---
title: 'Explaining Prompting Techniques In 12 Minutes – Stable Diffusion Tutorial (Automatic1111)'
source: 'https://youtube.com/watch?v=dlUpSEbbCho'
video_id: 'dlUpSEbbCho'
date: 2026-07-28
duration_sec: 726
---

# Explaining Prompting Techniques In 12 Minutes – Stable Diffusion Tutorial (Automatic1111)

> Source: [Explaining Prompting Techniques In 12 Minutes – Stable Diffusion Tutorial (Automatic1111)](https://youtube.com/watch?v=dlUpSEbbCho)

## Summary

This video explains various prompting techniques for Stable Diffusion using the Automatic1111 interface, covering how to structure prompts, use weighting, embeddings, prompt editing, and other tools to control image generation.

### Key Points

- **Prompt Ordering** [0:00] — Prompts are ordered from most important to least important from top to bottom and left to right.
- **Token Limits** [1:06] — Token limits refer to the maximum number of words in a chunk of 75 tokens; if you have 100 tokens, it processes 75 and then 25 independently.
- **Negative Prompt** [2:21] — The negative prompt tells Stable Diffusion what you don't want in the image, such as bad anatomy or artifacts.
- **Parenthesis for Weighting** [3:02] — Wrapping a word in parenthesis increases its attention by a factor of 1.1 per pair.
- **Square Brackets for Reducing Weight** [3:45] — Square brackets decrease the attention to a word by a factor of 1.1 per pair.
- **Prompt Weighting with Colon** [4:23] — Using a colon and a number inside parentheses allows precise control over word importance, e.g., (snowy weather:1.5).
- **Embeddings for LoRAs** [5:12] — Angled brackets are used for embeddings like LoRAs, with format <lora:filename:multiplier>.
- **Prompt Editing** [5:51] — Prompt editing swaps prompts during generation using [from:to:when] format, where when can be a decimal (percentage of steps) or whole number (exact step).
- **Escape Character** [7:22] — A backslash before a special character turns it into ordinary text, removing its effect.
- **BREAK Keyword** [7:56] — The BREAK keyword fills the current chunk with padding and starts a new chunk.
- **Alternation with Vertical Bar** [8:19] — Using vertical bars inside square brackets creates alternation, e.g., [long black hair|brown dreadlocks|orange curly hair].
- **CFG Scale** [8:55] — CFG scale determines how strongly the image conforms to the prompt; lower values give more creative results, recommended range 5-12.
- **Prompt Matrix** [9:36] — The Prompt Matrix tests the impact of individual prompts by generating a grid of images with different prompt combinations.
- **Prompts from File or Textbox** [10:30] — Multiple prompts can be tested by separating them with line breaks in a textbox or from a file.
- **XYZ Plot** [11:09] — XYZ Plot allows testing and comparing variables like seed, CFG scale, and prompt S/R (search and replace).

### Conclusion

Understanding these prompting techniques allows you to have more control over Stable Diffusion image generation, enabling you to fine-tune results and create images closer to your vision.

## Transcript

hey everyone I'm b-size genius prompting
instable diffusion can be a mystery and
it's sometimes tricky to know what does
what but there are techniques that you
can use to get the results you want and
in this video I'll be breaking them down
so you can spend less time reading and
more time creating but like the video
and give it to me bite-sized so let me
quickly cover some basic information
prompts are ordered from most important
to least important from the top to the
bottom from the left to the right there
were various theories on how you
structure your prompt for the best
result but it's worth keeping in mind
certain Concepts such as the subject
lighting photography style color scheme
doing words and much more which help to
build up your image when it comes to
style prompts can influence this
alongside your desired checkpoint as
stable diffusion was trained on a
multitude of data sets from across the
internet meaning you can draw references
to art style
celebrities clothing types and much more
to influence your image The Prompt
sections have these numbers on the top
right called token limits and these
refer to the maximum number of words you
could fit into a chunk of 75 tokens for
example if you had 100 tokens then it
would process 75 tokens and then 25
tokens independently into stable
diffusions unit this in plain English is
how the AI language model breaks down
and manipulates text for processing The
Prompt box is where the magic happens
this is where you will describe
manipulate and design your image through
the text to image section or alter
images if you're using image to image in
conjunction with your reference photo
you can put as much text as you like in
here but often it's better to keep
things short and sweet so your prompts
are easier to fix orine as you make
adjustments closer to your desired image
for example let's type out medium shot
of a woman with four lips golden eyes
and a white crop top long black hair
snowy weather octane render we can see
that stable diffusion will attempt to
generate its interpretation of that
prompt and this interpretation will
change depending on the seed image size
and other settings which change the type
of image generated from the same prompt
the negative prompt box is where you
tell stable diffusion what you don't
want in your image and this can include
literal Concepts items weather or
artifacts and bad Anatomy within an
image again you can put as much text as
you want in here but it's good to keep
it reasonable to make life easier for
yourself later down the line for example
let's type out bad dream unrealistic
dream nons saafe for work and we will
notice that we get a much higher quality
image that's safe for work it's worth
checking the example images of the
checkpoint you are using to ensure that
you are using the best NE negative
prompts for your chosen model the
parenthesis is used to put a greater
weight or importance on a word than your
prompt and for each parenthesis wrapping
a word it would increase its attention
by a factor of 1.1 and the further
parenthesis will multiply the attention
by 1.1 I'll be honest in saying that the
mass behind this isn't really all that
important as experimenting with your
prompts will yield better results but
it's still important to see this in
action so you can find two your images
so for example if I were to write out
our previous prompt but put snowy
weather in parenthesis then we can see
how the snow and even the clothes start
to turn either white or have a much
snowier presence the more parenthesis we
use square brackets are used to reduce
the weight or the importance of a word
in your prompt and for each square
bracket it will decrease the attention
to the word by 1.1 and of course we'll
multiply the detention by 1.1 for each
pair of square brackets WRA a word again
it's better to experiment when it comes
to fine-tuning your prompt but by
understanding what these functions do
you have another tool in your belt for
fine tuning images so for example if I
were to write out our previous prompt
but put snowy weather in square brackets
then we can see how the snow is reduced
especially around the hair and the
clothing the more square brackets we use
let's address prompt waiting you can
manipulate prompts to either add or
remove waiting or importance from words
within your prompts and what this means
in plain English is that you can control
how much impact certain words have over
others within your prompt and words with
greater impact will be visualized more
strongly in your image this is done by
wrapping your word in a parenthesis but
we will now add both a colon and then a
number which can be a whole number or
with decimal values so for example if I
were to write out snowy weather with a
variety of prompt ratings we can see how
the impact changes from no promp waiting
to a really high promp waiting value we
obviously get rather lowquality images
at very high waiting so keep the numbers
reasonable you may have seen these
angled brackets in certain prompts
perhaps some websites that provide
checkpoints and luras these are known as
embeddings and are common in luras where
a file and multiplier for the file needs
to be specified to determine the
strength of the Laura the typical format
would be Laura file name and then
multiplier and unfortunately you can't
do prompt editing with Aur in this
version of stable diffusion in this
example I've used the add details Laura
which adds or removes detail from our
generated images and we can see the
effects of changing the values within
our embeddings and how it is structured
now prompt editing is a powerful way of
controlling your generated images by
swapping the prompts being used for
generating an image during the
generation a good way to illustrate the
format is by using from to and when
where from determines what prompt you
start with two determines which prompt
you end with and the Step at which the
switch takes place is determined by when
you can also do from and when with two
colons in between to remove the prompt
being specified after a fixed number of
steps determined by when and don't worry
I'll show you this in practice so you
don't feel like you're in another Mass
class but remember this key piece of
information the weightings you provide
in decimal numbers are percentages of
your sampling steps which must total up
to one meaning 0.5 is actually 50% of
your total sampling steps anything l a
whole number above one will be the exact
sampling step you want to specify
meaning a figure of 20 means stop or
switch at sampling step 20 so for
example you can see the impact in these
images with our first image using
decimals which represent percentages to
transition from snow to Sun after a
certain number of steps the second image
uses whole numbers to represent the step
for the transition based on the 30 steps
we are using and lastly we can indicate
at which step we should stop using the
selected prompt s weather Al together
now here's a call trick using a
backslash before a special character
such as a bracket or par parenthesis
will turn that special character into
ordinary text so using it in practice
we'll remove the effect of the
parenthesis giving you snowy weather as
pure text with no effect you can see an
action here where we have the standard
prompt snowy weather compared to snowy
weather within the parenthesis and then
the literal parenthesis represented as
text alongside our prompt and you know
it worked because the snow is less
prominent now thinking back to the
tokens which form chunks I described in
the the beginning the break keyword in
uppercase will fill the current chunks
with padding characters and adding more
text after break will start a new chunk
I personally do not see any practical
use for breaking up your chunks
prematurely before hitting that 75 token
limit but if you wanted to achieve this
then this is your solution now the
horizontal line is used to trigger
alternation or a loop in prompts where
words are broken up with horizontal
lines and are given the chance to
influence the generation repeatedly as
stable diffusion Loops through the words
within the square brackets so if we were
to use long black hair brown dreadlocks
orange curly hair as a prompt then the
first step will be long black hair then
Brown dreadlocks then orange curly hair
this is another technique for
controlling the types of generations you
get during the generation process and
you can see in action how this impacts
the final result the CFG scale will
determine how strongly the generated
image should conform to the prompt you
provide with lower values giving you
more creative results extremely low or
high values may give you unpredictable
results so I tend to go between 5 and 12
sometimes it can be useful to generate a
few images through batch with a low CFG
scale allowing you to get a more varied
set of images and then running an image
that you like through image to image
with a higher CFG scale to make
adjustments more closer to your prompt
in these examples you can see a variety
of CFG scales being used and I think the
values 5 to9 are more accurate to The
Prompt The Prompt Matrix is used to see
what impact your individual prompts have
on your generated image allowing you to
remove unwanted or unimpactful prompts
and keep the ones nearing you to the
image you want in order to use the
prompt Matrix start your prompt with the
subject of your image and then follow up
with the prompts you want to test with a
horizontal line anything nested between
or after a horizontal line will be put
onto a matrix allowing you to see and
compare the impacts of that prompt the
more specific your prompt that the more
consistent the results you will get
across your images allowing you to
identify the prompts which are causing
issues by singling them out I'll be
doing a separate video breaking down the
prpt Matrix specifically as it's a
useful tool which could do of its own
easy to find dedicated video but in the
meantime you can look at this image to
see the comparison in action the prompts
from file or textbox section will allow
you to test multiple prompts at the same
time either from the text box or from a
file all you have to do is put your
prompts in and use a line break to
separate them then generate and you will
get an image per line of your prompts
allowing you to see the comparisons for
each and I'll probably do a separate
Deep dive video on this topic so it's
easy to find here are some examples and
while you look at these if you wanted to
use a file then just put the prompts
into a notepad on separate lines then
drag the notepad file into the file
section and it will copy the information
over to disable diffusion for generating
XYZ plot allows you to test and compare
a range of variables on your generated
images and make comparisons against
those valuables such as a seed CFG scale
and using prompt Sr or search and
replace which allows you to replace your
prompt with a different prompt during a
generation to see the results now this
script has a lot of options so I'll
probably do a separate breakdown video
of each option so you know what they do
and you can see comparisons between each
of the options but I've used this
feature throughout the video to generate
comparisons such as a CFG scale so
hopefully you have a good indication of
what it does but to wrap things up I
hope this video has helped you to
understand prompting a little bit more
like the video before you leave the
building and subscribe to be notified
when my next video drops drops there's
also a patreon for those who want to
support this is bite-sized genius and I
hope you enjoyed