Stable Diffusion Tips and Tricks: Automatic1111 Tutorial

recently a new release of stable diffusion automatic 1111 was launched in this video I will show you some of the new updates as well as some tips and tricks I have learned over the past year first check to see if you have the latest version if you start stable diffusion from the web uiuser dobat you can see the version number in the command window it says v1.6 but a new version is available in your web UI folder click on the address bar type CMD and press enter a command window should open at the exact location you are in now type get leave a space and then type pull this should start the update process however if you have edited some files it will not update and will ask you to commit or stash them if this happens you can type git stash and press enter then run the git pull command again I did this and now when I start stable diffusion I can see that it is at version 1.7 if you go to automatic 1111 on GitHub you can find more information about this interface to find out about the latest release look on the right side for the word releases and click on it to see all the releases you can check what has been added what was changed and uh what bugs were fixed I will talk about uh some of these changes in this video um for those who are using sdxl models here are some tips you can add some command line arguments to optimize your stable diffusion depending on what GPU you have and how much vram you have you add these arguments in thebat file uh which you can edit with notepad so after the equal sign you add your arguments uh as you can see I already have X formers there because I have more than 12 gbt of vram and that is what's recommended if you have lower vram check the other options you can add I also have theme dark added here this activates the Dark theme on my stable diffusion now we are in the automatic 1111 interface with the Dark theme enabled as you can see it has already opened with some settings that may differ from yours you can change these settings to your liking for example you can choose a prompt that should appear every time or set a negative prompt size or any other preference when you're ready go to settings and scroll down to others to find the defaults when you click on it you'll see several buttons click on view changes here you can see the values I've added then click apply and the reload UI button to reload the interface after reloading you should see all your settings applied by default for the next tip Let's test a prompt for a portrait of a cat hit generate to get your first generation but what if you want multiple generations and don't want to click generate each time you can rightclick on the generate button and choose generate forever this will start generating image after image until you cancel it to cancel you right click on the generate button and choose cancel generate forever now let's say you clicked generate and it takes too long or you already see that it's not what you want you can interrupt the Generation by clicking interrupt this will stop the process at that step and you can see the image is unfinished here you also have a recycle bin icon that you can use to delete your positive and negative prompts Additionally the arrow icon lets you bring back a generation from a prompt or the last generation if the prompt is empty uh let me show you how you can save a style for example if I have a prompt of a cartoon cat character in a digital painting style I will add to the negative prompt what I don't want like not look looking like a photo not being a 3D render or not being ugly after hitting generate to see if I like the result to create a style from this I click on the edit Styles button that looks like a brush here I click on copy the main UI prompt to style now think of a name for this style I will name it cartoon character for the prompt unless you want cats and all your Creations you leave only the Style Elements in this case I will remove cat once you click save the style is saved you can close the window now let's delete what we have in prompts to show how it works let's say I generate an image with the prompt dog you get a random style for that dog but now if I go to the style Tab and pick the style I just created and click generate you will see the result is in that style a cartoon dog character let's create another style for a photo portrait that looks cinematic maybe at Golden hour for the negative prompt I don't want it to be a digital painting a 3D render a painting ugly and I don't want bad eyes either after removing the cartoon style hit generate to save this style we do the same steps copy the prompt give it a name though this is a bit long it's just for testing and remove woman from The Prompt so we can make portraits of any subject save the style and close the window now let's test the style on a cute [Music] cat pick the style from the Styles drop down and hit generate and you get a cute cat portrait at Golden hour um you can also have multiple Styles selected at once so now if I run with both the first one in the list is most important in this case it's the portrait photo with a touch of cartoon if I change the order with the cartoon first and the portrait you can see it's now a combination of cartoon that is more realistic this way you can combine different styles to create unique new Styles look how cool this cat looks all these styles are saved in a file named styles. CSV located in your web UI folder it's a file with text that you can edit if you maintain the same format you can also add your Styles here directly just make sure to create a copy of the file beforehand in case anything goes wrong this precaution allows you to experiment with adding Styles manually while having a backup to revert to if needed click on your uh edit Styles button again now if you want to delete a style just just select it and hit the delete [Music] button as you can see the style is gone and only one style is left in this release they say they have added support for SSD 1B let's see what this means on civit AI you can search for that text as you can see it has found a few models with that name the one you want is the model with segmen in the name if you click on it you will see more details this is a distilled smaller version of sdxl offering a speed up uh I will download it and give it a try you need to put this into the web UI folder then look for models and then for stable [Music] diffusion it is quite big so it will take some time to download depending on your internet speed once it is downloaded you can hit refresh and you will find it in the checkpoints list this is a smaller sdxl model so if you have low vram it will run faster than a normal sdxl model before I test the new downloaded model Let's test a normal model like the realism engine to see how long it takes to generate as you can see it took around 12 seconds now I will switch to the SSD 1B model and generate again to see if it's faster it takes some time to load so wait until it's finished I will hit generate and let's see if it moves faster it took 8.5 seconds this might not seem like much but on older PCS with lower vram you can see the difference now let's send this image to image to image to show you another trick some people have asked where the interrogate button is under the generate button there there is this paper clip icon that says interrogate clip if you click on it the first time will take longer because it will download a model but after that it should be quicker this will give you a prompt based on the image you have in image to image you can see how it's loading and when it's done I get the prompt it's not perfect but it's a starting point you can copy and paste this prompt into the text to image Tab and try it now let's try with another model the same prompt to see how it Compares I will choose Juggernaut version 7 and after it's loading uh I will hit [Music] generate now I've got this not sure how to call it a thing I first assumed it was something wrong with the Jugger not model so I tried another one and got the same thing or should I call it it the thing H and I thought maybe the seed had some problem so I tried a random one when these strange things happen What do you do you restart stable diffusion by closing the command window now everything works okay I encountered this type of error only when loading a different model after the SSD 1B model so there might be a bug where it keeps something in the ram now for more tips on the interface as you can see there are various tabs at the top yeah and you might not need some of them because you never use them for example I don't use checkpoint merger and train if you go to settings and scroll down to user interface you'll find different options like a quick settings list where you can add more stuff to the interface you can change the tab order or hide UI tabs I will will choose these two options when you're ready apply the settings and reload the UI as you can see those tabs are now gone also what's new in this release is that you can now find things in the settings more easily by using the search function so you can search for a word and find all those settings easily you can change the path for saving as you can see I put all my images in the same folder for me it's easier this way but the point is that you can set the path to your folder here and organize them how you want say you want to generate or test multiple prompts for example you can generate a portrait of a cute cat if you want to use the same prompt but with a different animal you can go to scripts and select the XY y z plot for the X Type choose prompt Sr which stands for search and replace in the x value put the word you want to search in the prompt in this case cat and then add as many other subjects or words you want to replace it with like a dog lion or maybe a fish and here you have some extra settings now when you hit generate uh you will get get all those images first is the cat then the cat is replaced with a dog then with a lion and so on from those settings you can change if you want to see The Legend on top it's really good for testing uh different prompts if you open the folder where it's saved you can see the individual images for each prompt now what do you do if you want to generate multiple prompts but each prompt is different in the scripts select prompt from file or text box in this section you can add all your prompts uh with each prompt placed on a separate line and you also have the option to import a text file containing all the prompts so here you can have a variety of prompts such as one for a photo another for uh a painting and another featuring a landscape and so on I forgot to delete the prompt before running the script so uh all my prompts are starting with the text I added in the text box and they're ending with the cat prompt as you can see here therefore if you want the prompts in the script to be unaffected remember to delete the main prompt if you go to settings and search for settings in UI you will find some settings for text to image and some for image to image you can enable tiling on the text to [Music] image and then apply the settings and reload the UI you will see that tiling is now available uh in the interface let me test a prompt without the tiling option active so you can see the difference before and after um I will create a rose seamless pattern uh if I preview tiling you can see it's not tiled it's just a normal image now I will activate tiling and try again this time the image appears as a repeat seamless pattern but has some watermark over it I'll try a different seed to see if it happens again this is the AI trying to imitate human behavior by adding signatures and watermarks these are not real signatures I once tried to put my name as an artist and it created a signature with my name now a trick to reduce the chances of that happening is to add the following in negative signature Watermark text words now everything looks clean so we have a seamless pattern that tiles perfectly did you know you can add an icon to your stable diffusion in the web UI folder go to the batch file from where you start your interface and create a shortcut for it then give that shortcut a name right cck click and go to properties click on change icon and you can choose an icon from Windows icons or browse for other icons once you found one you like select it click okay apply and okay again now you can copy this shortcut to your desktop or any other folder and now when you double click on it it will start the interface pretty cool isn't it let's talk about composable diffusion first consider the result of a cartoon chair and a pillow when you hit generate you get what you expect now let's use a method to allow the combination of multiple prompts by using an uppercase and the result is a chair that is also a pillow let's try another prompt so this is the result for a cat and a dog and when I change the word and to uppercase this is what happens it starts as a cat then gradually changes a little to a dog so it is both a cat and a dog you can have multiple ants let's add a bat in the mix to see what [Music] happens and that's how monsters are born moving on to the last tip add more weight to a word uh or increase attention to a word in the prompt um I have a cat and a dog and as you know the words at the beginning of the prompt are considered more important now if I select a dog with a Santa hat in the prompt and press control and then the up Arrow I can add more weight to that section of the prompt making it more important use the up and down arrows to increase and decrease the values watch what happens with a value of 1.5 to the cat it transforms into a dog if I select the same Tech and add even more to that value the dog will become even more important now if you select the same text and press control in the down arrow a few times until nothing is left it lets you remove that weight alternatively you can delete the brackets and numbers manually you can also decrease the importance of a word for example by pressing control and the down arrow a few times I can decrease its importance so if I make cat the less important word the dog from The Prompt will have more influence [Music] you can find a lot of information and tips for stable diffusion on the automatic 1111 GitHub page here's an example of what I just talked about on how to increase and decrease importance so on the page of stable diffusion web UI you can click on Wiki then go down and click on features then scroll down and have fun I hope you enjoyed this video if you learn something new leave a comment if you know a trick that can help others also leave a comment you can find more tutorials on my pix Aroma Channel thank you for watching and have a great [Music] day

Channel: pixaroma

Views: 4,411

Keywords: stable diffusion, stable diffusion automatic1111, stable diffusion automatic1111 tutorial, stable diffusion tutorial, stable diffusion video tutorial, stable diffusion tips and tricks, automatic1111 tutorial, automatic1111 sdxl, automatic1111 stable diffusion, automatic1111 tips and tricks, ai image generation, stable diffusion scripts, ai art tips, stable diffusion xl, stable diffusion ai, Stable Diffusion, Automatic1111, edit styles automatic1111, a1111 v1.70, pixaroma

Id: J7h1l5zhfAU

Length: 18min 52sec (1132 seconds)

Published: Mon Dec 18 2023