Digital Painting & Spacial Frequencies. Nuke Tutorial

welcome everyone um it's quite quite a few people here so so thank you everyone for for coming and um and yeah this this this hopefully will be will be both interesting and and useful uh for me i i often often do things in compositing just because they're interesting but it really helps if they're actually useful too so hopefully this one will be both and um and feel free to kind of uh pop questions into the q a or into the chat i'll try to try to keep an eye on it on my other screen so so if i if it feels every now and again not keeping eye contact video is because i'm checking the screen so so yeah let's uh let's let's look at this so i want to share my screen press this button this button here and just check it on my screen that everything is working it looks all cool so groovy so this is um this is a talk that or a little tutorial that i did for for um the before and afters magazine uh in in december and um if you want i can drop this link into the into the chat as well actually see if i can do that and if that will be easy um yeah i'm struggling with it more than i thought um where are my meeting controls up here here they are uh q a and chat so i'll just i'll just drop drop them in here as well in case in case anyone wants to wants to check out that that that tutorial um but but basically what we are going to look at today is this concept of spatial frequencies and and how they could be could be helpful for compositing and what does it mean and and i will show you a few things that i didn't show in this this particular tutorial in here so so there will be some some additional things that we that that i didn't didn't cover in this um but anyways we'll get straight straight on to it just checking the chat as well yeah keeping other windows open in my in my other screen so um many many years ago uh this guy is called um adelson and his friends they adelson and anderson they they they wrote this paper about using image pyramids and and the idea is that if you scale an image down uh you should make an image smaller and smaller and smaller and smaller then basically every time you do this you are losing some details um so so if you were to then scale it back up you would see this image becomes kind of you know softer and softer and softer and as as you're losing detail but but what they thought is what would be interesting is if you looked at the differences between those images that they're scaling down and so what's what's actually left over and and uh what this gives you is um the details that were lost at every single stage when we when we did the scaling and then what we can actually do with those that is we can actually start applying them to our images for different purposes so a few years ago i read this paper and tried to build it myself and there are tons of um other tools already being built in wikipedia as well which i will show you in a bit um but let's let's kind of go through this through the kind of principles of this and what what do we mean when we talk about images and and and frequencies and and such so if you have a picture and i've got the same same picture of my our student here um looking gloomy um or intense either way um and um and um if you think of a every line in this image as some kind of a signal some kind of a wave so for example if i wanted to illustrate this if i just took one of these lines and just crop out let's say this line from here uh let's let's do this this track mark i just just crop out just this one line okay and um and then i stretch it out across the image then and if you just look at this brightness of this and if we imagine that that uh where it is darker it is further away from us where it is brighter it is let's say closer to us we could imagine that this is some kind of a curved curved surface so moving my um controls for zoom aside for the moment you can't see them but i can see them there's there's some controls on my screen um and anyways so if you go to look at this wave and this this sort of the the top view of this wave if you just view inside of nuke this tool called waveform monitor let's look at that we can kind of see what this wave might might sort of look like is uh you know there's this area in here which is a little bit brighter and then it is relatively smooth but not perfectly smooth you can see it's kind of noisy then we have sort of more detail and some other detail and again relatively smooth but not as smooth as this area over here and then i get more details and then another kind of brighter area in there so [Music] you can think of this now as like every image is actually a combination of these loads of loads of loads of waves there's another theory from a french mathematician called furia from at least two centuries ago that you can then decompose all signals into kind of individual waves um which is an interesting theory to follow uh in a different lesson but um but for my purposes this is kind of you know good enough and and then what we might want to do is apply another concept on top of this which is called filtering so filtering in the simplest form is called blurring so how blurring works or filtering is that you take some pixels nearby and you add them together and you get the average value and you put that in the middle so if i do this over here because these pixels are kind of similar to each other already the average value will be the same whereas over here because some pixels are kind of really bright some pixels are dark if average them together they will all kind of be gray so if i actually do it and and then basically what matters as well is is how big is this area that i'm averaging over so the more values average together the more average the result will be um so let's look at this if i do a blur and use a classic gaussian blur in the given case and if i make my image smoother you can see that the signal becomes smoother as well what's actually happening here is i'm using a 60 pixel wide kernel so it is roughly roughly this big actually so about 120 either side so it's about this big and salvaging all the values together in this in this region in here and then it puts that to the to the average value for that and you can see because um i have some you know bright and dark values in here but overall it's kind of bright i kind of i'm kind of left with this bright value um if this kernel was bigger so if i included more from the sides i would imagine that overall this middle part in here which used to be bright that would then become darker so if i if i make this bigger you can see that eventually everything becomes kind of more even more gray okay very fascinating so but important thing is that we can think of images as as waves because what's happening underneath is we have pixels of different values and when i average those pixels together bigger and bigger areas of these pixels together the the value just you know they average out and the signal becomes smoother so with this signal processing theory if you apply to two dimensions then i'm averaging these pixels in both directions and again this is just blurring this is nothing new to anyone who's using nuke or any other software so if you just blur things my image becomes softer happy days wow a softer looking image and just to illustrate this with another tool this is the slice tool from newpedia which gives you a slice over over you know your cert chosen range so for example if over here i have this point one and point two uh where i put them it kind of samples this area on this line between them so you can see i have some darker values then some brighter values and then some darker indent again so darker values brighter values brighter values so that's that's that's the values on the on this line over here and again if i if i blur this a little bit you can see the signal becomes smoother and less smooth so so how is that helpful excellent question so what we can do by doing these techniques is we can do what anderson and um and adelson did in their paper as well as we can look at what actually happens between these stages so anderson and the others and what they did is they made the image smaller and smaller and smaller instead of blurring it and you could do that but blurring also also works relatively well and it's easier so if i blur this image by a certain amount like let's say that much you can see the image has become smoother and i've lost a lot of the details from the skin so before and after i'm just kind of checking how much how much of this is actually visible through through zoom at the moment as well so i'll blur it a little bit more so before and after so image becomes smoother had lots of details now it is smoother let's look at some other region in here as well so for example if i look at this dot we can see overall relatively smooth region relatively uniform color with some noise and details and then there's this darker bit so that's the darker bit that's the that's the dropped over there and then again relatively smooth area after that if i blur this then eventually i can i might be able to blur this so much that this this dot might actually more or less kind of average out and kind of almost disappear completely but i still have some information in here i still have more general uh brightness information and and this is in these lower frequencies so what you can do similar to what adderson and adelson did anderson anderson is look at what's the difference between those images so what is the difference between this so if i look at this image and i take this value from the original i take the average value from the original what i'm now going to get is values that go below and above zero you can't see them mostly because they go kind of tend to go below zero then they're very very small but if i gain up a little bit then you can see they're actually there the other way of visualizing this is if you just add a little bit to this like 0.5 or something like that then then you can also see this this is this is the detail that we have taken out so the values that went um went below this skin color there so this is the values that we've taken out and um and the interesting thing is that now because i've taken them out i can simply put them back as well so in here i do have from i do a minus so whatever this value is uh i take it from whatever the original value was and we'll see what the difference is because the original value is often are smaller than these average values then then we get some negative values as the result but if i put them back with a plus i can get back the original result so yay we have managed to achieve nothing absolutely no difference between the original image and the new one so why is that any good because originally we have our image in a single pipe but now we actually have our image in this in these two pipes in here we have the image as low frequency detail the blur image and i have the difference between the blurry image and original image and that means i can edit those images separately without affecting each other so for example if i was to do a quick clone paint job in here it's very difficult for me to look at so i'll just look at my final result that i would normally look over here and if i just did something let it say a clone paint from here to here maybe from here to here and i just do that looks relatively okay not the best job but it's okay and um this is what i was actually painting i was painting some of these details from here to here so i paint some of those states from here to you know painting this kind of skin texture from one place to another and by separating it out in this fashion i didn't do any paint work on this softer blurrier image and this is important because if i had done it on both of them if i had just painted my original image i could have can subtle changes in the zoom but what you get by doing this is you're actually copying across some of the um values of the darkness of the oval image that are stored in these lower frequencies so in general this frequency separation allows you to separate the texture of objects kind of from the lighting of them it doesn't work fully but but that's the underlying sort of value of it so for example if i look over here and if we do a another blur and the from let's do a blur and merge in the from and actually start blurring so we'll actually get some output you can see this is the texture that we are removing from this image and this is this is the softer softer looking image without this texture so it becomes softer before after before after so for example if i if i blur it enough to to remove some of this um so that i don't see um the the moustache starting to grow so just enough like that and then i again extract this texture i look at this and this is the skin texture now so that's the texture of the screen separated from from the from the lighting of the skin so you can think of this in this way is that this contains now the lighting and the color and this contains the pure texture the all the wrinkles and that that's aspect of this game so if i then wanted to put them back together i can just with another merge and the plus inside so i plus them back together i arrive at my original result again no difference so seems like pointless but it isn't because what we have done is we have again turned one single image into two separate streams of the same image so we have a low frequency stream that contains the color and the lighting and the high frequency stream that contains the texture so [Music] if i paint this texture and just the texture again we can get hopefully relatively realistic looking results without accidentally copying across any shadows if i do the same paint but just on the original image you can see we have painted a lot of this redness from up there to down here as well so this patch up here used to be red so by doing this clone painting like that from there to there we have actually painted across there the red color as well but if you separate your image by blurring it into this is the color and the lighting and this is just the details we can now paint the details on their own so this is cool okay and and there are lots of lots of tools that allow you to actually do that in in silhouette um this is now built into into silhouette as well um in photoshop it's called high pass filtering um there's there's a bunch of gizmos on wikipedia that allow it to separate these gizmos but underneath what's actually happening is they're doing a a blur and a from cool now just to show you this blur and the from itself as a concept is um is something that already exists inside of nuke and in some other softwares as well so you can think of this as here it's only the low frequencies that pass you can drop another slice tool in here let's do another slice store so let me draw a slice over this area here so in this area we are letting only low frequencies pass just putting it right across this dot over there as well whereas in this area we are letting only the very high frequencies pass that's the difference between these low frequencies and the regional image in the original image we have both we have the low frequencies and the high frequencies so this is the low frequencies this is the high frequencies sorry these the low frequencies this is the low and high frequencies and this is just the high frequency so the high frequency are there going above and below zero containing just the detail information where is the skin darker brighter than the or like the wrinkles darker brighter than the overall skin if you wanted to make your image look sharper then when you put this image back together what you can do is you can amplify these high frequencies you can make them stronger and that will make the image look sharper you still have the same amount of low frequencies the brightness and darkness of the image wouldn't change but if if you wanted to amplify the state that you could make your image look sharp so when we put this image back together let's do a another multiplying here so another um merges to a plus so again we reach back our original image so you can see if it's the original image no difference there but if we plus it one more time if we do it one more time we add this image one more time to itself so we add the high frequencies to it again so we take these high frequencies this detail here these values and add them to itself keeping everything connected what you'll get is a sharpening effect again you can see here probably not a lot of this comes through zoom because you're getting compression on top of this but we're getting some artifacts in here we're getting this darkening around edges but these are levels of details here they are now looking sharp so you can see we can sharpen our image and basically what we're doing is if you look at the slice tool again is this small details in here we make the small details more prominent simply by adding them on twice it's like i have the original small details and then i add them on again and i have the small details are not twice as strong and this is exactly the effect that the sharpen node in nuke does so if you use the sharpen node you will get exactly the same effect if we were to use the same size as well let's do nine so identical okay so what's happening here is that we as humans we do it in our eyes as well we kind of look at we consider edges to be more you know more important than the than the the overall smoothness of objects uh just because we get more details out of this more information so our eyes tend to amplify this detail as well so when we when we amplify it artificially in the images then that makes makes our image you know images look sharper cool so just another kind of fun thing to do with with this tool okay so simply once you've extracted your high frequencies if you add them on again you get a sharpen looking image right so this kind of covers sort of more most of the theory that we and a little bit more than than what we what i discussed in this article um for that magazine um so we take the singles and we and we add them together but i'd like to show you a few more practical examples of this so let's look at this cinderblock wall so depending on how we do it do with diameter i'll build some of it to to show you how it works and then i'll show you the show you the already built version so another node inside nuke called laplacian is a fancy sounding name for something that we have already seen so applausion is a blur and the from so laplacian gives you the output of if you were to blur an image by this amount and then take it from the original what would be left over um just to prove it to you here is my image and i do blur by let's say 20 nice and soft and we do a from merge and from and we get this result and if i was to do a laplacian with a size of 20. we will get exactly the same result so laplacian basically that's the detail that's left over when we blur an image so it's just a quicker way to get this thing but underneath what it is actually doing is exploring an image and it's taking it from the original image so in a similar fashion because this contains the details of the image if i was to remove these details from the image i would be left so if i add them to the image i'm getting a sharpening effect whereas if i take them from the image i'm getting a blur blurring effect so laplacian and the from is basically a very convoluted way to do a blur if you do a blur on the front you're doing a laplacian if you're doing a plush and the flow you're doing a blur so get the same effect it's just this might look look fancier and more clever in your script until you you know show it to people who actually understand that it's the same thing so you can do it to the blur and the from doesn't matter so i'm going to replace it here as well we'll do it with the blur and the from just a blur and for this shot we might want to kind of replace this sign from the wall so we have somehow stabilized our image um so we've got our our image nicely stabilized and i'm going to drag my zoom controls around a bit um so image is nicely stabilized but what you should see is that um there is some lighting lighting changes happening in here but uh effectively a lot would kind of like this in order to get rid of this this patch on the wall i want to take some of this wall from underneath over there now let's see what would happen if i just tried to do that so if i just go somewhere here find my frame and if i just took those those um those things from down there let's do nothing so just try to take these bricks and move them up there should work right you know it's just just a very basic lamp i just want to take those bricks and move them up there so i can try doing that i can take them and start moving it further up but as i move them further up it feels like they keep getting darker and darker and darker and that's really annoying by the time they get up there it's like wow that's really really dark they they they definitely did not look that dark when they were down there so i moved them down there it's like uh no okay you know no they're not really that dark so so why are they getting so dark when they get up there um and you know they're not because it's really that the color itself hasn't changed i'm just moving it up and down it's just compared to the background this general area this patch is actually it actually is that much darker it's it's just that our eyes um compensate for this small gradient difference so we don't see the darkness difference but if i actually move it up it's like oh my god it's it's almost darker like uh it's so hard to believe if i wasn't doing it myself so what's happening there is um it's called a constancy effect where if you have something over a bright background it looks darker whereas if it is over a a bright you know dark background then it looks brighter and and and this patch itself is not changing color this this this um this gray ball is is is exactly the same same value of gray so if i zoom into it it says like it's that color 0.3 if i move it over here it's still exactly the same color 0.3 it's still the same amount of gray it's just just depending on what's behind it it looks like it's it's really dark but no it's not it's it's just just normal grades 0.3 um so whereas you know that's that if you move it over there it's suddenly suddenly suddenly feels feels like it's really bright so so yeah so what what can we do in order to be able to to still kind of quickly quickly paint this out and and move these things things up there this only works on this particular frame in here because it's not stabilized but if i wanted to just put this thing over here and get it to work surely this should be easier so what we need to kind of consider there is that we we might need to break this image then again into the details and into the lighting effect so how do we do this well basically we can look at this image and be thinking about what is it that we have to recreate so if i let's say i gain down a little bit so make it a bit more visible for you guys you can see there is there is um the the cinder block uh bricks and the like like the their edges so that's actually what i want to kind of recreate and another thing that i want to recreate is this kind of general smoothness of this of this lightning lighting going across the screen so if i blur my image just enough that i don't need to blur it is so much that this thing disappears although that's usually a good approach if you're doing some kind of a marker removal so if you're doing a marker removal and um like in this kind here if you simply blur it enough that the marker itself disappears then you've separated where the marker is and where is the lighting effect of the skin so if i do this you can see that the marker is in this in this details in here and the lighting and color effect is in here so i only need to so as long as i don't touch this color and texture effect then i'm fine to just work with this texture effect in here and just paint that so as long as i paint only this paint some texture in here i should be okay again if you were to compare this with if i was to paint from here to there it's like if i have the circles over you know then it looks like it's you know it's the same color inside of the circles but then you they they take it away like wow it's it this this thing is so much darker than the surrounding areas like i i don't believe that it is that much darker until i actually put it over there so um whereas here it's like you know if i blur it enough that this marker itself disappears from my lighting and color information then it is extracted only to this detail information so if i just fix the details i will have fixed my fixed my whole image as long as it don't touch my color now that's a good approach but sometimes the element that you want to remove is huge like in this case it's it's massive so when you think about cleanup you actually have to think about not what we are trying to remove but what we are trying to recreate so what i'm trying to recreate in this case is the wall and uh and and the thin lines on that wall so if i blur my image until those thin lines disappear then it means i have somewhere extracted i can somehow extract those thin lines and and keep them separate so over here if i don't know how to emerge in the from you can see in here i have extracted the texture of the wall the lines without the lighting so this means over here i can now do some quick paint work maybe i'll just do a similar kind of rotoshape that i did earlier over here just take the exact same one i actually can't take use the exact same one because it's broken because that one wasn't done on a stable image but if i just try to kind of roughly follow these edges in here it's always a good habit to to try to follow edges in that you already have in your image and maybe if i just take a whole bunch like that let's hope this is enough something like this q and we quickly pre-mult to take this patch give it a bit of corner pin to help with any fine fine adjustments of perspective just put these two to where i want them and and we just merge this over its original image in here and i'm just going to move them higher so in this in this high frequency image i'm just going to take it up there and try to try to cover my old old image that i had up there as much as i can i might need a slightly bigger patch i think over there but i'm just going to adjust it to place just so that this these lines would more or less line up with uh existing lines on the on the wall and i'm just going to make this the shape itself bigger to include a little bit more something like that so i've taken this patch from down there and moved it up there but obviously for now i have fixed the kind of the the high frequency details this is the this is the small details but if i'm putting back to the original image like we did in last time if i don't touch the blurry version if i just just add it to the original and let's do a plus then it doesn't work it's because you know i still have the the old image you know the old the blurry version still contains um the sign that i want to remove at least some of it it could it's it's here in this high frequencies but it's also here in the low frequencies so if you can blur it enough so it actually disappears great so which is why good vfx supervisors whether on set try to make tracking markers that are small because track marks that are small can be nicely blurred out so in this case we don't have to treat this blur image but in in a case like this it is there in the high frequency but it's also there in the low frequencies there in the you know blurry image too so what you just need to do is you just need to make sure that you you you clean this as well um luckily because this is a blurry image i can be quite um careless uh when painting this out so i can paint this um with fairly big brush strokes i'll still try to do some low opacity on this and i'll try to make it happen on all frames so i'll just quickly try to start you know painting this out adding some general color from its surroundings onto it and i don't mind if my result here is a little bit blurry because this is the blurry version of the image it's already blurry i don't mind losing some some you know detail in here because it's it's fine so i just want to kind of remove this thing from the middle soft and soft it's often soft and soft and soften you can sing your softening song while you're doing this if you have one i i do like annoying my students with my with my paint songs every now and again so just keep painting until until this area is nice and smooth and and we don't have what looks like the the unevenness that we had earlier so just try to remove a little bit more of that okay and this might just be enough for me for now kind of kind of happy with that getting some brighter color in there and then i'll just also soften it because it's already soft anyway i'll soften it a lot this should also happen on all frames i'll just blur it so i'm hoping to get kind of a fairly fairly even even gradient of out of this okay something like this so and you could you could even do even the switches just just take the take the whole area and just just blur it a little bit more um but then mask masking blur only to do to one one area in here so i just want to kind of blur that very end okay so now i've added this details onto it i'm getting a wall without the bad you know the design underneath and this this is giving me the details of the wall so again just kind of compare if i had done this paintwork just on its own so if i just take the paint and you know what we did earlier is if i just take this patch and i move it up there that's what it will look like and um it's crazy it looks it looks so so different and so bad whereas if i if i move it like that then you know it's the same texture but we treat it that the the smooth areas separately as well and so this is you know this here contains the texture and the original lighting from underneath this gradient from there and it doesn't work at all it doesn't match at all here it still contains the same textures and the same details but uh the smoother and the lighting we have painted out ourselves and it's a totally totally separate separate looking image and it's so much more better quality so in the end you just kind of need to make sure that you you capture just the part that you actually need to put on top of your image and you should have a reasonable result not going to cache all of it but maybe just enough to to show that after stabilizing and unstabilizing you get back to your original image so here i'm i'm not showing how i did the tracking for this um or how i did the stabilizing and unstabilizing but but what we are looking at is that this technique works and allows us to to extract these high frequency data from the low frequency details and treat them separately [Music] let's see if this plays back seems okay i can still see a little bit of a shadow in the of what used to be the the the sign but then again normally competex takes at least half a day this was just 10 minutes so for a 10 minute job i think this is this okay compared to this job over here which is not okay so you can clearly see there's a difference to these two techniques cool so yes so anderson and um adelson they did not sort of fully show show those ideas they did kind of um try extracting different techniques and um you know these different different frequencies on different levels and um and combining them um in kind of more automatic ways in order to yeah extract details that you can see a little bit more in some images than in others one thing that they did try which was interesting is um rather than blending together two images very softly or with a very hard edge they thought like what if you could blend images with the hard edge where there is texture but soft edge for the lighting um but they also say that actually it doesn't quite work that well um it only works if these images line up really really well so i tried this as well so first of all this is their concept of how you could um separate images into these smaller frequencies if you just were to scale an image so if you just make an image you just make an image smaller and smaller and smaller and smaller then and try to find the differences in in every single step um then then it's it's possible to do that yes you can get the difference in every step as well to get the details in every single level um and this is then if you if you did it as this whole kind of pyramid um and uh you can try to combine this you know trick that they said which is that rather than combining things with you know just a hard age and or a very very soft age you you combine it on sort of multiple levels you start with a little bit of softness a little bit of hardness a little bit of and you and you keep keep adding to it until until hopefully you will get something that has a smoother transition um so it's like it's hard transition where it is a hard edge and the smooth transition reality smooth edge doesn't quite work although built upon this idea is photoshop's healing brush except the difference is that um the healing brush also looks at um you know the the shape in here that we need for the for the hard edges is a very you know it has no relevance to the actual texture that's happening underneath the image so but but you know it exists so so they said this this exists and i tried it and they said they didn't work and i said and i tried it and this i agree with them it doesn't quite work um but in my um the conclusion of my sort of uh talking here where my tutorial i sort of said there's uh there's also potential other techniques that you could use um which uh soften your image better than just doing a caution softening and what do i mean by that so and this is something that i i occasionally come come across in some of the jobs as well and this is a technique that some of my past students who are working on a film now they were using on that film and that actually worked worked quite well for them so this blurring and finding the difference technique kind of works if your image is like what you're trying to remove is is quite separate from its background so if this this dot that i'm trying to remove from here um is uh is on a fairly even background then um then i can blur it until until this stop you know more or less disappears but um in other areas like if i wanted to remove this um this mustache then over here again i could blur this until until until it's you know it disappears and then effectively i have extracted it in this other area but but this might get a little bit risky when we get to the edge of the nose for example so where you have the edge of the nose you can see that we're blurring the edge of the nose as well so if i'm going to try some treatments on this and and i try to do some paint work in those areas in here so i'm just going to again separate it into my detail information and into my lighting and color information and then i will put them back together let's put them back together so if i put them back together you can see um it looks all good but then if i try to do some paintwork in here um i can try to you know copy this texture from here to here and that seems to work and from here to here as well that that seems to work um that's all fine here too copying the skin texture under there but if i get too close to the nose what you see is that it you know we are actually blurring the nose because what's actually happening in here is that i'm copying across potentially some texture where there is meant to be this sharp edge there is meant to be a sharp edge that tells me that this is the where the nose ends and where this lip begins so again if i if i try to do some paint work in here let's do that that's working for the for the skin but uh but when i get too close to the nose you know that that's going to what what's happening underneath is that uh i'm actually painting painting over this where i needed to have the detail of the nose so i i don't want to remove that detail of the nose because what i'm you know basically doing here is i'm saying that's just a normal lip texture but i'm painting over this so this is the kind of final thing that i wanted to show you and this comes from another great book by this guy called zelisky richard and it's called computer vision algorithms and applications and this guy very generously has made the online version of this book uh free free to the world so you can um you can find it on his website there's an online version of his book normally costs 50 or 60 pounds or something because it's a proper 1 000 page textbook and one thing that he describes um and you know it's not just him i just found a good illustration in his book um is um something from duran and dorsey it's called bilateral filtering and the idea is that you only filter when your pixels are already similar so i only filter colors that are similar to each other and then there's a sharp boundary then i don't blur across the sharp boundary this is expensive because this means that every pixel is getting a slightly different amount of blurring so it's only in the very recent versions of nuke where this became available because you need a gpu to to use that technique but there is a bilateral filter node so let's do a bilateral and the way the bilateral filter works is is it's still a blur node you're just blurring things and it initially seems to do nothing but what you're kind of telling is that let's blur across a certain visible range or a certain range in distance but not across a certain range in in color difference so here i can kind of choose make the color difference small and say that it's only these similar colors because this uh you know dark stubble there is similar to the dark skin nearby but this difference nearby is big so this is a you know dark area and that's a bright area pixels within the star carrier they similar to each other so what you're actually doing is you're saying when there's a big difference then then let's let's not in in colors then let's not paint across that big difference so so this actually helps you to preserve the edge um so when you do your same paint work like we did last time and you can still extract the details from this so if i now do the same from and plus so if i do a from what i'm now getting is effectively the kind of pure skin details and much less of the nose left in there so the nose edges is is like it's it's we're not blurring across that so we're not losing this this detail in this end and not extracting it so if i paint over this with exactly the same paint that i had earlier you can see one of them affects the nose and the other one doesn't let's view them with a with a wipe so we have blurry nose and knows where the sharpness has been preserved because the bilateral filter has never blurred across the edge so when i was painting my high frequencies i was never painting anything into that area because there was there was never anything there so so this this edge is already in this image in the kind of blur image so i haven't taken out this stage from a blur image it's whereas if i do the usual gaussian blur the gaussian blur blurs across that edge if i use bilateral filter bilateral filters this says that okay this is a relatively even area so what i'm going to do is i'm going to say that when there's only you know when there's a sharp difference in colors then then stop blurring if there isn't a difference then then continue to blur within so any difference is smaller than that we will continue to blur any differences bigger than that we will not blur across that edge and the position sigma that just tells us kind of how much do we blur in those areas so it's a nice tool if you wanted to build your own d-grain tools and stuff like that as well that might reserve hair detail and stuff um can be can be very helpful so this was the last tip that i wanted to kind of give to people if building these things is um i mean now you should see that building these things is easy you blur and the from and the blur and the plus that's it or if you want to take it more advanced bilateral blur and from bilateral blurring plus but only do this if you really need to because it's a heavy node that needs a good gpu but these tools are already available many people have put them together for example there's a frequency separator inside wikipedia there's a standard deviation frequency separation there is a sharpening tool which actually is sharpening multiple again levels high medium low ultra low effectively just describing the size of the frequency and there's wavelet which is quite nice um from what's high portland which is the starting of a bilateral filter so he he built this one before the one in nuke came out which is nice and also one of escape spa students artila was my storage system for a while as well he put together a nice tool that effectively underneath is allowing you to to do a transfer mask kind of but uh but only for the for the frequencies keeping the colors separate so so again you can kind of uh do things like wrinkle removal just you know putting some skin texture from here to here but not changing the color so the color remains as it as it was so a lot of these these tools are available and what i would like to conclude with is just this shot in here just because it's fun so now that you've you've seen how how this kind of different techniques work you can put them together for artistic effects so where it says escape in high frequencies it might say except in low frequencies escape in high frequencies except in low frequencies magical so now you know the techniques that we kind of saw in this nice webinar you can try playing with putting images like like this together yourself there's there's other ones in the internet that have like marilyn monroe and albert einstein when you look at it close enough it's advertised if you look if you're further further away it's marilyn monroe so now you know how how these things work so i will conclude with that

Channel: Escape Studios

Views: 1,444

Keywords: Escape Studios, Escape, studios, VFX, Visual Effects, CGI, Maya, Autodesk, Nuke, Foundry, I want to learn, tutorial, modelling, rendering, dynamics, texturing, animation, renderman, compositing, ncloth, motion graphics, design

Id: dw7SAUtYMVY

Length: 59min 55sec (3595 seconds)

Published: Thu May 06 2021