Krita AI Trick to Removing Objects With AI diffusion Plugin

hello and welcome to stream tabulous so I want to take a look at um Cryer and probably do a couple of little videos on um how to manipulate the AI diffusion to do certain things uh talked about in the previous video on um how to like add a cat if it's not doing it putting a little blob in there this one I want to talk about um perhaps trying to remove an object which is in the background uh C doesn't have a decent AI background removal tool and I want to take a look if we can find a way to um manipulate that AI into doing something so we'll go across to the intro and then we'll talk about um doing [Music] that [Music] okay welcome back now I've just opened up the um app roaming crer data file here um and if you've watched how to uh install it you would have shown I would have shown uh about the python Cryer file where your um add-on goes into now what we want to do is we're going to do something different we're going to I'm going to upload a file to my cloud that you can download and it's something that you'll be able to take advantage of when you're using the AI and it basically is a diffusion static and I've talked about how the um models don't really have a picture in it and the way diffusion generally works is it creates a static image and then it uses that static image to build a unique image every time sometimes s it gets caught in a loop so we're going to come into patents and then I'm going to upload a file for you all which will be one right here one static diffusion so this is the image here so um if you've done preview you'll find it starts with static and then it will build its image based on that static so this black section here for example might end up being a cat ear a cat face and a cat body so it's kind of like one of those magic eyes um is the way I see it um those old static images where you'd squint and try to distort your vision to actually see it so here we are so it's simple as going into uh the app data roaming CER go into patterns and drop the file in which I will link below onto my cloud now after you've done that then you can launch CR crer okay so now I've got crer open and I've loaded up the image that I'm going to be uh working on and for this system it's rather old and rather slow it's an old GTX 1070 uh so I'm not going to be able to use this at the full resolution that is an advantage that adobe has um has some secret Source where when you're actually doing a rendering you're select it it's working in its AI model the original model was train trained on 512 x 512 stable diffusion and the new model is trained on 124x 124 which is an sdxl model uh there are models coming out in the future in sdxl which will support 2K and higher so I believe you'll see that in Adobe Adobe certainly has its benefits um there's a lot of things that uh it just has some magic Source there that um the Cryer diffusion doesn't have so we need to know how to manipulate the Cryer diffusion to do what we want uh so I prefer Cryer diffusion and the only reason I prefer the um AI diffusion and CER is because I can actually select my models I love the live preview of being able to scribble I think that's great for kids and uh things like that and I generally like the tools I like the fact that uh it's a mix of what I know in stable diffusion and gives me the ability of a um a graphic user interface which is similar to Adobe or Corell um each has their benefits there's no doubt about that um and of course being on a pension I love the fact that this is free and I'm very very grateful to the open source people that have made it now I want to come through I'm going to come up to image here I'm going to come down to scale to a new size because this is way too big for my system to handle I'm going to come down to this one here and I'm just going to put a lower resolution of 1,000 actually might drop that a little bit lower just to make the renders run faster so I'm going to drop that to 800 and we have our image now we can always upscale this after and thanks to AI upscaling we're not going to lose um a lot of that detail which is good now what I want to do is I want to get rid of this vehicle here now no matter how I do it if I select this vehicle and I click on a 1.5 since this system will take forever on a sdxl so we come through to Absolute reality and then I'm just going to put Bush and trees and then I'm going to hit generate and um yeah this should not work now of course every time you load up a model and you run k for the first time it will take a moment to start its render because it's got going to load the model in the background so adobe's got some magic going for it um if you've used stable diffusion youve used something like automatic 11 or easy diffusion uh the in paint you're physically painting something it's not it's a selection but it's painting that selection I think what that is doing and um I'm taking some guesses with Adobe and uh some guesses with um well stable diffusion is the same across the board doy's model but with the way it works I think when you're painting it you're painting and it's generating static to actually create the image now it doesn't matter what I do here I'm going to end up with it reading the background and it going well there's a car there and it's really ignoring this so what I think is going on and we'll unhide this now if we come up to here our pattern fills we'll see our new one here which is static diffusion so we're going to select that we want to come over to settings Dockers and there should be one down here which is can't remember if it's tool options bear with me tool options is on ah tool options is up the top here okay so if you don't have your tool options I'm just going to bring that out if it lets me it's not going to let me it's locked in there we go okay this is a our tool options so if I come through and I click on something like the paintbrush you'll see that uh it has some options there if I click on the Paint Bucket you don't have that you want to come up to Dockers and you want to so settings Dockers tool options you see it's disappeared if I go settings tool options we get it here and then we can drop that anywhere on here so we can hold it over top and it will add a tab so I will hold it over here and just drop it back there and that gives me tool options and I can move that and lock it so it doesn't disappear once we have that tool options and we're on our bucket fill we've selected our pattern as the static diffusion which will be uploaded um on my cloud and you can download from the link below we want to click on patent okay let me change that we want a new layer so we're going to come up to layers new add a paint layer on that layer we're going to flood it with our static colors now from there we select our paint layer we can turn that preview back on it won't matter as long as we're selected on that now we'll come back across we'll hit generate now this will be completely different because what it's going to do is it's going to look at that static and attempt to um make an image out of nothing and uh this is the way the AI diffusion generally works so in my mind uh and the autistic thinking it's like well if I put static there uh technically it should read it as a new render and then try to build it but because it's looking at the outside of the image it's building based on the overall image as you can see that gives us our static there so we might want to come through and do some adjustments here now I mentioned this in the previous video for these selections that you may need to adjust them to do certain things I'm going to show you this is actually good that I made this error we're going to reset it which is going to redo its padding and its blending sections and then when we run it again we'll see we don't get this Shadow this Shadow is what we ended up with a gray Shadow when I was doing um my previous video on how to um add and change objects and what I found uh after doing that video and I put it in the notes that are modified underneath the um the video is that padding sometimes needs to be left as its default sometimes I do need to adjust it and change it but for the most part when it's on its default it Blends in a lot better to the actual surrounding information so so we'll give this a moment there we go so we can see here that we've got some beautiful blends with it so there's some padding issues but uh overall it's actually Blended exceptionally well and we can see that's removed it so that is how I'd recommend to do it now if we just drop that off again and we'll just select this area we'll select a bit more of the outside of it now what I find is depending on your model that you use you'll get um different renders again now this is a large model this one's 6 gig uh but it is a 1.5 model so it's going to take a little bit to render and we're going to click on that but this is the advantage um and what I like about using this over Adobe is depending on the artwork that you're doing if you're doing animations or something like that you can actually put the model in which is best suited to work with your art style um to what I can tell from Adobe and when I played with it and I only played with it for the week that it dropped out of um or become paid over here and that was before the November the 1st date um and when I contacted Adobe they said well it's a roll out it's by the 1st of November and some countries will be affected first so I found myself paying up for it and I think it was still in beta um someone said that the selection tool worked in Adobe uh or does now um I thought it was a snap so perhaps it was when it was in the beater when I first uh paid for it and um I assume there's a lot of implements to it and I certainly hope they fixed the white renders there was a lot of renders I was doing where it was just coming up with white pictures so uh but yeah I mean um that b experience put me off it and uh yeah I'll take free overid any day of the week especially when um the subscription is enough to upgrade my hardware and get a um use this for free and get better renders and faster renders by doing that and as I said as the models evolve it will be better and sdxl models uh if you're able to use them understand language better that is an advantage with the adobe's model training uh it has a better language model in there which is associated with the diffused images that it's done so it has more of an idea of what you're actually saying when you put the word remove in so we can see here this one's got a a bit of a um artistic style to it and of course these are Australian trees as well so I highly doubt many models are trained with the Australian trees so we can come down to realistic vision and we can press render again when you're on civity AI look for models which are suited to what you're doing most of the models that I've selected are for photo restoration so I'm looking at um skin textures eyes uh hand details um facial details facial hair um hair itself I'm not looking at uh nature so with this one you'd be better off going into CD Ai and looking for models which are uh more suited to the style of nature that you're actually working on so that's something to keep in mind there we go we can see there we're getting fantastic uh images that are blending in so you could um drop some of the saturation down on that which would uh blend it a lot more so we could okay so yeah you can come through and you can adjust the um the saturation to that to bring it in into the Australian sort of muddled up colors that we have over here of um no real greens so that gives you an idea um of how to use it there's a lot of information here in the rocks of static and colors so I'm just curious if I put reflective water in will it work as good as Adobe Adobe was really good for doing reflective water where it was reading the rest of the scene and doing it um now depending on how models are made and how that interface works it can be hit and miss so um easy diffusion I've done the um the water the reflective water and it's worked exceptionally well uh this uses comfy AI in the background um depending on the workflow they've loaded up as well can affect the way that it works with the overall image Im so this is more my curiosity and um yeah you'll notice I've left it as 100% as well uh you may want to drop that level down um you wouldn't really want to drop it down when you're working on the static because of the you then the static will start to actually show up on your image so you don't want to do that haven't tested it on buildings yet I've only tested it on nature so you can comment below on uh if this method is uh working well for you I do hope that um that Cryer gets out there more that's actually quite nice so we come through we do this we hit apply which adds our layer here and then we can come down to here and then we can apply that as another layer and then we have our reflective uh water in here so doesn't look half bad I mean question where uh some of these Reflections are coming in from um but nonetheless we could come through and easily clean that out if we wanted to or just leave it I don't think at the first glance anyone would um pick up on it unless they're worked with a lot of Photography and they got eyes like perhaps us that are doing this sort of thing so that 100% when you're working on your static image so I'm going to put that up in the link below to download that static pattern off My Cloud and you just want to come through and drop that in the patterns of your CER data folder uh as I showed in the beginning and hopefully that guy will help you guys actually um remove elements uh from a photo that you're actually working on and of course some of um the way I'm describing things is a uh a logical guess from patterns that I see uh which is something with autism you see patterns and I see when I'm running stable diffusion I see that uh when it builds an image it throws down static and seems to uh mesh and then create other static over top of that until it builds up its image to get its uniqueness and um I found in this application that perhaps putting down the cap for example in the previous one the white blob that that helps the AI and control what you're actually doing and then that lead me to I mention the static in the previous um video and this is the static that I talk about I did notice in one of the forums that someone did mention that putting down color actually helped and they mentioned that the diffusion works by Static as well which led me to well I'll create a a flood color and use it in this manner so there it is um that's my method for removing a object and uh using that static and hopefully this helps you out and if it does comment below and of course like share and get the Bell on and share this video to um you know your other social medias and get the word out I think more people that learn about the opsource uh crer um or creeder would be a good thing once more people learn about it and they learn about the AI diffusion which is on GitHub which means someone can clone the idea and change it further and get new plugins with more Tools in it uh such as perhaps selecting an area where it's automatically doing that um flood and like a scripted steps to render um might be something we could see uh and then we might actually see tools which are AI removal tools implemented in it I think that's exciting to see um something that is free and open source where the person that's made the plugins one person to best my understanding and I'll leave uh a link to their um reddits below so perhaps you can go in find some of their comments and thank them for their plugin the more people that actually thank them rather than the artists that hound and say this is a bad thing because they don't understand how AI works the more support that he gets the more he'll be motivated to work on it further is what my hopes would be so yeah go over and leave some positive feedback for the creator of that plugin for sure and I will see you in the next stream tabulous video thank you for watching my video and sticking around to the end if you like my videos it'd really help me out if you could like And subscribe it helps the YouTube algorithm to push my videos out there to more viewers which in turn helps me and helps everyone so thank you for watching my video and hanging around to the end and I will see you in the next video okay

Channel: Streamtabulous

Views: 2,141

Keywords: Android, photos, restore, art, Android photos, old photos, photo Restoration, fun, entertainment, Android apps, Ai, Ai art generators, stable diffusion, computers, krita, adobe, paintshop pro

Id: 0Uo4uU7RDco

Length: 21min 31sec (1291 seconds)

Published: Wed Dec 13 2023