[00:00]
what do all those symbols in stable
[00:01]
diffusion really mean how can we write
[00:03]
prompt words so that the pictures we get
[00:06]
are more what we want when you see those
[00:08]
complex stable diffusion prompt words do
[00:10]
you feel confused too don't worry this
[00:13]
video will help you with all kinds of
[00:15]
problems when you fill in prompts first
[00:18]
off let's check out the basic principle
[00:20]
of prompt words prompt words are
[00:22]
separated by commas and you can put
[00:24]
prompt words on different lines but you
[00:26]
still need a comma at the end of each
[00:28]
line as for weight each prompt word has
[00:31]
a default weight of one but the words at
[00:33]
the front get a higher weight so put
[00:36]
important words at the front if you can
[00:38]
finally keep the number of prompt words
[00:40]
within 75 if there are too many words
[00:43]
they won't have much control over the
[00:45]
picture next up let's talk about what
[00:48]
each symbol means first off there are
[00:50]
parentheses square brackets and curly
[00:53]
braces these are mainly used to tweak
[00:55]
the weights of keywords when you put a
[00:58]
hint word in parentheses it's weight
[01:00]
becomes 1.1 you can use up to three
[01:02]
parentheses at the same time when n
[01:05]
parentheses are used the weight of the
[01:07]
hint is 1.1 to the power of n curly
[01:10]
braces are also for increasing the
[01:11]
weight of a hint curly braces are used
[01:14]
to increase weight too each layer of
[01:16]
curly braces increases the weight by
[01:18]
0.05 the adjustment with curly braces is
[01:21]
much smaller than that with parentheses
[01:24]
square brackets are used to decrease
[01:26]
weight one square bracket makes the
[01:27]
weight of a hint word 0 point 9 times
[01:30]
less three square brackets make the
[01:32]
weight 0.7 to 9 the easiest way to
[01:36]
adjust weights is to add a colon after
[01:38]
the hint word in parenthesis and then
[01:40]
fill in the desired weight value
[01:42]
directly after the colon It's
[01:43]
recommended to set this value between
[01:46]
0.3 and 1.5 for example if you enter
[01:49]
cherry blossoms kittens green leaves and
[01:52]
set a different weight value for cherry
[01:54]
blossoms you can control the proportion
[01:56]
of cherry blossoms in the whole picture
[01:58]
the pointed bracket are mainly used to
[02:01]
call Laura the format is Laura trigger
[02:04]
word weight value after calling Laura
[02:07]
images with specific features can be
[02:09]
generated it should be noted that in
[02:11]
tensor Laura tritter words can be
[02:13]
directly used with one click which is
[02:16]
very convenient underscores act as a
[02:18]
link to make the keywords more closely
[02:20]
connected for example a milk cake might
[02:23]
be understood by stable diffusion as a
[02:25]
glass of milk and a cake but if milk and
[02:28]
Cake are connected by Under scores then
[02:30]
stable diffusion will understand milk
[02:32]
and cake as a whole next let's check out
[02:35]
some Advanced syntax now how do we
[02:38]
control when a prompt kicks in first off
[02:40]
you can use a combo of square brackets
[02:43]
and colons there's a number after
[02:45]
cologne it means the prompt starts
[02:47]
rendering from that point a double
[02:50]
cologne that means the prompt in
[02:51]
parentheses renders till that point when
[02:55]
there's one prompt it's like this but
[02:57]
there can be two prompts in parentheses
[03:00]
like this the two prompts are separated
[03:02]
by a cologne so for the first 70% of the
[03:06]
rendering stone is on and for the next
[03:08]
30% flower is on if the two prompts in
[03:12]
parentheses are separated by a vertical
[03:14]
line like this it's alternate sampling
[03:17]
so red hair and blue hair prompts
[03:20]
alternately and you'll end up with red
[03:22]
and blue hair after you get the hang of
[03:24]
the basic syntax let's check out the
[03:26]
recommended way to describe the overall
[03:28]
prompt words picture quality and
[03:31]
painting style words have a big impact
[03:33]
on the overall look of the picture so
[03:35]
it's a good idea to write them on the
[03:37]
top I've got some tips on picture
[03:39]
quality prompts mostly they work for
[03:42]
everything but for different styles of
[03:44]
pictures there will be some specific
[03:46]
quality words too like pixel style Pixar
[03:49]
style ink painting style and so on add a
[03:52]
comma at the end of this and you can
[03:54]
start a new line if you want then
[03:56]
there's the description of the main part
[03:58]
of the picture my character age
[04:00]
hairstyle hair color what they're doing
[04:03]
and so on the more detailed you are the
[04:06]
more accurate the generated picture will
[04:08]
be add a comma after writing and start a
[04:11]
new line next up is the description of
[04:13]
the lighting in the environment like a
[04:15]
Snowy Evening or a Sunny Meadow add a
[04:18]
description of the lighting used at the
[04:20]
end of this the last line is for adding
[04:23]
the trigger word of the Laura you want
[04:25]
to use after positive prompts fill in
[04:28]
the negative prompts
[04:30]
generally just filling in some general
[04:32]
negative prompt is fine if you're using
[04:35]
the flux model it won't let you fill in
[04:37]
negative prompts on your own when you're
[04:39]
using tensor's classic workbench you can
[04:42]
just fill in the prompts in the language
[04:44]
you know after that click the
[04:46]
translation button right below when
[04:49]
you're out of ideas for the prompts you
[04:51]
can choose to generate them randomly or
[04:53]
by recognizing pictures oh and you can
[04:56]
also check out the posts recommended on
[04:58]
the homepage to get some
[05:00]
inspiration there are a bunch of tensor
[05:02]
models for you to pick from all of
[05:04]
tensor's models are online no need to
[05:07]
download they're one click and don't ask
[05:10]
too much of your graphics card best of
[05:12]
all they're free you can get free
[05:14]
credits every day by doing some stuff
[05:17]
click the website link under the video
[05:19]
to give it a try any questions or things
[05:22]
you want to hear about next time just
[05:25]
subscribe me in the comment section see
[05:27]
you