Glossary

AI Image Generation Glossary

36 terms explained in plain language: prompt, seed, LoRA, ControlNet, inpainting, CFG scale and the rest. Each entry also says why it matters and where you will meet it.

A

Aspect Ratio

The ratio of an image's width to its height; it directly determines the composition.

C

CFG Scale

The setting controlling how closely the model adheres to the prompt.

Character Reference

A reference method that keeps the same character consistent across different scenes.

Checkpoint

The base file containing an image model's trained weights.

Commercial Use

Using a generated image in revenue-generating work; it depends on the tool's licence.

ControlNet

An extension binding generation to a reference image's pose, depth or edges.

Copyright

Legal protection granted to a work's creator; in most countries it does not apply to fully AI-generated works.

D

Deepfake

AI-generated content imitating a real person's likeness without permission.

Diffusion

How image models work: building an image by progressively removing random noise.

DPI

Dots per inch in print; the measure of print quality.

F

Fine-Tuning

Training an existing model on new data to specialise it for a particular domain.

H

Hallucination

A model generating details that do not exist or make no sense.

I

Img2Img

Generating a new image using an existing one as the starting point.

Inpainting

Regenerating only a selected region of an existing image.

L

Latent Space

The internal space where a model represents images in compressed mathematical form.

LoRA

A small add-on training file that teaches a model a new face, product or style.

M

Model Licence

The legal text defining how an AI model may be used.

N

Negative Prompt

A second instruction listing what you do NOT want in the image.

O

Outpainting

Extending an existing image's frame outward by generating its surroundings.

P

Photorealism

A generated image being realistic enough to be indistinguishable from a photograph.

Prompt

The text instruction that tells an AI what you want it to generate.

Prompt Engineering

The practice of writing and refining prompts to reliably obtain a desired result.

Prompt Weighting

A method for increasing or decreasing the influence of specific parts of a prompt.

R

Resolution

An image's size in pixels; screens and print need different thresholds.

S

Sampler

The algorithm determining the mathematical method used to remove noise.

Seamless Tile

A pattern image whose edges match perfectly so it can repeat infinitely.

Seed

The number setting a generation's starting randomness; the same seed with the same prompt reproduces the same image.

Steps

The setting for how many stages the denoising takes; it directly affects quality and time.

Style Reference

A reference method that carries one image's style into new generations.

T

Text-to-Image

Generating an image from a written description; the basic mode of every tool here.

Token

The smallest unit a model breaks a prompt into; it defines the prompt length limit.

Transparent Background

An image with an empty background letting whatever is behind it show through (usually PNG).

U

Upscale

Increasing an image's resolution by using AI to fill in the missing detail.

V

Vector Image

An image defined by mathematical shapes rather than pixels, staying sharp at any size.