Use Any Face EASY in Stable Diffusion. Ipadapter Tutorial.

Key Takeaways

today we'll look at how to render images with a specific face without training a model or Aura we're going to be checking out IP adapter and this is not the regular one this is a new IP adapter face ID plus version 2 that's a mouthful but it's going to enable you to create your own images with a specific face just have a bunch of input faces or input images with faces and you can output images well with that face it's going to work both for table Fusion 1.5 and sdxl and uh sdxl turbo so let's dive into that we're going to show you how to do it in automatic 111 but the workflow is basically the same for comy we just going to download a couple of models and uh let me show you how it works oh and some news I replaced my rooster with a duck so now I wake up at the quack of dawn AI so this is what it's going to look like when it's finished we're going to be using a control net and you got to make sure here that you have one of the latest versions I'm currently on 1.1.44 uh you can quickly see that if you have the multi input here then you have one of the newer versions if you don't have the multi- input you're one of the older versions then you're going to have to go into extensions check for updates and then apply and restart UI if you don't have control install i' recommend you check one of my previous videos on how to install it but basically you can find it in the available list if you updated and still can't see the mul input or an update in your controll version here you need to close down your terminal and restart your automatic 1111 just make sure that also you have the IP adapter phas ID plus here as your pre-processor if you are on all here which is going to be default and if you press IP adapter you will be able to find it in the list you will be able to find it even if you are on all but it's just easier to press it than it's going to be only the IP adapters here we're going to be downloading some models these are not yet available so these are some that we are going to download so this link is going to be in the description below I also have a detailed text and image guide in my Patron if you prefer that route if not you can just follow along in the video here now we're going to download the ones named plus V2 we're going to have the bins and the s so you might be thinking wait Seb why do I need both of them can't I just use one isn't the Laura just if you don't want the big one well no in this case the Laura is an addition so you're going to use them together so there are four files in total so you're going to press download file find your stable diffusion folder go into models control net and save here and that's going to be for the bin files now for the Laura files you're going to go into your stable Fusion rout then you're going to find models and Laura and just drop them in here so that's going to be in total two bins in the control n folder and two luras in the Laura folder now let me load a completely clean Dil Fusion here now so I've loaded a 1.5 model for this one I'm going to put in man portrait then I'm going to go into my luras here and I'm going to select the newly downloaded IP adapter plus V2 sd15 so this is which you're going to use for the 1.5 model that's going to be inputed right there I'm going to raise the sampling steps a little for this and that is because later on in the control net down here we're going to be changing the starting control step and the ending control control step and if we have too low on the sampling steps we're going to have a hard time adjusting these values so it's not necessary but I find it a little easier just raising the sampling steps a little bit you can use tomb Caris I found that uh SD Caris were works very well too we're going to be using the Default Resolution here now for some of the models we are going to be changing the CFG especially for the sdxl and the Turbo for 1.5 it's fine to use various cfgs but just for ease of use we're going to set it to 1.5 because that's what we're going to use in sdxl and sdxl turbo now you can either load one image here or we can go into the multi-input which is what we're going to do and then we're going to upload our images I have selected four images of me one where I'm holding my beautiful little YouTube plaque here and three older pictures we're going to select IP adapter we're going to change the pre-processor to phase ID plus and then we're going to make sure that the model is the phase ID plus V2 sd15 because we are using a stable diffusion 1.5 model if you are not if you're using sdxl model you need to select phase ID plus V2 sdxl now you can see this here current ly because it's filtering for 1.5 models the control weight is how much your images will influence the output face and the starting and ending control step is when it's going to start and when it's going to end I usually find that starting this a little later is going to help with creating the base of the image first and then applying the face on top of it and I also found that if you end it a little earlier you might lose a little resemblance but the quality might go up just a little bit but test this on your own and see what works best for your image and your resolution I'm just going to load a style here this is not necessary these are my predefined Styles and they can be found in the patron if you would like to use them I'm going to set this to four images and we are running a man portrait in the Cyber panga Style with the Epic rism 1.5 model and we are using IP adapter phase id+ V2 and you can see here coming in live there is a resemblance to Yours Truly now it's not as strong as if you would have a trained dream Booth model you can clearly see that it's trying to resemble the face and here we have it in a cyber Punk style I just love that beautiful teal pink lighting we got going on here the good part about this is it's so very very easy to get started with it it requires no training of models you just input your images you can even use one image and you just just output images that well resemble a person that is really really cool now if we would do this with another checkpoint so let's change into sdxl well let's take a random turbo model here we're going to make sure to remove this Laura and we're going to add the IP adapter face ID plus V2 sdxl Laura you can increase the size here because most sdxl models work fine with 1024x 1024 the rest relieving default remember CFG scale set it at 1.5 I found that two Works quite okay too but going higher will really mess up your image here now here we need to make a change because we need to select an sdxl model and I'm clicking this and wait I can't see an sdxl here I'm removing this I'm doing like this no nothing happens what can I do if you refresh this with the button here you're going to be able to see all of the models again now we can select face ID plus V2 sdx app on top here you don't need to change the pre-processor that's same for both of them now in general SD XL turbo models require less sampling steps we're just going to run 30 because feel like changing so it's going to be okayish anyway just remember that you can run sdxl turbo on Lower steps and still get a quality result and as you can see well you didn't maybe you didn't have time to see but you can see here that first we have a different phas coming in and then we're slapping on top the phase ID that's because that we have our starting control step here at 0.2 now if your images don't come out looking like the person then you can change the control weight here so if you look at this example I would say they resemble me they don't look exactly like me but they resemble my facial features if we increase the control weight here let's put it to the max here so this say two we're going to make new generations here we will get closer to our original however the images will start to break bear in mind you can see we're getting some very weird faces here you need to be very careful with now I found that values between 1 and 1.5 is probably your best bet here and as we can see from the results here this is probably not someone that you want to meet in a dark alley at night now let's set control weight back to one and now let's select an sdxl model I'm just going to use a merge that I have but you can use whatever model you prefer we're going to leave the settings the same as for the sdxl turbo and regenerating four new images I found in my opinion that for sdxl Turbo and 1.5 the results are well pretty good I think sdxl turbo actually outperformed sdxl but uh your results May different from mine these were some tests that I did and um the results that that I got looking at the previews these looked fairly good but let's check as soon as it's finished here see here we have four images and these are from sdxl so the non-t turbo version and I would say yes they resemble me they aren't me I think for right now it's close enough I would make a recommendation I would probably use the sdxl turbo settings that are similar to what we have right now the 1024 resolution about 30 steps steps CFG of 1.5 control weight around one leave a little room here for the starting control step and the ending control step is it necessary though with the starting control step and any control step no it's not necessary and again your results May differ from mine but after this guide you should be well set up to start your own Journey with IP adapter face ID plus version two so I hope you learned something today thanks for watching as always have a good one see you yeah