SUPER Stable diffusion 2.0 Settings Explained! Free Ai Art On Your Computer Made Easy!

Key Takeaways

hello humans my name is k you're a overlord and as promised in today's video I will be explaining every single options in settings that are present in the super stable diffusion 2.0 version that I showed in my previous video and before I begin I would like to plug my own Discord server that I highly recommend you to join because this is probably the easiest way to get an answer to your question you can create a support ticket or just you know talk about technical stuff or just showcase your art that you create in the stable diffusion or any other text to image AIS so definitely join the community if you are interested in AI or if you have any question about installation or any other AI related topics so this video will be separating in three different parts in the first part I will be giving you tips and tricks on how you can use super stable diffusion 2.0 the second part is where I will be explaining all the settings and options that are present in the super stable diffusion 2.0 0 and the third part will be kind of like a small jumbo shooting session when Ankara explain how you can solve the most common errors and problems that I see in the comments in my YouTube videos now the first question is how exactly do you update the repository so you have basically two ways of doing this the first one I've already showed it multiple times in my previous videos but basically if you come here you click here on the folder URL and you type CMD and press enter this will bring the command prompt window and here you're gonna type get pull and you press enter and this will update the repository and the second way is probably what I advise you to do because it's way easier you're gonna come here on the web UI user but you're gonna right click and edit with notepad or edit with notepad plus plus and here just above a line 8 called Web ui.bat you're gonna type git pull and then You're Gonna Save the file and what this is gonna do is that each time that you launch the web UI user bat file it will automatically update the repository which is really nice this is basically an automatic update each time you want to launch super stable diffusion 2.0 now the second trick is for people who want to create a public URL that they can share with other people and the way to do this you're gonna come here on the web UI user.bat and here in command line args you're gonna type Dash Dash share and then you press save and what this will do is that it will create an additional public URL that you can share and use with other people but that also means that other people will be able to use your computer resources to create images now another trick is for people who don't have a very powerful GPU don't have enough vram to run the stable diffusion at full power and one trick that you can do is again come here on web UI user but in the command line args and type dash dash Med vram and what this will do is then it will basically use less vram to process the operation that you're trying to do now this will take obviously a little bit more time but if you don't have a very powerful GPU this is probably the way to go and if Med vram doesn't work for you if you're still running into the Cuda ultimate memory issue what you can do is that instead of using main vram you can use low vram this will use even less vram to use their operation but of course it will again take way more time and here as you can see now if we double click on the web UI user web file since we input the command git pool in the file when I double click on it it will launch the automatic update as you can see right here this way you don't have to type gitpool ever again okay so now I'm gonna try to explain all the settings and all the options that are present in the UI but I'm gonna tell you this is pretty insane because I've been researching this video yesterday and this morning I wake up I update the repository and now there is a brand new thing here in the interface so one thing that I noticed which is a good thing but also a bad thing is that the author of this repository called automatic quadruple one is very very quick and very good at in implementing brand new stuff brand new changes to the repository which is great which is great absolutely fantastic so if you are for example a new feature that you want to request um he is very quick at implementing it but the problem is that it seems like each time you implement something new the documentation doesn't really follow so sometimes you might Implement a brand new feature but not talk about it and you can just launch the interface you're like oh there is a brand new stuff um how do I use it well I have no idea I don't know how to use it because this brand new thing was uh introduced right now just you know a few hours ago and it says here when you hover on it regeneration parameters from prompt into user interface um okay great but then when exactly that does that mean does that mean that if I put some parameters such as you know steps and stuff like that and when I click on this button this will populate the interface but unfortunately I tried it out but it doesn't really work if I type steps like I don't know 50 does it work if I type maybe column doesn't work maybe there is a I don't know something else I I have no idea I don't know what that means there is a button right here which for now doesn't really do anything I don't know how to make it work unless there is a new documentation trying to explain how that works uh this is for me right now I use this button so you know okay so that aside we're just gonna start with the first settings tab and then we're gonna go through all the other tabs um or maybe before I would like to say something about the PNG info Tab and last time I said that this PNG info time was kind of useless because all the internet is basically read some hidden metadata in a PNG image then in reality someone in the comments I'm sorry I don't remember your name told me that in reality this is actually super useful to be able to remember what kind of image and what kind of parameters were used to create this image for example if I come here and I take one of the images that I created and I don't remember what kind of steps were used and there you go if we put it right here just drag and drop you're gonna have here on the entire prompt that was used to create the image with all the parameters the step the sampler the scale the C the size and even the stable diffusion model that was used to create this image which is actually really really cool and super useful so you know I apologize to PNG info tab you are pretty great okay so let's start on the left right here we're gonna go from left to bottom then middle to bottom and then right to bottom okay we're gonna go through this pretty quickly we don't want to waste too much time because um there is only you know a very few options that you really need to take into consideration most of the other tabs most of the other options you're not really going to be using them okay so the first option here the first tab saving images and grids um checkbox here always saves all generated images obviously you want to leave it checked out file format for images if you want something else other than PNG like jpeg or something else but I highly recommend you just keep with the default PNG the images file name pattern is actually pretty interesting if you want your file name to follow a certain pattern here if you hover your mouse here on the images file name pattern you can have the explanation so you can use the following tags to Define whole file names for images are chosen with that you know the brackets step CFG prompt Etc um but to be honest I don't really think you really need that especially since now you have the PNG info tab that basically does that for you so if for example you want to remember what kind of prompts were used in an image you can just drag and drop here in the PNG info Tab and you're gonna have all the options then you're gonna have save all generated image grades which you can check it or not I'm Gonna Leave It by default because this way I can basically have all the representation of all the images that was generated on all at once you can uncheck it if you want if you don't need image grades and you just want to see the final images again file format PNG leave it on default here again an extended info see the prompt to file name so this is basically like an extension to the um the field right here the image file name pattern again as I said previously you can just use the PNG info tab you don't really need any of that do not save grids consisting of one picture obviously this will just create basically a copy of one picture now obviously this one just leave it on default you don't need a copy of another picture if there is only one green raw count just use the default Auto detect here save text information from generation parameter as chunks to PNG files very important to leave it checked because otherwise the PNG info time will not work if you don't want to use the PNG info tab you can check this box that says create a text file next to every image regeneration parameter and what this will do is that each time you create an image it will create another text file with all the parameters of the image I'm not going to be using that because it creates a very messy situation when you have a mix of images and text files it is again way easier and wins and way simpler to just use the PNG info tab again save a copy of image before doing face restoration um this could be useful so this one you can either really check or uncheck um I'm Gonna Leave it unchecked because I usually really like the end result that I get with face restoration but if you are afraid that it will mess up the final result you can just check the box and this way you're gonna have a copy of an image without face restoration and an image with face restoration quality for safe jpeg images leave it at 80 is pretty good you have a good balance of quality and compression now this one you can leave it checked you don't want a big dimension image taking too much space on your drive now this option right here is really up to you I'm just gonna leave it unchecked wherever I don't really care about the name of the output and file names and stuff like that but if you need absolutely to have the same exact file name so that you don't kind of mess up with your project you can check the box this way every single image will have the exact same as the original one that you input in the uh into a stable diffusion path for savings uh just leave it all by default I think this is really good the way it does it really creates a different output for like text to image image to image from extra jobs it really just it creates a very good um it creates a very good organization it's very organized it's very clean um this is probably the best way to kind of organize your images into different directories so just leave it at that unless you want to create of course something different but seriously I don't think you're gonna be using this okay so here uh pretty much exactly the same thing just name it and default unless you want to create a some directory with a very specific name pattern because you have some sort of project that needs to be a very precise very organized in a very specific way but otherwise just leave it at the default okay so here in the upscaling zone you have title size from es organum scanners um basically you don't really need to worry about any of that here because the default version the default values are good enough but basically this has to do with the way the op scanners work it basically works by tiles and larger tiles larger size styles basically produce less visible seams as you can see here this is an upscale with ESR again and you can see here some seams that are visible along the red guide markers and the kind of the bigger the tiles are the less visible they are but the same time you are using more power but the default values that you see here works really well at balancing quality and performance so you won't be able to see any visible seam so just leave it by default here Sonic which real is our game models you want to use uh I highly recommend just keeping these ones um these ones are really just the best one I really like I really enjoy my favorite is probably the plus anime 6B um it really creates a very sharp image um it really works really well when you have an image that are a little bit more fantasy or anime but for realistic landscape or realistic images you may want to use this one but these two are just perfect at what they do so here again tile size and tile overlap for swing image restoration basically the exact same thing as the one I explained earlier just leave it on default you don't need to touch any of that it works right out of the gate and here are the options for the ldsr which is another upscaler and here again just leave everything by default everything works perfectly well and here you can choose by default the up scanner for the image to image option I personally recommend just limit by default and choosing when you're creating your image but I really just use the real Sr again ones they are the ones that are faster and create really good output images but mostly because I just really like the look that they produce but this is just my opinion of course some people will probably prefer like maybe ldsr Lexus so just try it out and see for yourself which one do you prefer now here you have the face restoration model between cut former and gfpgan and apparently cut former is better than gfpn so I highly recommend that you check cut former as default and that new input software blank 0.6 for the wave parameter this way by default it will use the code former face restoration model in instead of GF began it will create a phase that is more close to reality than if you do not use the face restoration model and here this option here if you don't have a lot of vram in your graphics card you can basically offload some of that work into your computer RAM but of course it will take longer and here when using the save button only save a single selected image I would probably leave it at the default and don't check it because you want to save all the images that are available so here in the system Zone don't touch any of that everything is good by default this is really not important it just creates more information for you but you don't really need any of that okay so now we're in the Staples diffusion Zone this is kind of where the fun stuff start a little bit because here we start with the stable diffusion checkpoint you can see here that I have two different models the model.c kptpt and the waifu model.c kptpt yes just when here you can click and check which model you want to use for your in this generation and to do this just to be able to change the different models you need to go into your folder models and here you're gonna put all the ckpt models that you want this is my 1.4 the normal version and this is my wifu model version that I tried in a previous video and once all your models are here in your models folder it will appear in this section right here and you can change on the Fly which model you want to use for your next Generation which is really really cool you don't need to rename anything you don't need to change anything you don't need to move anything just put them in the models folder and then you can select them in the settings tab now here the apply Color corrections to image to image results to match original colors is basically a color correction automatic color correction that basically applies the original colors of an image through the new one so for example here and created a new prompt and put this image a portrait of Christina Hendricks and put in this prompt which did not really work very well because I asked for a portrait of Christina Hendrix with a mustache I don't really see any monsters right here but whatever so this is basically the image without color correction and this is that same image using that color correction option so as you can see here this is before this is after the color correction and as you can see here the same colors that you see here were then applied to the new one so it's really up to you to see what kind of results you want I'm just gonna leave it off when if for some reason you want to try it out you can simply check this box but I also I highly recommend that if you do check this box that you check this box right here that will basically save a copy of the image before the color correction this way you will have a copy of the before and the after but I'm just gonna leave them unchecked now this one with image to image doing exactly the amount of steps the slider specifies and it's not quite clear what that means it probably means that if you don't check it the system will basically optimize the generation so that it don't necessarily use the amount of steps you specify on the slider I could not find any documentation on it and I did not see any differences with and without it so you know just leave it at default now this one enable quantitization in case simplers for sharper and cleaner result basically it improves the image but not really by much it is not as impressive as you might think like for example this is an image before and this is an image after um if you see any differences let me know because I don't and here's another difference before and after the difference is basically none almost unnoticeable but again it's a matter of opinion maybe you do prefer the slang difference it's really up to you but I'm just gonna leave it at default and unchecked okay so next is use text between the parentheses to make model pay more attention to the text and text between brackets to make it pay less attention this is awesome this is absolutely amazing I will show you an example right now what that means so basically if you use parentheses on a word the model will pay more attention to that word between parentheses and if you put a word between brackets it will pay less attention here's a great example of a photo in eggs and bacon on a frying pan and you can see here that bacon is between parentheses a lot of them and you can see here on this image right here that means that you want the final image to contain more bacon and the less parentheses you put the less it will pay attention to that word so here you have the example with the bacon which the more parenthesis you put the more baking you see and here with the eggs if you put more parentheses you will see more eggs in the final image and vice versa by using brackets so this is really an awesome addition so this one make candy Fusion Samplers produce same images in a batch as when making a single image I'm not quite sure when exactly that means again not a lot of documentation on the subject so just leave it by default the filter and the CFW content that you can check if you want to I don't know why but you can and here's the other categories for random artist selection when you use the roll button this is something I will show you later which is really really cool if you don't have any idea you can basically click on a roll button here and this will add a random artist to The Prompt so if you click here it will you know just add Adolf Willette or matte fraction Clements Archer and stuff like that so if you don't have any idea for what kind of image you want to create what kind of artist you can basically click on this button and it will add on random artists from a list so that's you know that's pretty cool here you have the interrogate section for if you want to basically use the image to text option um it's again it's not really that useful I mean you can use it if you want but I rarely get very good results you can basically leave everything by default you don't really need to touch anything unless you heavily use that option for some reason you can for example increase the amount of the minimum description length so that you get basically more details or more artists for the final prompt but to be honest I didn't find that this option works very well and finally you have the user interface none of it is really interesting except this one the progress bar because I remember someone in my Discord server saying that he liked to see basically the progress the generation of the image on the screen kind of like the it creates a new animation and if you want to enable it you you can simply increase it to 1 and this will basically create an animation when you create an image so I'm going to show you right now just you know click apply settings I'm gonna go into my text to image and here I'm just going to create a brand new image I'm just going to take a prompt at random um I'm gonna take this one for example paste it here and then click on generate and as you can see here there is now a little animation of the image being created from scratch and I'm just gonna leave it run until the end and there you go so you know if you like this you can leave it at one I personally don't really care so I'm just gonna put it at zero so that's really up to you okay so now let's talk about the three remaining tabs that we have right here the text to image image to image and the extras and let's start with the text to image which is probably you know the tab that you know the most so obviously here is when you put the prompt here as I showed previously is when you can add a random artist into the prompt so this will create a brand new image using that r artist style that you see right here here is the negative prompt so this is where basically you can input an information that you don't want to see in the image this is something that I showed you in the previous video also if you remember the example that I showed you in the previous video here's the original image here is the negative prompt using the word purple and it creates an image without the color purple the negative prompt tentacles so that it doesn't have any tentacles in the funnel image so let's say for example that I don't want any grass or adult 12 any green I'm just gonna input here negative prompt green press generate so as you can see here it did not work exactly as intended since you can see a little bit of green but it has less green than the previous image so it's not that bad but since it kind of goes against this style of the image that we wanted to create since this is supposed to be a castle in the forest trying to delete all the green is probably gonna be quite difficult so that's probably why it did not work and you can definitely try it out yourself see what works one doesn't just have fun with this you know so here the restore faces option is basically to use either gfp Gan or code former as we saw earlier in the settings tab so if you want your final image to have a better face I highly recommend you keep it checked so the tiling checkbox basically creates a seamless texture that you can use for a texturing for example I'm going to try to show you what it looks in the end so this is what it looks like with the tiling option and if you think this image and you put this image into the seamless texture checking website you can see here that this image is indeed a perfect seamless texture you can increase the size and you can see that this texture it basically is perfect and you can use this for infinite texture so it's pretty cool so the iris fix is actually really cool because if you're ever had pictures with like much multiple heads on multiple images of the same character in the image checking this box right here will basically get rid of the problem and eliminate all the extra heads that you might get if you increase the width or the height of the image so that's a pretty cool fix here you can choose a random seed or reuse a seed from the previous generation and then you have the script options here and you have the option to either choose a prompt from a file or a text box and prompt Matrix or an X Y plot so the problems from file or text box is well kind of what it says it does so as it says right here you can simply upload the text file and it will read the problems that are in send a text file now the Matrix file I'm gonna show you what it does it create basically this kind of grid image with all the different prompts written here on the left and on the top so you can better see what words influence the final image in what way and to do this you need to separate the words that you want to see in the final image by a vertical sign character so for example you can see here in the example is a busy City street in a modern city then vertical line illustration vertical line cinematic lighting and the result that you get is basically four different images when you have a busy City with cinematic lighting here you have a busy City with cinematic lighting and illustration here you have only a busy City and here you have a busy City illustration so only by using two arguments here it creates four different images and the more arguments you use the more images it will create as you can see here here's another example with five prompts and 16 different variation so that's really useful if you want to study how stable diffusion works and what words influence the funnel image so the X Y plot creates a grid with different parameters that you can put right here so you can see here in the example you have an initial prompt right here portrait of Medusa gorgona then you input The Prompt FSR with the values here they have chosen different artists I've seen the final results and here they wanted to see if by increasing the scale by 5 each time they wanted to see what kind of final image they could get so with all of these parameters this create this final image this funnel grid image with all the different artists right here with that same exact prompt that you see here the initial one and for each additional scales by 5 this creates some brand new image by using the word of the artist that you can see on top it's a little complicated to explain but I think you can see what kind of results you can get again this is really useful if you want to study or stable diffusion works and what kind of results you can get with different words okay so now let's proceed with the image to image tab so here is where you add actually implement the image that you want to use as a base here is when you interrogate you can use the image to text option to try to understand how the image was generated well at least in hearing because in practice it doesn't really work that well here you have three tabs image to image in paint and batch image to image so image to image you know is when you transform an image into another one using the prompt right here batch image to image is the same but this time it's multiple images at a time the in paint I've already shown it on a previous video as when you basically create a mask on an image and then create a new image from that for example I'm going to show this one and I want to change your face maybe I want like a I don't know like a mustache I'm gonna draw a mask here so you can either draw directly on the image or upload the mask that you created somewhere else so implant mask if this is selected basically everything that you mask will be modified if you check in paint nut mask is basically everything else outside of the mask that will be modified obviously since it's about a mustache about what I'm asked to be modified so the mass content feel basically feels with colors of the image the original Kip whenever was there originally the Latin noise fill it with latent space noise and let it nothing filled with latent space zeros basically look like this this is the mask this is the feel this is the original latest noise in limited nothing so it doesn't really matter what you choose it will be in painted anyway so in paint a full resolution basically allows to kind of get a better image in the end so I'm just going to increase it for this one it's only this this one basically just resizes to the Target resolution but you might get incorrect half speed ratio this one will basically crop the image that sticks out and this one will fill with basically the image colors now the sampling steps right here I'm just gonna leave it at maybe let's take 75 so apparently Didi I am is based at impainting so I'm going to choose this one I'm not going to check restore finishes or dialing I'm gonna leave everything here by default and just click on generate and there you go here is the final result which is not bad actually it's a pretty good result I'm still not as good as what you might get with Delhi for example but it will surely get better with time but in on this version of him painting is free so you can use it as much as you want kind of training trying it out a little bit trying to see what kind of result you can get so it's always nice to have you know especially if you don't currently have access to daily 2. now here in the bottom in the script option you have a lot of options right here now we've seen some of them we've seen the XY plot the SD upscale is basically a simple upscaler The Prompt Matrix we've seen it the prompts from file or text box we've seen it also now we have here the poor man's old painting and out painting MK2 these are basically two different different scripts that allows you to Output an image and output it basically creates a brand new image outside of the initial image that you created right here so here for example this is the initial image and this is the out painting created with the script for men's alt painting and this is the same result the same exact prompt with the outpainting MK2 so it doesn't work as well technically the outputed image actually looks better and this one integrates the initial image way better so it feels like if you want to have a better alt painting final image that integrates the initial one better you might want to use the poor man's out painting now you can play around a little bit with the the options the number of pixels to expand the max blur Etc only want to increase the maximum or a little bit more you have to see what works what doesn't the outpainting direction also if you just want the the left portion the right the up down so this is really nice to have so the loop back option is actually pretty cool it allows you to basically automatically feed the the finished image back into stable diffusion so that it creates some brand new image so for example here in the example it's starting out with this crude drawing of that a bunny girl this created a brand new image and using the loopback function this image was then feedback again into the stable diffusion to create this image that image was then filled again in stable diffusion to create this one and etc etc to finally create this final image right here and here is where you basically choose the amount of Loops you want the amount of times you want your final image to be reintroduced into the cycle and this is the the noise in strength that you want on each Loop if you don't understand something you simply hovering your mouse on the option and it will give you an explanation on what this option actually does finally you have the image to image alternative test now the image to image alternative test is kinda strange I have to admit I never really managed to make it work but basically it's supposed to make it easier to modify an initial image now there is a guide here on how you can use it but I never really managed to get the result that I want and basically as you can see here you have an initial prompt with that image that was reconstructed and then modified with different prompts but to be honest I try to make it work and I just couldn't so if you do manage to make it work please let me know because I just don't really get the results then it says it does and finally we have the extra tab which is pretty self-explanatory it doesn't really have a lot of options there is basically an upscaler and a face restorer here you can open one image or in the batch process process multiple images is here you have your choices of upscaler again as I say you should probably use the real laser again or the plus anime one these are the ones that I use most of the time and here we have the visibility for the face restorer Jeff began a code former and as I said confirming is probably the one that you want to use because it apparently creates better results so I highly recommend you kind of play around a little bit and see what kind of results it does for you because it really depends on the kind of images that you have initially so just play around a little bit and see what kind of final image you get so here's a little troubleshooting session for all of you so if you have any error with python when you try to launch the web UI user.bot file someone sending the comment that instead of clicking on the install now button you should click on customize installation where you can choose the location and all the features of python and again don't forget to check these two boxes right here 2 installation for all users and add python to path as is very important and if you have any problems if you have any trouble if you have any errors with python that says that python is not found or something like that just click on customize installation check every boxes and then install python in the default folder now another issue if you have a graphics card in Nvidia in the 1600 series or if you just have black images so one thing that you can do is right click on web UI user.bat file click on exit with notepad plus plus and here on set command lines argas you can type dash dash precisionfull dash dash no half now this should solve any black images output but unfortunately it will also use more vram to produce the images so you can add another argument that we saw earlier on and that is Dash Dash net and here I am and then you can save the image and this will basically cancel out the additional vram that this operation requires and if it still again doesn't work instead of putting made vram you can use low vram and then save the file now yes again this will take more time for each image generation but you know at least it will work and there you have it folks you should now have a complete understanding of all the options that the settings present in the super stable diffusion 2.0 so if you like this video don't forget to subscribe and smash the like button for the YouTube algorithm and I'll see you next time bye