Stable Diffusion OpenPose Beginner Tutorial | Step-by-Step Tutorial

what if you want your AI generated art to have a specific pose or if you want the art to have according to a certain image then control Nets open pose is your friend today I'm going to go over the model open pose in control net how to extract pose from an image how to create your own poses and I'll even dive a bit into these settings this is a complete beginner friendly tutorial but if you are new to stable diffusion please check out my previous stable diffusion beginner guide to get used to the interface first off the installation in your Google search box type controlnet web UI and the first result you'll get will be the link to a GitHub page you can find this in the video description as well this page will take you to the official GitHub page of controlnet's web UI extension now scroll up and copy the link of this GitHub page now go to stable diffusion go to extensions and you'll get a list of extensions you've installed here go to install from URL and paste your link in here now click install now if you go to your install tab you'll be able to see control netx extension in here just like mine but we are not done now go to apply and restart UI for the changes to take effect now the newly loaded web UI should have a control net extension right here if this didn't work make sure to completely restart the web UI by closing the command prompt and then doing a fresh restart and now we are done we have control net now in your web UI you should be able to see the control net extension if you expand this little section you see you get a ton of options I'll be diving into the main options we need very soon but first we need to download a model for this to work yep just like how you download models or checkpoints to generate art controlnet as models as well to download a model type control net hugging face on Google then go to this hugging face model right here here you can find details about the original control net now go to files and versions then models and here you'll see a list of models we need for our control net preprocessors as you can see we have a ton of preprocessors and all of this needs a model now click on this open pose.pth model and hit the download button then copy this file and go to your web UI install directory go to extensions sdwebui control net models and paste your model right here I have it pasted right here now we are fully done so let's take a deep dive into the settings first is the place where you need to upload the image remember when I said you can extract a pose out of an image this is where that happens you can upload your image right here now to enable the extension you can click the enable check box right here without checking this control net won't work if you have a low vram let's say four gigs or six gigs then it's best to check this low vram checkbox as well I have 8 gigs of vram and this seems to work fine without checking this on so always try out before clicking that next is the Pixel Perfect mode when using a model you should have a preprocessor resolution and Pixel Perfect make sure you don't have to manually set the resolution because it will automatically recognize it next is the allow preview button which will show the preview of the open pose model of your image control net works by using a preprocessor with a model preprocessors take your image and extract information for the model to use which is then applied on the image generated now the control type shows the preprocessors we can use and in this case we should only focus on open pose clicking on open posts and then going to this preprocessor tab will give six options including a none option the good old's open pose option and then open pose for various parts like face and hand right next to it is where you should load the open pose model if you click on the drop down you should be able to see it and if you can't make sure to click the blue refresh button down here the next important setting is the control weight which is like denoising strength in image to image tab since that controls how much the control map or output generated is followed relative to The Prompt the preprocessor follows the resolution of the image and control mode tells whether to put more effort to the control net model or the prompt or to keep both balanced now let's get to work and generate an image with a certain pose here I've written a set of negative prompts that I often use and in the prompt box I'm going to write the prompts I want in this case I want an image of a girl with long brown hair wearing a white blouse with raised arms since this is just a tutorial I'm going to go with a simple background I keep the sampling steps for 50 since I need some detail set the resolution height to 768 and hit generate as you can see the keywords I used hands up only worked to a certain extent I don't need this I want her to raise her hands above her head even if I use a keyword like raised hands you can see it doesn't give me the results I want okay so this is worse than the last one so what can we do we ask help from control net I open control net and load this image right here now wait let's not get ahead of ourselves images downloaded from the internet can have different resolutions and to get the best results it's easier if the width of your image the pre-processor resolution and the width of your generated art is the same to crop your image you can go to a site like berm which I'll include a link in the description and then enter your width and height then click save files all of this is completely for free no hidden fees like a ton of other online editors now load back the image make sure the open pose is selected and from the preprocessor select open pose and for the model select the open pose model and finally click on enable option to activate it now wait for the generation and there you go it has the exact resolution as our image and now if you click this second box right here you can see the pose they have extracted from our image see those raised hands that's the pose we need now that you know the basics you can play around with the values for example increasing the control weight and focusing more on control net will give the same result with a very more accurate pose but with a little bit of quality sacrificed now let's say you don't know the resolution of the image or it's an image that cannot be cropped since the subject fills the image check this image out as you can see this is a completely different resolution now I click the Pixel Perfect checkbox right here it generates and there you go it adjusted the settings and perfectly used it on our image but still make sure not to use images with too much space in my previous try I use this really large image and you can see the result I got remember even AI has its limits now what if you don't have a picture with the pose you need what if you have opposed you want to create yourself well there's a trick for that in Google search for open pose editor now go to this GitHub link and follow the previous installation process copy the link go to extensions paste it on the tab go back to installed and click on apply and restart and after a short restart you should be able to see a new tab called open pose editor right here if you click on it you'll see this new screen with an open pose Skeleton on the right you can change the positions of the skeleton any way you want in this case I'm going to go to the usual hands raised position if you want you can even add another skeleton and create two images if you want to save a particular pose go click on Save preset and enter the name you want you can always reset the pose using the reset button and load a pose using the load preset button pretty cool right just send this to text to image click on the send to text to image box and this will get loaded on your control net tab now to generate the art you can click on enable box click on open post and in here make sure to keep the preprocessor empty remember we only need a pre-processor sort of find the pose of an image we don't need it when the pose is provided make sure to keep the model the same though now hit generate and there you go the face is obviously bad but that's because we didn't use hires.fix and didn't have the generation settings relevant to the model but you can see they gave us the exact pose we had we can also provide our own image in this tab and generate the pose from that image as well remember that little tip so that's the end of this tutorial we obviously didn't touch the face of the hands but I'll be doing a video on that soon I hope you gained something from this video as usual make sure to like And subscribe and stay tuned for more AI updates like this one that's it for today AI gang and I'll see you all in the next video [Music] foreign [Music]

Channel: The AI Outline

Views: 32,624

Keywords: stable diffusion tutorial, stable diffusion, controlnet stable diffusion, stable diffusion ai, stable diffusion ai art, stable diffusion art, stable diffusion controlnet, stable diffusion browser, stable diffusion online free, stable diffusion installation, stable diffusion gui, stable diffusion beta, stable diffusion openpose, stable diffusion controlnet openpose, stable diffusion img2img, tutorial, stable diffusion prompt guide

Id: ALnmwCpM2xE

Length: 8min 21sec (501 seconds)

Published: Mon Jul 03 2023