Photorealistic inpaint in Stable Diffusion.

hello my name is Laura meshopen in this video we're going over in painting tools inside the stable diffusion with the magic 1111 installations we're going to look on the workflow and what is best to use I'm using RPG version 5. you can download this from save it AI easiest way to do this is by utilizing extensions you can then download it extension Civic AI inside the extension tab I will provide for you down below link how to install this extension and how to use when you installed save it AI just going down and Search terms type RPG click refresh and after short time you'll notice right here the model will be displayed what we're looking at this with specific RPG mine is currently installed and you can see it's in the green frame if you don't have it any frame it's mean this model or checkpoint not yet installed or if it's orange color then it's been outdated because this is RPG version 5. to update you can select them model you want it if it's already installed and you want to update just click delete Model and after delete you should have the button says download model then click on this model and download it after download complete going on your model selection you can click refresh which is update list and from drop down you can select it as well you can select it by going inside the texture text click on the textile inversions actually checkpoints sorry and in checkpoints you can also select directly from here the model which will upload after upload complete we ready to work let's go ahead and image to image and we're going to utilize one of the image to using in painting for the test I'm going to use this image of stone gargoyle and I just want to replace the face if we just utilize image itself we're going to paste the beautiful girl portrait face mask out of stone it's what I want to replace here and without in painting just straight out we'll surely select properly model let's go down let's go ahead and use about 50 steps we're going to utilize zuler a just for the faster processing the current model actually recommend to use the GPM 2 plus M cars but again the Euler will work just fine in this case and size I'm going to be sure we're matching our original size of the image and we'll go down most important it will be the noise string if you're using by 75 it will overlay with heavy noise over our image let's click to render and you can see it does not look at all what we needed the good balance if you just want image it will be somewhere around 0.3 0.2 and you can see it did create beautiful face similar but a problem when we created will last a lot of details so it does created the portrait of the girl as a stone similar what we have it with our gargoyle but not as we lost all of these beautiful details in the background to preserve these details it's why we want to use it in painting tools to do this we'll just need to go down below and click on in paid this will take our image and all parameters that we set and passed inside our image next what we want to do we want to select the mask for this I'm going just use mask tool and select the face that I want to replace or area where I think it's applied it's very important not to go over this Edge we want to be kind of close but leave enough the face you want to run here let's go down below mask blur it will determine blur on the edge of the Mask how much we want to blend in also mask mode in painting it's mean inside the mask tool it's where the impending happen if you want preserve just a place where you pink select the in painting knot mask on this case it will reverse our mask next we have it mask content feel original like noise and Link nothing this is different ways applying the noise I found in many cases latent noise or original work the best next one only mask padding pixels we leave it for now as 32 default if we want to decrease it will create a bit sharper look for the picture next in painting areas responsible for whole picture or only masking it is analysis of the image if we're using all image it will analyze also around the areas and create more realistic if we have it only mask I'll show you shortly you will see it will be different between outside and inside so let's leave it this as default steps and everything will leave it as default here including the noise string and let's click generate as generate completed we can preview any notice right here we still have it all as high details preserve and face is beautifully replaced with new what we wanted instead let's check if we select only mask it on this case analysis will happen only inside in some cases it's work okay but also it will create a different coloring and you'll notice right here how it's come closer and it's great it will look it's not as we don't have the same textures as before it did try to match the coloring on the background but it does not match all the details that we have it in previous and make little bit smoother look another nice options about creating you can utilize additional extensions like for example a rope to replace face for example if we're going extend a rope and we'll take it and paste some of the face and currently I will just use this image we'll go click enable leave it everything as default and let's go click generate notice we're still inside so it will analyzing inside and the last step should replace this face with our Roop as it recognizing the face as well and right here we have our image with face replacing notice again because we're using in paint we don't have this enough details we can go back and switch to whole picture in this case to analysis and let's run this one right here we have our render and notice we have a little bit more details however anytime when you have it a rope it will try to replace with original face which it does not it does very good job but we still lose some original of details this is just example let's uncheck group and another options I want to show you because in some cases you'll find these techniques maybe does not work as you want in this case we want to use it in paint upload it will perform about similarity only give it as ability to upload mask instead drawing and in many cases it may work better because you can create little bit more precision and another tools like for example Photoshop created additional mask more accurate zoom in and have it all these details I already used some other selection tools so in this case what do you do same things we'll posted image here and we'll take our mask and usually I create a specie with transparent white color representing our mask we leave everything as default at this point included our the noise string and after this just click generate and we don't cover the roof anymore so you can see we generate it perfectly with this mask as a sign and our face even called it same texture as a rest of the body which is much very good of course you can play around a little bit more to try find right seed then you can post it inside as a recycle so it's already uses the same face for you if you desire to do so okay let's go showing a little bit the experiment in this case and I'm going back to just the in painting what we've done before and I'm going to modify the noise string so you can see the difference that you can achieve the higher the noise string the more noise will apply over and as a result we can have it more flexible effect but you'll notice it start playing too much out of the image so for example let's go to increase to 7.5 and we'll go render this and as a result you can see a big difference on the faces because it's allowed more flexibility in some cases it's maybe what you're going but many times I found it's a little bit more unpredictable and if you're going too low for example 0.5 let's go around maybe seven eight you'll notice it's not enough noise and image will stay and resemble too much our original and even maybe lose some of the tail so right here you can see kind of like a regional image because not enough noise for our image to work around and extrapolate data from that and you'll notice 0.3 work very well you're not just necessarily need work just with the face you can also create very nice animations inside your still images by utilizing animate GIF on converted this to PNG and what I was meaning by nice animations you maybe see sometimes they have it like moving water or moving cloud or other elements so let's show what example you can do for right here we can see we have a house and we have it chimney so if I just go ahead and draw this area around on the top of the chimney and instead this I can put it smoke it's all what we need to do let's go ahead remove our seed because we want to do I'll leave it everything as default and without animations I'm going to click generate just to preview what we have it and in some cases because it is not enough noise information will need increase it's maybe work very good on face but in this case we need add more flexibility and we'll go back to about 68 7 and click generate this maybe take few tries till you find the correct one in this case I found that the late noise work a little bit better as well I increased to 7.75 so we have a more permanent smoke visible in our case and when you like it I do recommend for you click to lock so we have it similar effect up there next when we're done let's go have our animated div expanded and click to enable animations I leave it everything as the Founder's case if you don't have it animated install as extensions I will provide another link down below for you to also check and install animated div on your computer after this completed let's go ahead and click generate this will take a little bit longer time because it is neither regenerate our 16 frames that we have selected before and combine them to create our smoke now let's create animations we can preview all the frames it's preparing be sure they're coming very good it may take a little bit time depending on how many steps you selected and what resolution and how fast is GPU on your computer okay and after some time right here is completed and we can preview you can see small animation smoke of course we can done like left and right a little bit move it to create more interesting but I think this way you can very easy to add nice small torch and live bring to your images the options you can also animate leaves sky and smoke in this case if you have the water the water ripples but I highly recommend for you to animate one element at a time because if you just type clouds and smoke you'll get literally mess around because it's one recognized so just be sure you animate one element at the time another thing someone point to you that images we created they are much smaller size than original images to increase size because we can go and send them to the extra way inside we can upscale the images but I found out there are another applications like gigapixels work much better and produce greater result on a quality it does utilize internal Ai and I have very good review about hard work and you can see it's add quite a bit details beautiful details because it's utilized AI operation on the backend to analyze what is is inside the image and creating sharpness without destroying like in this case for example booking or blur effect I'll provide link down below for the newest version of the gigapixels that was just recently released and thank you for watching this video let me know um what next video you want to be about thank you for your support your subscribe and sharing this video

Channel: Vladimir Chopine [GeekatPlay]

Views: 2,278

Keywords: Geekatplay Studio, Vladimir Chopine, Digital art, How to, Learning Photoshop, 3d tutorials, Digital compositing, How to do in Photoshop, How to do compositing, How to do in 3d, AI art, MidJourney, Stable Diffusion, Dreambooth, Dall-e, Free resources, Free learning, Digital art for begginers, Free tutorials, artificial intelligence, Digital Photography, Photography

Id: 1ts60WB-XDo

Length: 14min 10sec (850 seconds)

Published: Tue Sep 19 2023