[0:00] hey everyone I'm b-size genius prompting
[0:02] instable diffusion can be a mystery and
[0:05] it's sometimes tricky to know what does
[0:07] what but there are techniques that you
[0:10] can use to get the results you want and
[0:12] in this video I'll be breaking them down
[0:14] so you can spend less time reading and
[0:16] more time creating but like the video
[0:19] and give it to me bite-sized so let me
[0:21] quickly cover some basic information
[0:24] prompts are ordered from most important
[0:26] to least important from the top to the
[0:29] bottom from the left to the right there
[0:31] were various theories on how you
[0:33] structure your prompt for the best
[0:35] result but it's worth keeping in mind
[0:37] certain Concepts such as the subject
[0:40] lighting photography style color scheme
[0:43] doing words and much more which help to
[0:46] build up your image when it comes to
[0:49] style prompts can influence this
[0:51] alongside your desired checkpoint as
[0:53] stable diffusion was trained on a
[0:54] multitude of data sets from across the
[0:56] internet meaning you can draw references
[0:59] to art style
[1:00] celebrities clothing types and much more
[1:03] to influence your image The Prompt
[1:06] sections have these numbers on the top
[1:08] right called token limits and these
[1:10] refer to the maximum number of words you
[1:12] could fit into a chunk of 75 tokens for
[1:15] example if you had 100 tokens then it
[1:18] would process 75 tokens and then 25
[1:21] tokens independently into stable
[1:24] diffusions unit this in plain English is
[1:28] how the AI language model breaks down
[1:30] and manipulates text for processing The
[1:33] Prompt box is where the magic happens
[1:35] this is where you will describe
[1:37] manipulate and design your image through
[1:39] the text to image section or alter
[1:42] images if you're using image to image in
[1:44] conjunction with your reference photo
[1:46] you can put as much text as you like in
[1:48] here but often it's better to keep
[1:50] things short and sweet so your prompts
[1:52] are easier to fix orine as you make
[1:55] adjustments closer to your desired image
[1:57] for example let's type out medium shot
[2:00] of a woman with four lips golden eyes
[2:02] and a white crop top long black hair
[2:05] snowy weather octane render we can see
[2:08] that stable diffusion will attempt to
[2:10] generate its interpretation of that
[2:11] prompt and this interpretation will
[2:13] change depending on the seed image size
[2:17] and other settings which change the type
[2:18] of image generated from the same prompt
[2:21] the negative prompt box is where you
[2:23] tell stable diffusion what you don't
[2:25] want in your image and this can include
[2:28] literal Concepts items weather or
[2:31] artifacts and bad Anatomy within an
[2:34] image again you can put as much text as
[2:37] you want in here but it's good to keep
[2:39] it reasonable to make life easier for
[2:41] yourself later down the line for example
[2:44] let's type out bad dream unrealistic
[2:47] dream nons saafe for work and we will
[2:50] notice that we get a much higher quality
[2:51] image that's safe for work it's worth
[2:54] checking the example images of the
[2:56] checkpoint you are using to ensure that
[2:58] you are using the best NE negative
[3:00] prompts for your chosen model the
[3:02] parenthesis is used to put a greater
[3:04] weight or importance on a word than your
[3:06] prompt and for each parenthesis wrapping
[3:09] a word it would increase its attention
[3:11] by a factor of 1.1 and the further
[3:13] parenthesis will multiply the attention
[3:16] by 1.1 I'll be honest in saying that the
[3:19] mass behind this isn't really all that
[3:21] important as experimenting with your
[3:23] prompts will yield better results but
[3:25] it's still important to see this in
[3:27] action so you can find two your images
[3:30] so for example if I were to write out
[3:32] our previous prompt but put snowy
[3:34] weather in parenthesis then we can see
[3:37] how the snow and even the clothes start
[3:39] to turn either white or have a much
[3:41] snowier presence the more parenthesis we
[3:43] use square brackets are used to reduce
[3:46] the weight or the importance of a word
[3:48] in your prompt and for each square
[3:50] bracket it will decrease the attention
[3:52] to the word by 1.1 and of course we'll
[3:55] multiply the detention by 1.1 for each
[3:58] pair of square brackets WRA a word again
[4:01] it's better to experiment when it comes
[4:03] to fine-tuning your prompt but by
[4:05] understanding what these functions do
[4:06] you have another tool in your belt for
[4:08] fine tuning images so for example if I
[4:11] were to write out our previous prompt
[4:14] but put snowy weather in square brackets
[4:16] then we can see how the snow is reduced
[4:19] especially around the hair and the
[4:20] clothing the more square brackets we use
[4:23] let's address prompt waiting you can
[4:26] manipulate prompts to either add or
[4:28] remove waiting or importance from words
[4:30] within your prompts and what this means
[4:32] in plain English is that you can control
[4:35] how much impact certain words have over
[4:37] others within your prompt and words with
[4:40] greater impact will be visualized more
[4:42] strongly in your image this is done by
[4:45] wrapping your word in a parenthesis but
[4:48] we will now add both a colon and then a
[4:50] number which can be a whole number or
[4:52] with decimal values so for example if I
[4:56] were to write out snowy weather with a
[4:58] variety of prompt ratings we can see how
[5:00] the impact changes from no promp waiting
[5:03] to a really high promp waiting value we
[5:05] obviously get rather lowquality images
[5:08] at very high waiting so keep the numbers
[5:10] reasonable you may have seen these
[5:12] angled brackets in certain prompts
[5:14] perhaps some websites that provide
[5:16] checkpoints and luras these are known as
[5:19] embeddings and are common in luras where
[5:21] a file and multiplier for the file needs
[5:24] to be specified to determine the
[5:26] strength of the Laura the typical format
[5:28] would be Laura file name and then
[5:31] multiplier and unfortunately you can't
[5:33] do prompt editing with Aur in this
[5:36] version of stable diffusion in this
[5:38] example I've used the add details Laura
[5:41] which adds or removes detail from our
[5:43] generated images and we can see the
[5:46] effects of changing the values within
[5:48] our embeddings and how it is structured
[5:51] now prompt editing is a powerful way of
[5:54] controlling your generated images by
[5:56] swapping the prompts being used for
[5:58] generating an image during the
[6:00] generation a good way to illustrate the
[6:02] format is by using from to and when
[6:06] where from determines what prompt you
[6:08] start with two determines which prompt
[6:11] you end with and the Step at which the
[6:13] switch takes place is determined by when
[6:16] you can also do from and when with two
[6:19] colons in between to remove the prompt
[6:21] being specified after a fixed number of
[6:24] steps determined by when and don't worry
[6:27] I'll show you this in practice so you
[6:29] don't feel like you're in another Mass
[6:30] class but remember this key piece of
[6:32] information the weightings you provide
[6:35] in decimal numbers are percentages of
[6:37] your sampling steps which must total up
[6:39] to one meaning 0.5 is actually 50% of
[6:43] your total sampling steps anything l a
[6:46] whole number above one will be the exact
[6:48] sampling step you want to specify
[6:51] meaning a figure of 20 means stop or
[6:54] switch at sampling step 20 so for
[6:57] example you can see the impact in these
[6:59] images with our first image using
[7:02] decimals which represent percentages to
[7:04] transition from snow to Sun after a
[7:07] certain number of steps the second image
[7:10] uses whole numbers to represent the step
[7:12] for the transition based on the 30 steps
[7:14] we are using and lastly we can indicate
[7:17] at which step we should stop using the
[7:20] selected prompt s weather Al together
[7:22] now here's a call trick using a
[7:25] backslash before a special character
[7:27] such as a bracket or par parenthesis
[7:30] will turn that special character into
[7:32] ordinary text so using it in practice
[7:35] we'll remove the effect of the
[7:36] parenthesis giving you snowy weather as
[7:39] pure text with no effect you can see an
[7:42] action here where we have the standard
[7:44] prompt snowy weather compared to snowy
[7:46] weather within the parenthesis and then
[7:48] the literal parenthesis represented as
[7:50] text alongside our prompt and you know
[7:53] it worked because the snow is less
[7:54] prominent now thinking back to the
[7:57] tokens which form chunks I described in
[7:59] the the beginning the break keyword in
[8:01] uppercase will fill the current chunks
[8:03] with padding characters and adding more
[8:06] text after break will start a new chunk
[8:09] I personally do not see any practical
[8:11] use for breaking up your chunks
[8:12] prematurely before hitting that 75 token
[8:15] limit but if you wanted to achieve this
[8:17] then this is your solution now the
[8:19] horizontal line is used to trigger
[8:22] alternation or a loop in prompts where
[8:25] words are broken up with horizontal
[8:26] lines and are given the chance to
[8:28] influence the generation repeatedly as
[8:30] stable diffusion Loops through the words
[8:33] within the square brackets so if we were
[8:35] to use long black hair brown dreadlocks
[8:38] orange curly hair as a prompt then the
[8:40] first step will be long black hair then
[8:43] Brown dreadlocks then orange curly hair
[8:46] this is another technique for
[8:47] controlling the types of generations you
[8:49] get during the generation process and
[8:52] you can see in action how this impacts
[8:54] the final result the CFG scale will
[8:57] determine how strongly the generated
[8:59] image should conform to the prompt you
[9:00] provide with lower values giving you
[9:03] more creative results extremely low or
[9:06] high values may give you unpredictable
[9:08] results so I tend to go between 5 and 12
[9:12] sometimes it can be useful to generate a
[9:14] few images through batch with a low CFG
[9:16] scale allowing you to get a more varied
[9:19] set of images and then running an image
[9:21] that you like through image to image
[9:23] with a higher CFG scale to make
[9:25] adjustments more closer to your prompt
[9:27] in these examples you can see a variety
[9:30] of CFG scales being used and I think the
[9:33] values 5 to9 are more accurate to The
[9:35] Prompt The Prompt Matrix is used to see
[9:38] what impact your individual prompts have
[9:41] on your generated image allowing you to
[9:43] remove unwanted or unimpactful prompts
[9:46] and keep the ones nearing you to the
[9:48] image you want in order to use the
[9:50] prompt Matrix start your prompt with the
[9:52] subject of your image and then follow up
[9:54] with the prompts you want to test with a
[9:56] horizontal line anything nested between
[9:59] or after a horizontal line will be put
[10:01] onto a matrix allowing you to see and
[10:04] compare the impacts of that prompt the
[10:06] more specific your prompt that the more
[10:09] consistent the results you will get
[10:10] across your images allowing you to
[10:12] identify the prompts which are causing
[10:14] issues by singling them out I'll be
[10:17] doing a separate video breaking down the
[10:19] prpt Matrix specifically as it's a
[10:21] useful tool which could do of its own
[10:23] easy to find dedicated video but in the
[10:26] meantime you can look at this image to
[10:28] see the comparison in action the prompts
[10:30] from file or textbox section will allow
[10:33] you to test multiple prompts at the same
[10:35] time either from the text box or from a
[10:38] file all you have to do is put your
[10:40] prompts in and use a line break to
[10:42] separate them then generate and you will
[10:45] get an image per line of your prompts
[10:47] allowing you to see the comparisons for
[10:49] each and I'll probably do a separate
[10:50] Deep dive video on this topic so it's
[10:53] easy to find here are some examples and
[10:56] while you look at these if you wanted to
[10:58] use a file then just put the prompts
[11:00] into a notepad on separate lines then
[11:02] drag the notepad file into the file
[11:04] section and it will copy the information
[11:06] over to disable diffusion for generating
[11:09] XYZ plot allows you to test and compare
[11:12] a range of variables on your generated
[11:15] images and make comparisons against
[11:17] those valuables such as a seed CFG scale
[11:21] and using prompt Sr or search and
[11:23] replace which allows you to replace your
[11:25] prompt with a different prompt during a
[11:28] generation to see the results now this
[11:31] script has a lot of options so I'll
[11:33] probably do a separate breakdown video
[11:36] of each option so you know what they do
[11:38] and you can see comparisons between each
[11:40] of the options but I've used this
[11:42] feature throughout the video to generate
[11:44] comparisons such as a CFG scale so
[11:47] hopefully you have a good indication of
[11:48] what it does but to wrap things up I
[11:51] hope this video has helped you to
[11:52] understand prompting a little bit more
[11:54] like the video before you leave the
[11:56] building and subscribe to be notified
[11:58] when my next video drops drops there's
[12:00] also a patreon for those who want to
[12:01] support this is bite-sized genius and I
[12:04] hope you enjoyed