FINALLY! Its Easy Step To Fix Hand For Your AI Image In Stable Diffusion

hi WhatsApp everyone in this video I will provide one of the easiest and best techniques so that we can get an image with a fixed hand position like this in stable diffusion but before that there's nothing wrong if you subscribe to this Channel first because your subscriptions mean a lot to us many thanks okay let's get to the tutorial actually there are several ways that are usually done so that our generated image in stable diffusion can get a fixed hand position some use postex in paint depth control and so on but in this video I will show you one method that I often use in stable diffusion the trick is to use a combination of open pose control net and soft edge control net but before that make sure you have updated control net on stable diffusion to the latest version so that you can use open pose control net and soft edge control net if you have updated the control net make sure you have activated the multi-controllenet option you can check the video on how to update control net and activate multi-controlenet in the description of this video okay before going to stable diffusion here we need a reference image we really need this reference image to be able to use open pose control net and soft edge control net we can look for interesting and complete reference images on Google Images but actually I would rather suggest that you look for image references on the Pinterest website there are lots of reference images with various kinds of interesting poses that you can download look for reference images with interesting poses for you on Pinterest or on Google Images [Music] when finished now we return to stable diffusion as usual I have set up a prompt here I will try to make an image based on photorealistic deep Shadow Dark theme with the subject being one girl smile black hair and so on I also use Laura Mecca girl to make the image more detailed and specific than as usual I copied this prompt on Google documents and I have included the link in the description of this video complete with a link where you can download the Laura model next we go to the settings actually for free settings it's up to you what you want and like for the model I use Photon one of the best realistic models in my opinion then for the sampling method I set it to DPM plus 2m caras then I set the sampling steps to 32. for the size I set it to 512 X 768. meanwhile I set the CFG scale to 7.5 like this then next we scroll down in the control net unit zero tab we drag and drop the reference image that we have prepared like this this is the image that I downloaded earlier I want to make an image with the subject of a girl with her hands posing to form a heart return to the control net tab we click enable then select open pose on the preprocessor tab we can select open pose full or open pose hand then click open pose is more important then click run preprocessor to see the preview then on the right side a preview of the open pose preprocessor will appear like this when finished then we click control net unit 1 next to it then drag and drop the same reference image as we entered in control net unit 0 earlier then click enable and select softage in the type control in the preprocessor options we can choose one I often use soft Edge pdnet or soft Edge head to get a more detailed reference image here I choose softage head then click run preprocessor to see the image review then on the image review tab we click download in the upper right corner like this then the review image will be downloaded to our PC or laptop like this the next step we open the image editor that we have on our PC or laptop can use Photoshop Pixlr or photo p in this example I use Photoshop then we just open the reference image from soft edge control net that we downloaded earlier in Photoshop then we can select using the pencil tool or brush tool we use black base color here we will delete or replace some of the white line and Shadow objects with black and what we will be left with is the hand object only the purpose of leaving only the hands is that later the soft edge control net preprocessor will only process and generate the hands while the other parts will be processed based on combined data from open pose control net and from the basic prompts that we enter make sure there are no white lines or Shadows left except for the sides of the hands like this when finished we just save the image then back to stable diffusion on the control net unit 1 tab close all the images that we pinned earlier including the preview image reference then just drag and drop the hand image that we edited in Photoshop earlier to the control net tab like this then on the preprocessor tab we set it to none because we already got the image reference that we will use make sure the model tab control sd15 underscore soft Edge is selected like this then on the control weight we can set it at one or below here I set it to 0.75 like this when finished scroll down and make sure control net is more important is selected as well then we can scroll up and generate and here it is the result we find that the hands can follow the soft edge control net well and the other parts are generated following the open pose control net flow and the prompts sometimes we don't immediately get the right image on the hand we can find the best by adjusting the control net weight there are various variations to make this image more interesting one of them is making the image widescreen within paint underscore only plus Lama control net this is the generated image that I have developed from the base image that was generated earlier I made a complete tutorial some time ago on this channel I have embedded the video link in the video description okay now we'll try again with the second reference image here I have set up a simpler prompt I want to generate the image of Arena makihara a girl who is a Manga and Anime character for that I also need the help of Laura Arena makihara which I put at the end of The Prompt the model I use is still the same Photon V1 then I set the image size to 500 x 800 like this scroll down then on the control net unit 0 tab we click enable then we drag and drop the reference image that I downloaded earlier to the control net tab like this then on the control type we select open pose then on the preprocessor we select open pose full then in the control mode section we set it to control is more important we try to click run preprocessor to see the preview eat later the preview image of the model will appear to the right of the reference image the next step is to click control unit 1. then drag and drop the same image again like this then we click soft Edge and don't forget to click enable the control net as well then in the preprocessor section I select soft Edge head like this also click control net is more important then run the preprocessor to see a preview of the model after the model preview appears then we click download like this the next step is to open Photoshop and open the reference model image that we downloaded earlier then we change or delete all the white lines or shadows and leave only the steps here there are times when we have to cut the white lines manually like this so that the shape of the hands is really appropriate and realistic and doesn't cause the AI to misread the flow of the soft Edge preprocessor later so the results we get are really precise on this part of his hand When Everything feels right we press save the image then return to stable diffusion and close all images in control net unit 1 and drag and drop the image that we edited earlier like this then scroll up and generate and here are the results as seen the side of the hand really fits and looks pretty precise like this the side of the hand that looks bokeh also adjusts to the reference image that we entered at the beginning the last step if necessary we can do upscaling so that the image becomes more detailed and Smoother by using image 2 image I've often practiced this method in my previous videos we select the image then click Send to image to image in the image 2 image tab we select resize by then we increase the scale according to our wishes here I raised to two times then we set the denoising strength to 0.3 or below and for the CFG scale we can set it to default or we can follow the CFG scale setting in the text to image here I set it to 10 then for the sampling method I set it to DPM plus plus 2m Karis like this then we scroll down in the script tab we select ultimate SD upscale then we change the target size to scale from image size and for the upscaler we set it to 4X Ultra sharp or R erzgan 4X you can also use other upscalers that are most suitable and what you normally use when it's finished we scroll up and generate wait until the results are complete and this is the final image we get I've also made several other images with this technique and the results are as you can see right now I think it's cool and very easy to practice you can try it with a variety of poses and hand styles that you like my tips are to make sure the reference image has good detail and if necessary you can edit and add detailed lines and white shadows in Photoshop like I did earlier it will really help AI to grind the image with Precision so what are you waiting for give it a try right now see you in the next video and many thanks

Channel: SahiNama

Views: 83,188

Keywords: stable diffusion tutorial, stable diffusion, ai art, fix hands stable diffusion, fix hands ai, perfect hand stable diffusion, fix hand tutorial, sahinama, sahinama stable diffusion, waifu model, ai, paint, ai art generator, a.i., draws, ai-drawing-illustrations, drawing, sketch, tutorial, ai drawing, ai paint, cosplay, lora, vae, controlnet, openpose, open pose tutorial, softedge tutorial, combine multi controlnet

Id: UgVOQTjahxc

Length: 10min 10sec (610 seconds)

Published: Thu Jun 29 2023