Unlocking the Future: IPAdapter FaceID V2 Revealed!

welcome back to my Channel today I'm going to look at IP adapter face ID there's a new version released version two I haven't played with version one so I'm going to go straight into version two and let's start this off uh face ID IP adapter app face so we got this and that's going to go into K sampler K sampler there model there so now I need to get an IP adapter IP adapter get loader I ad model loader there we go and add this to this hopefully that's the right one I think it needs to be a bin so let's look for version two plus version two okay I adapter I need a clip vision clip Vision loader clip Vision loader there you go I'm going to connect that up to there and for clip Vision need model safe tenses then we need inside face inside face loader Cuda Nvidia graphics card the one I have if you don't have an Nvidia you can use CPU I'm not sure what RM is I've never used it before okay so we need the image image so load image image TR this one's actually not an image a web p and we actually need two images image how can we uh get two images image badge image image blend okay ah it's got two images so image one and this goes here copy this same image flip do a flip image flip connect this to this and this to this so we got that and we need a model and that would be Laura loader model load Laura La load mod on zoom out a bit put this down here and we going to stick this to the model and what we're going to use use IP adapter I there's the version two and I'm going to give it a strength of 65 that should do it blend Factor 2 is 5050 that's fine one is flipping to the other we're just going to add a preview to see that it is actually flipping okay so that's our apply IP adapter done now we're going to do case we need um positive and negative uh clip Tex and code prompt do the same again so one goes to positive one goes to negative um we need what do you need here need to load a checkpoint load we have checkpoint loader checkpoint loader there you go and I'm going to use realistic v51 V8 config no that's the wrong one check another simple there you [Music] go stick you the others you can you can use use those but this is the one I know that works for me we got the model I'm going to click Jo the clips together to there that's that one we need lat an image lat actually lat image there you go I'm going to change this to 1280 set uh 720 and that's um uh 16 by 9 I believe check that into there there we go that's that and on this side we need the latent to be decoded so load we need a VA loader that can be the one that everyone uses um and we need the vae decode okay so VA goes there goes there just going to do this image image goes to uh let's do a preview preview preview image done we got these two and for positive we want um let's see solar woman in a garden for high quality um police woman de light highly detailed photo of film photograph ultra realistic photo realistic Ultra photo okay and we got negative um work trying what else uh bad quality okay let's try this fingers crossed I don't know if it's going to work that's the preview of that being flipped okay that didn't work only okay so check I'm push this for to connect this let's try it again hopefully it works this time we can mess about with the details afterwards so what do we say uh solo a g looks like Garden May a preview image open that up that's uh what's up this up no it's not going to let me open that up uh actually what I can do is uh save image and load that image okay save image going see if this works keepy prompt save image that's another G let's load this anyway okay so my confy on Drive H uh output so that's the the one that saved let's load this one up there you go that's going to actually we don't need to save this anymore get rid of this open image does that look like this sort of but it's a bit artificial like so what we can do is pass this into a face detailer so before we do that we're going to upscale it add node comy R upscale upscale go image image upscale there to there I'm going to do ultra sharp two times one two 4 eight so let's try this FL should just up scare this image app sces finished this is already getting a bit um messy uh big so we got that upscaled uh what resolution is that 2560 by 1440 so uh standard resolution is 1280 by 720 and that was whatever that was okay so I'm going to add face detailer so we got the up scale image going into here the model where the model um there the model the model there got the clip going there and vae going there positive going there um where did hide the negative there's the negative no that should be posit negative we need the positive here okay so we need alrc detect provider going put the B box there so that's that one done um I'm going to do another preview here preview image and uh change these this value to 512 um actually let's do seed everywhere and we want to change this conver seed to input seed goes here need to change the seed to input as well SE to input I randomize start at one and we going randomize it um change this to DP DP MPP 3M sde and cross do the same in K sampler denies uh we're going to do that to say five okay CFG to three CFG to three and let's run this and see what that looks like that might have been a bad idea CFG on that one or do is actually do that should have been one generate new new image see what it looks like anyway yeah it's going to give it a better face uh RG three and we want the image compare I'm going to go from upscale and image here keep I wonder if it's possible to do three images we got that image we're not saving any images yet which is a good thing well if you got enough space it doesn't really matter that's the upscaled open image don't need to be that okay let's zoom in so we got the original upscale and details does it look like this person open image it does not actually does it well which one do we need to load up hold this one the same one different one I sort of okay so we got that now we can uh generate lots of different images from that one image and they all will look fairly similar say [Music] in C on a bch outside the building solo sitting on the bench iide building policewoman I guess that might be let's change this to police [Music] woman see that one first now is go to face detailer let's open this up you can see the cartoony style what you get then you got this one open image bit more realistic and we'll check the difference they look like Judge Dread uh police okay now to make it actual individual we're going to go to where's our latent change that to 512 and 512 and that should give us one person so i' got face ID taking both images here taking a face and putting sending it to case ala using these details and does a pretty good job and you can get consistent images thanks for watching and in the next video we'll do something else

Channel: Grafting Rayman

Views: 4,087

Keywords: comfyui, stable diffusion, realistic, ai, artificial intelligence, model stacking, stacking models, comfyroll, workflow, comfyui workflow, comfyui tutorial, ipadapter, faceid, face id, ipadapter faceid, faceid v2, face id v2, tutorial

Id: TEvVM6ekhHo

Length: 18min 3sec (1083 seconds)

Published: Sun Jan 07 2024