I Discovered An INSANE New Tool To Relight and Perfect Your Stable Diffusion Images! Check It Out

hello there engineers and welcome to this video for prompt geek let's start this video with a question have you ever had a situation where you've created an image using stable diffusion and you love everything about the subject the person looks amazing the composition is perfect but the lighting lets it down all the background is not what you wanted the worst thing you could do is have to redo it all in hope for a similar result while in today's video I'm going to show you a piece of software that will let you relight the image completely using AI I'm going to show you how to use it to make amazing AI generated images where you control the lighting so let's get into it the software I'm talking about is called switch light by bieber.ai and as you can see here in the example image you can fully re-light your photographs in a very convincing way but here's the thing we can all also do it with our AI Generations from stable diffusion and then we're going to actually take it back into stable diffusion afterwards to make it all feel a lot more grounded in the surrounding environment so how does switch light work switch light uses AI to turn your image into a very detailed normal map it also does something called Albedo which takes all of the shading and the lighting information out of the image and makes it look very flat then what it does is it's able to figure out how the light would interact with that face based on the normal map so a normal map is something that's not new technology and normal maps have been around for quite a long time in visual effects and gaming and in gaming normal maps are used to visually add more detail to a model without actually increasing the polygon count the best way to show you is to show you it actually with a 3D model as we can see here the normal is effectively the direction that surface is facing and then that normal is turned into a texture which the AI is able to figure out which direction that bit of your subject is facing and then calculate how the light would interact with it so if you sign up for switch light you actually get given a hundred free credits to use and then of course you can buy more as you see fit to relight an image is just one credit so you've got a hundred images to play with before you have to pay a single penny if you want to you can even relight video on here though it does cost 20 credits having looked up the results though it does look pretty worth it so let's have a look at this in action we'll start off by of course creating a beautiful waifu image in stable diffusion and my prompt here is photo of a young woman with brown hair wearing a dress harsh direct lighting okay it is much more photo real than anime way through I didn't want you getting too excited did I so they've got this image here great pose great subject but I don't like the background it's not what I was envisaging the answer to that would have been prom geek why didn't you actually prompt the background Well I obviously messed up there but I don't want to waste this image so let's get it over into switch light to fix that right now once you've created yourself an account at bible.ai and logged in to switch light you get this menu screen here where you can search through the different types of scene that you would like you're subject to be in you've got from Studio through to Neon so let's go with Studio first to have a quick look at what the options are now all of these are panoramic images but they are not just panoramic they are hdri this is a type of file which contains all of the lighting information for the area that the photo was taken in and from that information the AI is able to completely relight your image using the actual environment as the lighting source so we probably want to go with something which has quite a bit of lighting in there just for the purpose of this demonstration so let's go with Studio small one we select relight image it's going to load up the hdri then we just click here to import the image from stable diffusion that we just created it'll take a few moments to load up but watch the result and there we go look the background has completely been removed and you can see instantly that there is a difference a massive difference in the lighting on our subject like I said the way that it's figured this out is the AI has obviously removed our subject from the background created a normal map from our subject created this Albedo version where they've removed all of the lighting information and then applied the lighting from the hdri map on top of air you can export all of these layers individually so we could export the re-lit foreground image of her fully masked and then just put her over a new background but here's where it gets even more incredible over here on the left we can move move the hdri map and effectively move the lighting we do this we press apply and watch this transformation absolutely amazing look at that change there let's do it again with the light coming from the top left it casts shadows and everything it just looks brilliant now it is of course AI so we are getting some artifacts especially in this angle here but let's say that we really love this lighting effect so you can see there's a bit of noise in here now and it is definitely getting some artifacts in the textures so here's what we're going to do we're going to export just the relit foreground we tick that box and hit download so with that image now saved out from switch light we want to come back into stable diffusion we want to create our background image so I've switched over to rev animated for this because I want to get a science fiction kind of look and so I've just created this very quickly but we want it also to be blurry as well because it is going to be a background and Euler will work fine for this and we probably don't even need to high-res fix it because we are going to be blurring it out let's generate that and see what we get and we've got our image here it's maybe not the right kind of lighting in terms of the color but I've been this will do fine for the purpose of our example so let's just save this one we'll copy it and bring it into veggie shop so I've opened Photoshop you may be using Critter or another piece of software but the principles are the same I've got a new canvas here I'm going to paste in our background image resize that to fill the space and I'm just going to add a blur to it we don't need the detail it is just a background for our subject we're then going to bring our relit character into it now yes this definitely doesn't look like they fit together but the power of image to image is going to do that for us and we can also if we want to create a mask layer too say we can save the mask out separately if we think we might need it for some in painting but for now let's just export this image and now we're going to pop back into stable diffusion and put all of this together in the image to image tab we want to really bring over our original prompt there but we need to obviously add in here that it's a woman stood there with side lighting so I'll just type that in very quickly we're going to bring in our very basic composite image one make sure that we have the right sampling method sampling steps are up high you are going to need a lower denoising strength and you will need to experiment with this I'm sure it'll take me a few tries to get this right but let's just start off with 0.25 to get going and generate I ended up having to put the denoise instrument Pub to 0.5 and I kept the Original Seed number into the generation and the result pretty damn decent now I will need to impact her face a bit to make it a little bit more realistic but you can see that we've done incredibly well this was our original image and this is our new one with the re-lighting so really fantastic now if you had a particular face that you really wanted to have on the model you wanted the same face perhaps you could of course crop that original image down to the face and then bring that face into rupe in order to make the face the same but this definitely works as a way to be able to re-light your images and to have different backgrounds for your characters and subjects a really fantastic tool but it doesn't just do the hdri maps either I'll just quickly show you one last feature over on switch light we go back to the home you can also relight your images using other images and I'll just show you that right now let's go to Neon there you see here these are not hdri panoramic images they are just photographs but if I go here to re-light image I can drop in our very first generation here and look at that we've taken the color information and lighting information from that original photograph and we've been able to actually recreate that with our model here and actually the lighting on this one is probably more convincing to be honest with you again we've got the normal map the Albedo map and then the foreground as a separate layer that we can take and use in our Photoshop or other photo editing software now it isn't something that I can get to work right now so I think it's maybe a feature that's not been fully activated but it looks like you will be able to upload your own photo images here eventually in order to copy the lighting from your own pictures the same is also true of hdri maps which you can also upload your own panoramic images too so overall this is an incredibly powerful tool that lets you fully control the lighting of all of your images and get amazing results from stable diffusion controlling light has always been a bit of a task even using something like control net this tool makes it super easy and gives you a great level of control over it so make sure that you like this video And subscribe to the channel because if I am working on this amazing look book prompt guide for everyone so that's in the works at the moment and if you want to be the one of the first people to get hold of it make sure you like this video And subscribe to the channel and you'll know when it's been released so I'll see you all in the next one prompt engineers

Channel: PromptGeek

Views: 8,157

Keywords: ai art, stable diffusion, switch light, relight images, lighting effects, hdr, normal maps, ai lighting, ai tools, beeble ai, studio lighting, neon lighting, ai thumbnails, ai youtube, ai art tips, prompt engineering

Id: xzTgDvcSRXU

Length: 11min 18sec (678 seconds)

Published: Sun Jul 23 2023