Nine Tips and Tricks for Automatic1111

welcome back I'm silicon thomaturgy and today we are going to cover nine tips and tricks for improving your experience in the automatic 1111 GUI for stable effusion while I hope each and every viewer finds several valuable tricks there are bookmarks in the video description so you can skip through ones you already know how to use now let's get to it have you ever wanted to reuse the settings for one of your Generations then gotten frustrated trying to find the precise values while there is an option to generate text files with settings for each image this is a bit of a red herring because there is a much easier way all you have to do is drag the generated image into the prompt box and the settings used generated will appear in that box to load these settings in the GUI you'll need to press the blue button with the diagonal Arrow which is just below the generate button one crook this method is that it doesn't load the checkpoint used in the drop down instead it appears near the bottom of the interface as a small box saying model hash while this will ensure that the correct model is used it's easy to predict it's there so I would recommend exiting it off and then loading the checkpoint from the drop down manually interestingly this also works for grids showing not just the settings for a single image but all the settings for that XYZ plot even better when you press a blue button automatic 1111 will populate all those variables in the XYZ script plot unfortunately this is not true for control net while the control net settings show up in the prompt box they do not import with the rest of the settings have you ever wished automatic 1111 let you know when it completed a long job so you could do something else in the meantime you can have automatic 1111 play a sound when a job completes by adding an M3 file with a name notification in the base folder for automatic 1111 after you place it there you'll need to restart the GUI for it to take effect have you ever gotten sick of clicking generate over and over again waiting for an image that's good enough to use as a starting point to keep stable Fusion churning out images with your selected settings right click on the generate button and select generate forever it can be quite mesmerizing to watch it make image after image however to stop you can't just click interrupt as the only sub generation of the current image instead you need to right click with the generate button and select stop generating forever as many of you have experienced certain settings like the upscale for image to image are buried in the settings tab even though you might want to change them fairly often to get around this you can set up quick settings which will display in the row at the very top of the GUI by default the checkpoint drop down is the only quick setting to set these go to the settings tab then go to the interface sub-tab and scroll down until you see a box called quick settings click on the box then type in part of the name of variable you want to change and it will filter down the variables by your input my personal three favorites are image to image upscaler vae and clip skip once you've added all that you want you'll need to click restart GUI button at the bottom of the screen for them to tick effect as most users know prompt emphasis is an essential tool for fine-tuning your prompts especially if you're using loras the Syntax for doing this is string the words with a parenthesis for emphasis brackets through de-emphasis or a parenthesis with a colon and a number for using a custom value however you don't need to type this out each time instead you can just highlight the words into your prompt or negative prompt then press Ctrl up to increase emphasis or control down to decrease emphasis by default this increases or decreases emphasis by 0.1 for regular words and by 0.05 for lures and Hyper Networks however you can change the increment by using the settings in the user interface tab as an example I increase the increment for regular words to 0.15 and increase the increment for loras to 0.1 let's say you've found a seed in prompt that gives you the perfect composition you want but the details leave a bit to be desired changing the prompt could ruin that perfect composition you work so hard to get so what do you do as it turns out you can add to the prompt without changing composition by using prompt editing this is done by surrounding a section of the prompt with brackets then after the text add a cone then a number which is either the number of steps or a fraction the total number of steps to Define when those words take effect since the image composition is defined pretty early on you can probably use values as low as 0.3 without issue for that however you will probably want to wait until 0.5 or even higher if you want to keep facial resemblance however this is a trade-off the longer you delay the less impact the extra section of the prompt will have if you are going to run XYZ plots with multiple models or vaes you should consider enabling caching them in Ram to reduce the amount of time it takes to load them this setting can be found on the stable Fusion tab in settings and adjusting it is as simple as moving a slider switching between models still takes a second or two but it is much faster than loading from disk even for a solid state drive like mine loading it takes about 10 seconds for a 2 gigabyte model and proportionally more as the model size increases with 64 gigabytes of RAM my PC is able to max out both of these without any issue even with 32 gigabytes you should be saved to sit to that three or four unfortunately this setting does not work for sdxl models so you'll have to endure those long load times for those as we all know using stable Fusion has a tendency to fill up your hard drive with all the checkpoints and vaes taking up a substantial amount of space each suggested by Duran PT and my vae video using symbolic link is an easy way to reduce the amount of hard drive space needed at least if you chose to define a vae for each model so what a symbolic link well basically it makes an imaginary file that points to the file you really want to use this way you only need one copy of the file but can access it from multiple locations typing out the necessary text command line is time consuming and error prone so I made a spreadsheet to help you out which is linked in the video description you'll need to make a personal copy of the spreadsheet in order to use it all you need to do is paste four things into the spreadsheet the folder which contains your automatic 1111 models the file name of the model without the file extension which in this case is safe tensors the folder for your automatic 1111 vaes and the VA file name with the extension once that is done type CMD into the windows search bar and open the program called command prompt finally copy and paste the contents of the green box and press enter you should get a message saying that sabalt link has been created and a new file will appear in the folder I also made a second variant so you can do the same with models if you're using a UI that you can't just point to a folder here is a trick for essentially photoshopping an object into an image though it does require some work an image editing software that supports transparency and layers this trick is very useful for making people hold objects which stable Fusion isn't great at first you start with your base image and an image of the object you want to add to it for the object you need to make sure the entire image is transparent except the object you want to add now open your base image in your editor and add a new layer add the object you want to add to that layer and make sure its position and size precisely how you want it now save it this will be the version We import to controlnet now we are going to make a second version with transparency so it gets automatically masked and in paint use the magic wand tool or similar to highlight the whole object next switch to the base layer and delete that area finally make the object layer slightly transparent then save this version as well now we go to automatic 1111 and go to the end pinning tab import the image with a partially transparent area as the main image automatic 1111 has a bug so it doesn't look like it's masked but it is set the denoising proportional Talent you want to change the object then import the full image into control net and set your model and preprocessor of choice enter a prompt for how you want that object to look in your final image and click generate now the object should be inserted and integrated into your image if this video taught you some useful tricks for automatic 1111 like And subscribe to support the channel if you really like these videos and think I deserve a couple extra bucks consider supporting me on patreon where you can view ad-free versions of these videos and get recognition here like my patreon anonymous as always thank you for watching and see you next time

Channel: SiliconThaumaturgy

Views: 4,106

Keywords:

Id: TV5a-OXXk-I

Length: 9min 51sec (591 seconds)

Published: Sat Aug 26 2023