ComfyUI Modular- Ultimate Starter Workflow Usage

okay hello and welcome to another comfy UI tutorial uh today you get to actually see me in this image um it is a terrible webcam but someone said that I needed to do it so now you get to see me talk while I talk I guess anyway um to not waste too much time this is going to be a long video anyway I'm going to cover a workflow which I've created um and normally I'd have you learn how to build workflow normally I like to have you um work through the system with me so that you understand why each thing does what and why each node does what and you know how to set these things up and some of the background behind why are these things are the way they are um but in this one I'm going to do some of that so he'll he'll he should understand this afterwards but um I wanted to make a workflow that is adaptable modular because people were saying that they wanted to just have small units and learn just on a small thing and um so this is going to build modular units um and shouldn't show you how to set them up and how to hook up all the controls and then wire everything I mean look at this thing um there's a lot of stuff here um this is actually an active thing I've been using this to make art today and so it's actually in the middle of uh all the stuff I was doing earlier and um as you can see we've got controls we've got um you know we've got loaders we've got switches we've got image editing we've got gradients we've got colors we've got image loading we've got batch image loading oh we've got pre-defusion we've got image analysis and turning image analysis into a prompt and go to the bus and this stuff is just your bog standard K Samplers and upscale stuff there's also control net and Laura over here so it's a pretty fully featured workflow which I'm going to be providing you so um hopefully it makes sense now um it's a big system it's complex it's got controls in certain place and so on and it's designed mainly to do all the stuff before you actually generate stuff so this is where I'm first actually using the sampler which is all the way over here and you can see there's a whole bunch of stuff up here that comes beforehand um there's a little sampler over here but that's not being used as you would possibly expect instead um basically this stuff just makes a seed for you to sample off so you can see at the moment it's sending out kind of this red figure um it's sending out along this latent line here and goes to the Samplers and they all riff off of that one image and then if I want to if I find one of these that I like what I'll do is I'll copy it to the clip space and I'll paste it in one of these so I can paste it here and these are just load image nodes and the cool thing about them is they can just hold your image so when I find something good I put it in here and it just sits here until I'm ready to run up scales on it so this section is not even plugged into this stuff like there's one connection and that's for the the prompting but uh this stuff runs pretty much by itself like it when I click go it runs after the other stuff's done and that's pretty it's like a good way of just kind of setting stuff up while I'm messing around with settings up here I can um I can just be working on an upscale in the background while I'm doing other stuff that's pretty cool but you can set up however you want like you know so I'll show you so as you can see they're in groups and the reason they're in groups is because each group is its own little set of nodes and they're connected via um noodles to whatever else they need to be connected to now this one is connected with a whole bunch of stuff down the bottom all these and we've got these ones connected with staff we've got heightened width being sent all over the place uh we've got a whole bunch of stuff going in and out of this one you know these are the all this connection kind of goes between the different groups and each group is kind of self sustaining like its own thing if you grab the notes from each one of these groups and then you could put them elsewhere in your workflow if you wanted to um you know you could put image analysis way further down the line that kind of thing so may as well get started because we're already pretty far into it and I haven't even explained it all right so starting off uh let's talk about initial resolution so you can put these controls anywhere they'll drag and drop them wherever wherever you want to use them and initial resolution is how you set the resolution of your image like normally when you're setting up um a111 or whatever you set your resolution and then it generates an image so this is what goes into your first sampler so this is the resolution you'll go and everything in here it's set to that resolution and that is because if you're loading images um you could be loading 4K you'd be loading 8K or whatever so it's a good idea to make sure that they're a reasonable size because Bay decoding and encoding takes ages on big images so it's a good idea to keep these however as small as you can get away with or whatever um diffusion you're gonna be doing I'm using sdxl so it's using the uh the preset one of the presets from in here from this little list here if you're using 1.5 you want to be sticking to your 1.5 standard kind of razors um or you'll get clones and all that kind of stuff Okay so this sets the resolution of everything cry keep it just be careful dragging around groups because if you let this go on top of there and then move it it'll drag all that stuff with it as well which is annoying so um essentially all of that comes down to this thing here changes that resolution it changes it all over the place so you know you shouldn't if you mess up resolutions in here or in here or wherever it should fix itself in theory in theory and uh so resolution is done uh the next thing that we need to be aware of is let's go with what each of these things does so the first one is the color noise group color noise is used to create a color and noise and it has one other ability It's ability to create gradients and it can combine a gradient with a color and it can combine those two things with noise so what it's doing here if you have a look at this we've got a gradient and it's being blended with noise which is purely in noise which is a kind of noise it kind of looks like clouds and it uh it's blending with what's called multiply which is it'll only blend darker colors together so um you know if you had like a a scribble on a piece of paper and in Black Ink and you blended it with that it would just have the black from that paper would appear Blended here the white paper would be gone so the question is how do you use this setup well first off uh you've got an uh all right gradient no no so generate gradient is pretty simple it doesn't look simple when you're looking at it you're like what does this do so the first number is where the color is located at 100 percent so for instance this bit red is it 20 25 along the line of the image this could be vertical or this could be horizontal and top to bottom left to right and um so it's at the 25 mark on that and it is you know two five five zero zero which is red so RGB is the also the other numbers on these so you know red green in the middle blue on the end and so that's kind of how you set colors that's how you create a gradient because the gradient is formed by the um the space between those little nodes and it'll kind of smooth between them essentially and if you've got two that are the same the area in between is going to stay at 100 percent so if you look at this bit here I've got 25 to 40. they're in between those two points we keep stay at 100 percent so this is really good gradients are fantastic for making Landscapes if you want to make a landscape and you want certain colors it's amazing like really good it's like add a little bit of this to just before your sampler and it will fully influence the whole image like um that color is running through everything and kind of underlying all of these images a little bit not so much this one because this is at 0.8 denoisy but more this one anyway so that's how you do gradients and we blend them together in this system up here and it's separate because you don't want to be hunting around in these for things and you'll lose track of them and so here I have them here and I have it also notated what these things all do next to it um so we've got the float the the um sorry first blend node uh which is this one number one we have a look here it says blend B1 is blend one if you set it to one it's going to use color so it's going to use this orange color and if you set it to gradient it's going to use the gradient now if you set it between it will be a blend of the two and this is the case on all nodes which are using Blends like this because a blend is basically an image switch with steps in between I said if you put it on you know one or the other it's one or the other and if you put it halfway through you've got images information from both if it is set to 1 will be fully noisy so at the moment it is actually set to one however it is not showing the full noise because the blend mode over here is set to multiply if I set this to normal which is the default setting on these and then I run it it goes black and white because that's the default that's what the this purely noise outputs by default but we don't want that we want the other thing now I could switch this to green which is going to blend only the white um only the white Shades will blend with the color hang on still getting over a cold Weeks Later as you can see it's it's Blended the white with the background instead of going with the dark I hope we switch it back to multiply it should do it again but cancel that you'll see these images switching occasionally and that's because of this set up here is it's batch loading from my um my output folder like all the stuff I've generated in the past and uh so generally you can just ignore that but anyway so that's how the color noise thing works and once it's Blended those together it outputs from this little node here and where it outputs two is it outputs to the image processing group and it also outputs straight line all the way through to this switch now this switch is important because it's how you choose what comes out the end of this it's like it's how you choose what goes into the the bus or you sampler or whatever and um it's so everything everything usually has there'll be a color noise there's a diffusion there's an empty latent in here as well and image to image should probably actually renamed to what is it image processing to be honest but you also get damage to image through this so it's doing like four different things um which is why we have two switches to run whenever we're actually using anything through here we're using another switch to Define what this means which oh yeah it's complicated but I can only handle so many switches at once I'm gonna have noodles noodles everywhere speaking of noodles they are everywhere so this one can bypass everything and just send the straight up radiant noise here um and if we go down to here to the input sources which is where you get your image loaders uh your batch image loader and so on and they output obviously they'll output the blue they'll output to the same place so it sends one damage processing and it also sends another one over to this place over here I'll just reconnect it this is the blip analysis node and it is amazing it can turn your images into prompts it doesn't it doesn't grab metadata it looks at the image and says what does this look like and then it it's like a reverse clip basically so it'll pull words out and I'll say this is what this image looks like and it's it's not it's certainly not always right let's put a show texting over here and we'll run it there are text um nodes are amazing and then from the python gosses um for UI Scripts they there's also a comfy UI manager is awesome um it's what's giving me these green names next to things and you can activate that yourself if you want it's under the manager settings badge nickname all right so this is actually showing what it thinks this image over here is and it is being sent uh I think it's being sent the load image at the moment judging off of it and it is being sent it thinks it's seeing a train station with people in the station and waiting for their trains that's what it thinks it's saying it's not really it's about bunch of people around shops like a mall but you know it's image recognition is not perfect so that's what uh that is doing and uh the reason I'm doing that is I'm actually building my prompt with it um you can plug anything into here but currently this is like a little cool thing I've been messing with um I've been doing some prompt tutorials and uh this is going to be used in the next one to kind of show what what kind of words it thinks it should be included in everything it's complicated anyway um anyway I can build prompts for this and the other things and stuff I'll go on this later anyway sorry now we know where these things setting stuff but we don't know exactly probably you're familiar with load image I doubt you're familiar with load image batch and load image batch is an interesting note it's from West suit and it allows you to basically load a folder so you can go find a folder it could be on your network it can be on your hard drive and it will sequentially load all the images from there and then it'll run through the index of those images so starting from one the first image it picks I think they're alphabetical I'm not 100 on the auto it could be date I'm not sure but sort of run through them all one by one and then um the thing is normally it's normal kind of setting is to be manually you index it so it's kind of like a batching without it doesn't automatically batch more images like next image is not always progressed which is why we have this number counter here and then number counter just it just counts up every time we run Q prompt it adds a number and then we send this to the index so it adds one onto the index or two onto the index which gets us to the next image and so on every time we click so through doing this every time I click go it's loading a different image from this folder basically hopefully there's no risque images in there I don't think there is I think we're good I think we're good and there's also um the reset counter thing here which is just a Boolean constant node and it is sending either a one or a zero constantly and if it's sending a zero it's saying I don't want to reset so ignore me and if it's sending a one it's saying reset the index to one or zero zero I think it is and so it'll go okay this this will reset if it's set to one and you click queue prompt so make sure this is zero if you are trying to cycle through images and one if you don't want it to cycle basically anyway and these all get sent to an input switch over here and then it goes to an image upscale which is used to down scale stuff that's too big and make it the right size for the rest of the workflow and so the switches are sent through this little note here which goes all the way up underneath everything all the way over here and here we can choose which one of those is to be running so we could choose the batch or we can choose the image and as you can see up here that will tell you which one is which this area also allows you to pick um the options for this particular switch as well and they're also noted over so yes hopefully that makes sense now describe that describe that pretty Fusion is the next one I've described previously pre-diffusion a few times I've been using it for a while to make images and two quickly prototype ideas for things um it's very simple it's just a case sampler low on the step count with its own prompt its own um checkpoint you want to use the same checkpoint as the main one um simply because um if you do that you are not going to use as much RAM so it's not going to have to load and unload models and stuff now um the other thing is the prompt has to be really really simple as um be foggy um you know be simplistic be very vague just describing kind of composition really like you want gray in the middle of the thing or you know certain shapes and like a room and that kind of you don't have to be like describe a scene fully or anything because that's not what you really want from it you just wanted to sketch out an idea of where things are as you can see here it's just making a figure in a red um on a red background and it's just picking where it wants to be and I keep this unfixed because if it's on random and you run this and it'll just run this every time even if you're not using it so it's better to keep this on the fixed until you find when you want to send anyway all right now this one also gets sent all the way up here all the way up to this switch and then comes out here as pre-defuse so you can use that image by itself just like you can use color noises now if we want to use the other options which you'll notice all of those are also getting sent to image processing so image processing is where this kind of setup really shines and that is because you can cross blend all of those images together in here and you can also you can do visual effects so I'm just using a blur but you could add a sharp and you could add a color correction you could add color effects in in this kind of section here there's no real limit to what you can do with this setup um the controls are all up here using the same kind of layout I describe the the node and what it does here um hopefully I'm not going to go through each one of these because it'll take me like 20 minutes like it did the last time I recorded this and uh it should be pretty self-explanatory but basically mixing and matching these you can create different outcomes and they'll all come into this one spot here blending you know color the color noise and the 3D Fusion or the images you know all that stuff and you you pick which one is coming through up here essentially so if I'm going to use pre-defusion be on uh two if only use um this one it'll be image of image is one so this is now going to try and use whatever is being sent by this switch which should be the train station and it has it's it's centered through and it's blended with that gradient noise thing using these settings up here the light blue ones that's like 40 percent um of the noises all the noise is coming through and uh yeah so it's creating this image and then sending it out this area yeah and sending it through to my samplers and as you can see they are now interpreting it based over the prompt and they do make some pretty cool images but this is just the first step this is the bit where I send images and then I pick the ones I like can I put them up here and then I upscale them I don't really like to run batches I'd rather run individual um Samplers with um different keys and settings that's just my preference all right so once it's run that through and it's got a hold of an image that you want to make with all of that stuff and all that um you get to run the prompt over it and so it's going to use image analysis to make part of this prompt and you can see it's using inside the train station with people walking in the station waiting for their trains um it's adding that to this which is the subject which we have said ourselves a photo sends from 1970s dark survival movie nature survival in a different world bioluminescent wet alien Flora and then I'm saying in the background and you'll notice the name of this is where is text C that's because this is being taped to the front of this oh to the front of this text here in the background the inside of a train station is what it's doing and it's doing that because we want to ensure that this is our foreground and everything behind this is the background and that allows us a little bit more control over things prompt wise so we're not just mashing a bunch of random words together this way we can make sure that certain things come first because the things that earlier in the prompt are more important you can use um you can weight words and all that kind of stuff if you want as well and I'm also duct taping the quality of words on the end of it because the quality of words help with refining images and uh you know just putting the style on there after as well and of course there's a negative prompt as well and it runs through here and all that's duct taped together with the context concatenate node which is amazing I love these things where scenaries are probably the best because it gets you four of them instead of just two which you'll get with the quality of life suit it comes through texting code goes into your bus and you've got everything sorted now um and so as you can see it's able to generate images based off of those kind of settings and you can also send it up to here so what I'm going to do now is I'm going to make my own image and you can follow along if you want or you could uh stop watching because it pretty much explained everything I'm just gonna make my own thing and I'll kind of talk my way through it if you want to follow along that's cool all right first off let's load an image because we want an actual image that we want to use something specific um Batman stuff no we'll do yeah these these will work these will work much better got some landscape let's run yeah this one will do now we have a landscape um and we are running it so uh let's go up here and we'll be sure that it is being chosen so we've got one going through uh our processor uh resolution is good we are using image to image over here good this is using image loader good and now what do we got what kind of gradient do I want so there's some good sites around where you can grab RGB values pretty quickly um I'm going to grab about yeah well there was a okay yeah we'll go over this color puppy and put it in here and if we delete these ah I need one at 50. 50 and what color are we going to use um we'll use a darker color yeah that'll do it and what are we going to make the ground what color shall be the ground let's make it maroon whatever the color that would be it's kind of that coloring but all right good so now we have a different gradient set up so that how much of the color are we I want less of the color to you mostly the Great point one and blend wise how much noise do I want oh yeah why not we'll keep the amount we have so now we have the color noise set up with the image selected we're not using preview well so we can ignore it and we are running through the image to image setup which is the light blue top row here and I am happy with the settings for that pretty much it is ending calinoid to the image and we're going to change the actual way we're going to blend it we're going to blend there with yeah I think overlay is the one I want I'm just going to check all right all we want soft light that's what we want a soft light yeah and this way I can change how this first set of images is blending together you don't have to just always use normal you can actually change some of the settings and get different stuff and I don't want to blur it sir this should now output correctly the other stuff should be sorted what's my prompt looking like yeah yeah I think I'm kind of happy with that but I'm going to add some explorers group of explorers okay Explorer is Explorer yes okay and we got that and we get that okay okay all right it is now set up and ready to go so if we just click go it should go runs through here creates the gradient Blends the things together duct tapes them together and sends a preview here which is not really changed enough for my liking I'm going to fix that by changing this back to normal why not there's no reason not to anyways we'll see what output it's doing so as you can see it's um it's still able to make some really good images this way um which may be difficult to make um just putting prompt words in and getting the kind of colors you want getting the style you want could be much more difficult just using words because it's not going to understand when you tell it you want colors in certain places or that kind of stuff it really sucks at colors you said red Sunrise you'll have red mountains red people and that stuff so you can see it is still able to make some cool images even though it didn't do exactly what I wanted I'm going to copy this one actually and paste it what is this one there's some oddly sized human one okay that will run it again because I want to run I want to run this out and have an actual bit of color now it's still not quite there I need more of it so I'm gonna put this at six actually blending even more of the gradient in with it and that's going to create more of the same because the color profile and image is pretty much the same so I'll cancel that and we'll rerun it yes I am actually recording this I just I literally had to check because I have done like a 40 minute tutorial before and looked up and realized that I hadn't clicked record which is not Optimum okay now it's still not really good enough to be honest not really got enough of the color I want but I guess it'll do that's pretty cool I like I'm going to copy that one now um because I can copy these things and just paste them in here I can then copy them here and paste them here and this a particular load image loads directly into my upscalers and you might notice I have this this one is grayed out so if I control m i can unmute it which will activate my upscalers before it was disabled which stopped it from running this when I was running the other prompts and now it should run this it is using a canny control net it is running on running realistic Vision instead of sdxl and it is running with a iterative impact pack upscaler using pixel upscaling and it is also using the add detail 1.5 Laura in case you're wondering and if I run it it should it will try and run your workflow from where you are selected where you have a selected node by the way so if you have an area that's different from your main one and you click into that area it will go backwards and try to find way to where it should start and then it will run forwards and it should run your isolated bit before it runs the rest of it and sometimes I won't even run the rest of it all right so we are now refining this image and upscaling it and it has created a Kenny image over here and is loading correctly into the apply control net which is a good way of stopping it from wildly hallucinating when you're upscaling have control net applied and control net can be good fill um redoing an image at a higher level of denoising but I'd like to use it just to keep kind of the borders in place at like a mid level of denoising but anyway this is going to take a while or I may as well end it there and you can see that uh this system can be used to create all kinds of setups you can do all kinds of stuff with this

Channel: Ferniclestix

Views: 6,747

Keywords:

Id: ppE1W0-LJas

Length: 36min 51sec (2211 seconds)

Published: Thu Aug 17 2023