PERFECT FACE - NO Lora or Extension - for A1111 and ComfyUI

how to get perfect faces in AI images hello my friends how are you doing today I want to show you a method and I'm a little bit proud to say that that I developed for a friend now I do photography also myself and one thing I can guarantee you is that the model is never going to be happy with the AI face because there are subtle differences it just doesn't look completely like them but also there's another thing that is important for artistic purposes AI simply cannot create an expressive face and cannot create these subtleties that a human face has in the expression so for example things like and and stuff like that with an AI you can forget about that so with this method well I'm not creating the face I'm keeping the face and I have created this for automatic 1111 and then a much simpler method for comi let's check that out so first let's have a little look at the image so so here we have a beautiful face it's not abstracted but you can see it has a very beautiful expressive smile this little subtleties in there just amazing okay now here is one thing down here we have this little fashion detail which is very beautiful but it is not so great for AI rendering so you might want to paint that out because it can hinder the automatic face detection a little bit and create a little bit of an edge in the image afterwards also what you want to keep in mind is that AI still is not great with complex hand postures so maybe find a posture where you have no hands or the hand postures are simple enough so thei can recreate that so as you can see here in the second version I have remove that little fashion accessoire here within painting in Affinity photo very simple so now that we have prepared our image been going to go over to automatic 1111 first of course you write your prompt and your negative prompt your describe you want to have in the scene and we're going to load the image into in painting that is very important now you want to mask the face maybe even a little bit around the face but stick closely to the face very important Next Step scroll down here you want to turn on for the mask mode inpaint not mask this means it will inpaint everything but the mask then of course mask content original in paint a only mask select your sample method and the step count that is perfect for the model you want to use down here I'm setting the same resolution that I have for the photo I'm not using a high resolution for a photo because that can actually break your AI rendering there might not be enough run to actually do that so I have here a smaller resolution of 800 by 1,200 but you can upscale afterwards no problem now here for the D noise strength I'm going very high within this cas Cas 0.89 you can go even higher experiment with that to give me a better result for my image I'm using two control Nets now the first control net I'm using is open pose and I'm using it with a control net weight of one in this case and then the second one is going to be depth and I'm going to using that with only 0.35 you might want to experiment with that depending on how complex the body pose is but also how much you want to cover or bring over from the other photo then we are off to the rendering and you might want to create multiple versions to see what is the best and play a little bit around with the settings now in the first step there is a little bit of a caveat and that is that in automatic 1111 the skin color of the person often does not match the exact color of the skin in the face of the original image now for that I have a second step for you so you want to click down here on this pallet icon to send it again to in painting now what we're doing in here now in the second step is we are going to mask only the eyes why only the eyes well the reason for that is human eyes are very expressive they are the gaug to the soul and they have very subtle details you cannot create that in a nice way that is feeling human in AI so you you want to mask out the eyes so you keep the original eyes still and then we have the same settings down here but here is the big difference we are going to set the D noise strength to 0.01 the lowest possible value and what this is doing as you can see is that it's rendering the exact same image again but the color of the face now is closer to the body color so here we have the image from before you can see here we have a red edge and a red face and and this is the new version where we have a face that is closer to our body color now you can try to set the D noise a little bit higher you can also rerender that multiple times to get different colorations of that face to see where it perfectly matches for you there's a little bit of experimentation involved of course as a last step you want to send this over to your upscaling and then you want to set here your upscaler of your choice and you can even use gfp and code forer but use it very slightly so it doesn't change the face too much and then you simply upscale it to any size you desire now my friends that was an automatic 1111 as you can see it's a little bit cumbersome you have to chump between different settings and between different tabs so how can we actually automate everything of that you can do that in comu ey and of course my Patron supporters and my YouTube membership supporters get that workflow tonight as a free download let's check that out all right so here we are in com VII you can see I beautifully prepared everything for you in different areas so you know exactly what is going on let's start here on the left you can see here we are loading the image and then I'm applying two different control Nets again the depth map and the open post we have here a preview for both and in here in this area for apply control net you can see that you can adjust here the weight or the strength that each of these control Nets has and I actually will rename these so you know which is depth and which is the open posst now this is then going into our prompting over here so here you set the positive and the negative prompt you're choosing in here your model and your vae and then this is going here into the next part and here I'm using f face detailer but I'm not using it to detail the face to make it better I'm only using the masking function so for that I'm tracking the face automatically and then we're going to invert the mask in here and down here you can see a preview where we have only the face tracked beautiful now this goes into two case Samplers in the first case sampler we are rendering the image as you can see here we're using a high DS of Z .89 this is also using both the control Nets as you have seen before and this is the version we get now here you can see inside of comi the face color of the face and the body both match perfectly for some reason here it works right away and that is really nice so at that point you might wonder why am I doing a second case sampler here well it is called face color merge so it is still adapting the face color a little bit to the rest of the body but then also we are rendering over everything again to get a little bit of a better detail in this case I'm using a d noise of 0.14 here on the right side we have a preview of this result but of course we are not done yet because we want to have the best quality so over here everything is automatically going into an upscaler you can of course choose your upscaling method in my case I'm using the 8X n mkd super scale 150,000 I'm doing a little bit of image sharpening after the upscaling and down here you have the beautiful finished result and you can see it is very nice high quality high resolution just fantastic and here you can see both faces side by side as you can see it's the same face beautiful quality and it looks super human because it is the real pH I really love this method if you're one of my YouTube supporters check out the membership tab on my YouTube channel and if you're a patreon supporter I'm going to mail out to you and also you have an exclusive post for you where you can download my comi workflow and also some extra information about the models and UPS scalers I'm using thanks for watching my friends leave a like if you enjoyed this content and see you soon bye oh you're still here so uh This is the End screen there's other stuff you can watch like this or that really cool and yeah I hope I see you soon uh leave a like if you haven't yet and well um yeah

Channel: Olivio Sarikas

Views: 27,377

Keywords: affinity photo layers, affinity photo adjustment layers, affinity photo learning, oliviosarikas, olivio sarikas, olivio affinity, olivio tutorials, ai art, stable diffusion, stable diffusion tutorial, stable diffusion ai, comfyui tutorial, face ai, face a1111, face comfyui, a1111 tutorial, face tracking, better face, perfect face, ai face render, no lora, lora, roop, reactor, best face method, realisitc face

Id: 9UOdXLYwhYY

Length: 9min 35sec (575 seconds)

Published: Tue Oct 24 2023