Stable Diffusion - Face + Pose + Clothing - NO training required!

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
generating AI images is fun but if only there was a way to mix a face a pose and an outfit into a single image all without having to make a fine-tuned model or any lauras the initial reposer workflow did quite a good job of that here the character has been transferred into a new pose using only a few basic negative prompts and those are mostly to help keep things YouTube safe the face the outfit the style you can see them all there in the newly generated image however to change the character's outfit would mean updating your prompts here I've asked for a blue t-shirt another way was to alter colors in the original face image like in this example here I've also used prompting to create her as a photorealistic character rather than the original graphic novel style she was created in want to take it that one step further and make costumes or clothing easier to apply yes hello and welcome to More nerdy rodent geeker where today I present the reposer plus workflow this updated version now comes with a supporting image which you can use to transfer clothing exactly like you see here face plus pose plus outfit no prompting no having to make a Laura no making a custom model just pick your images and create of course you can still guide your images through prompting it's all up to you as I've already been through the setup for this workflow in the original reposer video I won't bore you and go through all that again here instead this video is going to focus purely on usage which things work which things don't options you may need to change to get the best results and all sorts of things like that as usual this workflow is also available from the a very comfy nerd web page which is linked in the video description okay so a quick breakdown of what we're looking at here I've done a bit of reorganization compared to the original reposer so now all your image inputs are at the top below that is where you can set your prompts the positive and the negative as well as weights for things like your pose and inputs of support image and finally at the bottom is the funky new section for Segment anything here you can provide additional prompts to pick out the things you want from your supporting image as well as to remove the things you don't this is because the supporting image uses the IP adapter Plus image encoder just like the one used in the various instant Laura workflows meaning it will see everything in the image the pose the face the style whatever object positions are in it everything in there just like with the include pose option if your supporting image has a face in then it will merge that face in too giving you a different look of course this may be something that you're wanting to do but either way you should be aware of it the same thing goes with hairstyles too if your supporting image has long dark hair then it can result in your final generation having long dark hair as well this is why it's really useful to have segment anything so you can cut out things like faces or hairstyles segment anything automatically downloads models making it really easy now this may not be the best way of doing things but it seems to work so hey you could just provide a supporting image already done perfectly in your image editor of choice but using segment anything simply means you don't need another app and can do things quite quickly some optional bits are over to the right here disabled by default but there you can add a little bit of extra detail and do some upscaling there's also a black or white background option for the segmented supporting image which can make a difference in some cases and that's it in essence as easy as 1 2 3 select images select prompts and segment all right let's put it through its Paces so you can see some of the different fa es poses and outfits to get the best results from this workflow you'll need to pay close attention to the images you're using like in this example there is a painting for the face as you can see he's got a red hat a beard and he's wearing some sort of coat the pose is such that our generation should have a face torso and some lower body visible as well with the supporting image having a crop top and blue jeans checking the the segment anything previews we can see it's segmented nicely to leave only the clothing you've got the thresholds there for Segment anything and that will impact its detection rate the setting there usually around .3 is good but you may have to raise or lower that a little bit if you go all the way down to zero it will crash if you go far too low it will just detect everything and if you set it too high it may not find anything at all if you're not getting your outfit transferred be sure to check these previews to ensure everything is segmented okay and change the threshold and prompts to match your needs in this case everything looks fine and in the output you'll see it's mixed all of those things the generation has a red hat and beard let's have a little zoom in on there so we've got the red hat we've got the beard but if you look at the outfit you can see it's sort of a mix of the one he's wearing and the one from the supporting image well at least so far as the top goes he is wearing those jeans rather nicely and well that's because the face image over here doesn't actually include any leg wear so it's making a sort of combination of the stuff that he's wearing combining it with that and then turning it into a new item of clothing keeping that same outfit but swapping the face image to one that only includes the face and has no items of clothing means it has an easier time generating she now has a crop top and there aren't any other clothing items to clash if the pose you use means that your character wouldn't have the items of clothing you've provided in your supporting image then sometimes things can get a bit weird like you see here it really wants to give her that pair of jeans but she doesn't have any legs causing much Mayhem and confusion tiny details don't remain consistent because well this is stable diffusion and it just kind of works that way this means you'll get the best results if you don't use complex clothing designs things with words patterns flowers or t-shirts with complex designs on but of course as you can see here that may be something that you're looking for here we're essentially doing a redesign of a jacket so there she is she's got the sort of orangey bits there and that's glowing now if we do another prompt There She is again in the same pose same face but the jacket design has changed and of course we can keep queuing up those prompts and giving her different jackets all very much inspired by that supporting image but slightly different each time due to those very complex designs here you can see the arm tattoo has copied across as well so there's a sort of jaggedy pattern there and there and they've both both got a dragon t-shirt of course if we do another generation then complex things such as the tattoo and dragon have all changed there you can see it's on a different arm and the dragon t-shirt design is different simple designs on the other hand do remain fairly consistent such as with this top even if we do another random generation then the clothing Still Remains fairly consistent and we can keep going obviously you could find a few little changes but for the most part even as you keep randomly generating she will still be wearing that same item of clothing you can change the pose too so let's generate another random one here but I've got a new pose and hopefully she'll have much the same top on fingers crossed and there we go yep so she's changed the pose same face different pose but she's still got the same top and once again in a new pose and she's still wearing that same outfit one thing you must may find with simple clothing items however is that you may not be able to think of prompts that work well with this segment anything exclude section for these cases you'll need to do something like I've done here bypass both of these things so bypass the supporting image prompt and also rightclick and select bypass so if I turn it on that's how it is default and bypass group nodes will bypass those that way the preview image the one there where it says excl won't be used and it will just use the positive prompt simple but realistic images work well as clothing items too like in this example with a leather jacket once again you can change the pose and the outfit will remain much the same remember everything from the supporting image will influence your output even the style such as if it's a photo Vector art painting or whatever if you don't want the full level of influence then you can lower the the image strength for the support image that will help things like vector art items of clothing remain a little bit more realistic assuming your face input is a realistic one and of course you can always guide it by adding photo realistic style prompts into the positive prompt box while I haven't really used any prompting here that is indeed also another way to influence anything in your Generations so why not give it a go today and see what you can create you could even try strange things like negative weights now the stuff down at the bottom here way way way down is a little bit of a mess and there is likely a much better way of doing this given the segment anything also leaves those little outlines that kind of messes stuff up a bit sometimes so maybe filling these sections with noise would give better results who knows either way download it now from the a very comfy nerd website the link L for which can be found in the video description have a play and see if you can improve it any or you could just watch another nerdy rodent video instead
Info
Channel: Nerdy Rodent
Views: 22,766
Rating: undefined out of 5
Keywords: nerdy rodent, ai, ai art, reposer, stable diffusion reposer, reposer plus, stable diffusion consistent clothing, stable diffusion clothing, stable diffusion outfits, stable diffusion face swap, AI face swap, AI face generation, consistent characters with stable diffusion, comfyui, generative art, generative AI, stable diffusion consistent outfit, Stable diffusion face swap, Face swapper, Stable diffusion swap clothes, Ai clothes swap, Ai face clone, Instant LoRA
Id: ZcCfwTkYSz8
Channel Id: undefined
Length: 11min 21sec (681 seconds)
Published: Sat Oct 14 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.