginner-friendly explanations)

Once you start making AI art, you run into lots of unfamiliar words: prompt, seed, CFG, LoRA, VAE and more. This glossary sorts the most common terms into four groups and explains them simply for beginners. For any term you want to learn more about, follow the link to its guide.

Last updated: 2026-10-22 approx. 4 min read

日本語版

The short answer

AI art terms are easiest to remember in four groups: words that tell the AI what to draw (prompt, negative prompt), settings that shape the result (seed, CFG, sampler, steps), add-on data that changes the look (LoRA, VAE, embedding), and ways to fix an image (img2img, inpaint, upscaling).

How should you group AI art terms to remember them?

Sort them into words you write, generation settings, add-on data and fixing methods, and you will know where each term is used.

There are many terms, but each one belongs to a fairly fixed step. Prompts and negative prompts are written before you generate; seed and CFG are settings chosen when you generate; LoRA and similar add-on data change the look or character; img2img and inpaint fix an image afterwards. Find the group of the term you are curious about in the table below.

Group	Main terms	When it is used
Words you write	Prompt, negative prompt, tags, weights	Before generating
Generation settings	Seed, CFG, sampler, steps, resolution	When generating
Add-on data	Model (checkpoint), LoRA, VAE, embedding	To add a look or character
Fixing methods	img2img, inpaint, outpaint, upscaling	To fix a finished image

Anime girl checking terms in a notebook at a desk, image for an AI art glossary — 用語をノートで確認

Anime girl looking things up in a book in a library, learning AI art terms — 用語をノートで確認

Illustration usage rules

What do the terms for describing the image mean?

The prompt describes what you want drawn; the negative prompt lists what you do not want.

Term	Meaning	Detailed guide
Prompt	Written instructions describing the image you want	Prompt writing basics
Negative prompt	A field for things you do not want in the image	Negative prompt template
Tags (Danbooru tags)	Short English words such as 1girl or long hair	How to look up Danbooru tags
Quality tags	Words such as masterpiece or best quality that raise quality	Quality tags explained
Weights (emphasis)	Syntax such as (smile:1.2) that strengthens a word	Weight syntax guide
Natural language prompt	A prompt written as ordinary sentences instead of tags	Tags vs natural language

If you are starting out, read the basics of writing prompts: Prompt Writing Basics: Tag Order, Emphasis, and Tips →

What a negative prompt is, with a copy-paste template: Recommended Negative Prompts and a Copy-Paste Template →

What do the generation setting terms mean?

Even with the same prompt, the seed, CFG, sampler and step values change the image you get.

Term	Meaning	Detailed guide
Seed	The random number behind an image; the same number and settings give the same image	What is a seed value
CFG scale	How strictly the image follows the prompt	What is CFG scale
Sampler	The calculation method that gradually finishes the image	CFG, steps and samplers
Steps	How many times the finishing calculation repeats	CFG, steps and samplers
Resolution (size)	Width and height in pixels; the ratio also changes the composition	SDXL resolution list
Batch	How many images to make at once	-

How to fix a seed and reproduce the same image: What Is a Seed? Fixed, Random, and Reproducing the Same Image →

How to think about CFG, steps and samplers: What Is CFG Scale? How to Choose Steps and Samplers →

What do the add-on data terms mean?

The model is the foundation of the look, a LoRA is small add-on data for characters or styles, and the VAE affects the final colors.

Term	Meaning	Detailed guide
Model (checkpoint)	Large trained data that forms the base look	Comparing and choosing tools
LoRA	Small add-on data for characters, styles or outfits	How to use LoRA
VAE	The part that turns the result into an image; a mismatch makes colors dull	What is a VAE
Embedding (Textual Inversion)	Small add-on data packed into a single word	How to use embeddings
Trigger word	A keyword that activates a LoRA or similar data	How to use LoRA

How to write a LoRA and choose its strength: How to Use LoRA (Syntax, Strength and Trigger Words) →

The VAE to check when colors look dull: What Is a VAE? Why Colors Look Dull, and How to Choose and Set a VAE →

What do the terms for fixing images mean?

img2img redraws the whole image, inpaint redraws only part of it, and outpainting extends it beyond the edges.

Term	Meaning	Detailed guide
img2img	Redrawing a new image based on an existing one	Using img2img and inpaint
Inpaint	Redrawing only the masked area	Using img2img and inpaint
Outpaint	Extending the image by drawing outside its edges	How to outpaint
Upscaling (Hires.fix)	Enlarging an image while adding detail	Hires.fix settings
ControlNet	A system that guides the image with a pose, line art or other shape	How to use ControlNet

Steps to make one image while learning the terms

Write a short prompt for the image you want and put a standard template in the negative prompt
Choose a size and make a few images with a random seed
Fix the seed of the one you like and add to the prompt a little at a time
To change the look, try another model or LoRA
Fix broken spots with inpaint, or use img2img to change the whole image
Finish with upscaling

1girl, solo, long hair, smile, school uniform, classroom, upper body, masterpiece, best quality

Negative: worst quality, low quality, bad hands, extra fingers, blurry, watermark, text

Where do beginners get confused? (common misunderstandings)

Many people trip over terms with similar names, or terms that each service names differently.

Common misunderstanding	What is actually true	What to do
A fixed seed gives the same image whatever you write	Changing the prompt changes the image	Use it when changing things little by little
Higher CFG is always prettier	Too high makes colors and lines harsh	Start from a middle value
A stronger LoRA works better	Too strong breaks the image	Start low and raise it gradually
Longer negative prompts are better	Too long weakens important instructions	Keep to the standard words
Every service uses the same terms	Names and fields differ between services	Check the official documentation

Step-by-step for making your first AI illustration: Getting Started with AI Illustration (Beginner Steps) →

FAQ

Frequently Asked Questions

Does prompt mean the same as a spell?

Yes. In Japanese AI art communities a prompt is often called a spell (jumon), but it means the same thing as a prompt.

What is the difference between a model and a LoRA?

A model is the large base data that sets the look, while a LoRA is small data that adds a character or style on top. A LoRA is always used together with a model.

What is a seed?

It is the random number an image starts from. The same seed with the same settings and prompt gives the same image.

Why do services use different names for the same terms?

Each service names its own screens and features. The same idea can have a different name, so check the official documentation of the service you use.