Reposer = Consistent Stable Diffusion Generated Characters in ANY pose from 1 image!

a single face is all it takes to create a consistent posable character yes hello and welcome to More nerdy rodent geekery where today I present my latest comfy UI workflow which has both an imaginative and descriptive Name reposer by combining the IP adapter face model with an open pose control net this workflow can create a consistent character in any pose you like on along with of course prompt control to guide your Generations you can probably do much the same thing in automatic 1111 just not as a single workflow so let's take a look for example here I've got my original character she's sort of a 1970s style detective the image contains mostly face but with elements of the clothing included to such as that brown large collared leather jacket notice how here just by changing the background color everything else in the image has changed too but her face the whole character there is still consistent but what we do have is this more purple and green sort of cyber Punky athetic while it works best with a decent face image you can use any image you like in there just got part of a face looking left no problem as you can see there looking right full body half body painting and anime photo maybe even no face at all just get experimental and see what happens sure it's not perfect but it is very quick and easy no need to fine-tune a model or even create a character Laura as all you need is just that one input image if you've used comfy UI before setup is Trivial nothing complicated in here at all just load it up as normal download your models and off you go if not then do check out my installation and basic usage video guide which is also linked in this video description you'll then be aware that in comfy UI the image is the workflow so if you drag or load the reposer image from the a very comfy nerd web page there it is then you'll be halfway there this web page also has links to all the models you'll need in the res sources section at the top the workflows are in the table below and next to the workflows you'll find a description also linking particular tutorial videos as well as any information about models you may need to download and where they go also when you click refresh comy UI will update all the names in those loaders to match the ones on your own personal computer this means the names you see in these loader boxes may not match exactly as to the names you have on your own computer this is absolutely fine and not a cause for concern as the name of a file will not change the way it works for example you may have your own naming conventions for models and perhaps even custom locations too let's have a look over here as an avid comy UI user myself I found that organizing things into subdirectories helps me when searching particularly if there are lots and lots of options for example let's have a look at this control net pull down now if I want to find a model in there it is a little bit difficult isn't it all right but don't worry because here is one tip for you you might not have noticed but you can actually filter the results by typing stuff in so so here I go depth and there it will give me anything with depth in the title so now I can see as I've put everything into different directories exactly what is what these are all stable diffusion 1.5 those are all stable diffusion XL that's a TI adapter these are control auras thankfully I've colorcoded and labeled all of the loaders so whatever your names and conventions that you use at home it's going to be really easy to tell which model you need for witch loader just by looking at the description at the top okay let's check through them one by one to understand all of the requirements for this workflow first of course is your stable diffusion 1.5 checkpoint loader this is your choice of stable diffusion checkpoint and the loader says load checkpoint it will load checkpoint models from the models checkpoints directory of your comfy UI installation as mentioned you can see there I've got mine split up into stable diffusion 1.5 and sdxl directories now even with the IP adapter the model you use will influence your image generation for example if you have a model tailored towards photo realism there's a a photo realistic one but you want to create a cartoon character then you're still going to get a little bit of that realism coming through because that's what the model generates so you may want to choose a more cartoon oriented model instead if that's what you're trying to generate okay the next one we've got here is the load control net for stable diffusion 1.5 and that is the open pose model now if we have a look back at the directories again so we're still in comfy UI models the default directory where everything goes and here we've got one for control net and it's called control net once again I've got mine split up into stable diffusion 1.5 and sdxl models be sure to use an open pose model of your choice here and also remember comfy UI will update the names to match those on your own computer if you have your own naming and location standards things will look different I've tested that with both the standard and control Laura models and both of those seem to work fine personally I would suggest using a control Laura as it should be faster and use fewer resources all right next up is the clip Vision so we've got load clip Vision IP adapter stable diffusion 1.5 image encoder this will load clip Vision models from the models clip vision directory and as you can see once again I have different subd directories in there this is particularly useful as they seem to have called both of their models model so if you want to download both of those you are going to have to rename or choose a different directory you've got no choice there whatsoever okay now the PowerHouse behind this is this load IP adapter this is the stable diffusion 1.5 face model this one doesn't load things from the default location so we have to go back up one and then we're into custom nodes and we've got IP adapter come for UI and there we have a models directory with the IP adapter plus face for stable diffusion 1.5 those other models will work in here but the face model is what will really give you those face results now the final two there for upscaling so here there's a tile control net load control net stable diffusion 1.5 tile and also an upscale model there as well if we have a look back at the default location comfy UI models here we've got the upscale models so put your upscale models in there for the tile control net that's exactly the same location as we saw previously for the open pose models so that goes into the control net directory upscaling as mentioned is totally optional so if you don't want to do any of that you can just go over to the upscaling group and just click bypass nodes and then none of that will be run and you'll just get your one reposer image in there links for models can be found on the a very comy nerd web page in the resources section also there is a description column next to each of the workflows which has links and more information pertaining to the workflow to the left of it great now you've loaded your models it's usage time in its most basic form you simply need to drag an image onto the face box there it is load IP adapter image face and a pose into the pose box and click Q prompt to generate wait a couple of seconds and then you will get that face in that pose in the little save image box there while optional prompts can help maintain consistency too maybe give your character a name or specify clothing such as a suit or jacket things like that if you swap the face from Real to something a bit more cartoony then that face and style will also be transferred as you can see there it's coming out as a a more cartoony image switch the pose image and the character will take on that new pose instead you've also got some other controls at the top there you got the weights for the prompt so there's the prompt strength by default that should be at one IP adapter strength so how much does the face influence your image Again by default that's one and you've got another option here which is image or batch default mode is image so that's just going to use the one face in there but if you change that from image to batch then it's going to use both of those images in the IP adapter what happens in this case is that you basically then get a blend of the two so even though that's cartoony it's taken the realistic bit out of that image and sort of merged their faces together as well all right you can get some really interesting things out of that mode in the pose pre-processor you can disable or enable hand body and face detection should you wish I suggest just playing around with loads of different images to start with and you will soon get the hang of how things work more information and links are down in the video description though if you enjoyed this then why not watch another nerdy rodent video

Channel: Nerdy Rodent

Views: 45,610

Keywords: nerdy rodent, reposer, comfyui, comfyui tutorial, stable diffusion consistent character, stable diffusion any pose, stable diffusion pose character, Consistent character stable diffusion, Stable diffusion consistent character, Face swap, Easy face swap, Stable diffusion face swapper, Roop, Simple face swapper for stable diffusion, Face clone, Stable diffusion face swap, Consistent faces stable diffusion, Easy LoRA

Id: SacK9tMVNUA

Length: 11min 34sec (694 seconds)

Published: Thu Oct 12 2023