Face Swapping in A1111: Ip-Adapter Face ID Plus V2 (Better than Roop, Reactor and InstantID)

hey there creative minds and digital art enthusiasts welcome to my AI Force where we dive deep into the mesmerizing world of AI art unlocking the secrets of stable diffusion and Beyond Today we're diving into the fascinating world of face swapping with the latest model face ID plus V2 if you've experimented with models like plus face before you're in for a treat because the updates from IP adapter continue to impress us through my tests I've discovered that the face swapping capabilities of this new model significantly outshine its predecessors and it seamlessly operates on both automatic 1111 and comfy UI interfaces imagine transforming Angelina Jolie the Hollywood icon into various characters from mid Journey generated images this is a medieval warrior woman generated by mid Journey this is the warrior Angelina jol and here's a portrait of a beautiful Renaissance woman let's make Angelina travel back in time how about we Crown her a queen the results are astonishing right now let's walk through how to install and use this groundbreaking model in the automatic 1111 many of you might be familiar with control net the PowerHouse extension for automatic 1111 today we're focusing on two control types within it IP adapter for face swapping and open pose for maintaining the original head pose IP adapter represents a more advanced iteration within the control net compared to other control types if you're working with an older version of control net you might find that only open pose is visible an IP adapter is missing from your list of control types this means it's time for an update to your control net to access the full capabilities of the newer models to get started we'll need to download the corresponding model and Laura from hugging face don't worry I'll drop the links in the video description below in some scenarios you might encounter a missing Insight face issue if that happens the solution is outlined in an article Linked In the description as well first let's fetch the model needed for open pose after entering the hugging face link into your browser you'll be directed to the control net model download page where you can find all the base models and configuration files related to control net let's navigate through the page and look for the file labeled with open pose here you'll notice two types of files one ending in pth which is the model file and another ending in yo known as the model configuration file in most cases the yaml file is already included in your control net setup so there's no need to download it again focus on downloading the pth file that's the one you need to bring the open pose functionality to life simply hit the download button located to the right of the file listing to begin the download process next up we're looking at the face ID model and Laura required for IP adapter to ensure face ID's consistency once you land on the hugging face download page search for the IP adapter face ID plus sd15 and sd15 Laura files you'll find a download button on the right side of each file listing click on these buttons to start downloading the files the sd15 in their names indicates that these files are designed to work with the SD 1.5 model rather than the sdxl model it's worth noting that the SD 1.5 model not only delivers excellent performance on its own but also offers enhanced compatibility with other control types within control net once downloaded place the model and Laura files in their respective folders within the automatic 1111 first navigate to the root directory of the web UI where you'll find the file named webui user dobat although I'm running stable diffusion on a server and the appearance might differ from a local deployment rest assured the file structure remains consistent once there proceed to click through the following folders extensions then SD web UI control net and finally models within this location I've placed two essential control net models one is the open pose model and the other is the face ID plus V2 Laura files have a specific storage requirement return to the root directory of the web UI then navigate by clicking on models followed by Laura this designated location is where the Laura file should be placed with the model configuration and Laura files all properly placed in their respective folders you're ready to start face swapping in the web UI before diving into the face swap I took a moment to prep Angelina's headshot the original photo turned out a bit too luminous using it as is for the face swap would lead to an overly bright result in the final image to tackle this I turned to photoshop to dial down the brightness to a more natural level and toned down the Highlight areas on the face once the image is prepped and ready to go head over to the automatic 1111 and pick a realistic model tailored for SD 1.5 for this demonstration I've selected the Cyber realistic model but feel free to explore and compare effects with other models to see what works best for your project Pro initially you can skip adding a prompt next let's utilize the Laura we downloaded earlier to further refine the consistency of the face swap simply click on Laura locate the face ID plus V2 sd15 Laura file we've just downloaded and select it this action will insert the Laura into our prompt input box to fine-tune the influence of Laura on our project hold down the cotal key and press the down arrow key this will decrease Laura's weight ideally setting it within a range of 0.5 to 0.7 return to the Generation section as we're about to change a face simply click on in paint to allow stable diffusion to take over the task of repainting the face for us then proceed to click and upload the image whose face you wish to replace following that grab the brush tool and gently paint over her face it's a good idea to cover a slightly larger area than the face itself this technique ensures a smoother blend allowing the new face to seamlessly integrate with the surrounding features enhancing the natural look of your final image next let's adjust the in paint parameters increase the mask blur slightly this helps in reducing any harsh seams that might appear after changing the face ensuring a smoother transition you can leave the other parameters at their default settings for now choose a sampling method that aligns well with the realistic model you've selected consider increasing the sampling steps as well a higher value here can significantly improve the quality of the final image ensure that the width and height settings match those of the original image to maintain consistency in Dimensions adjust the CFG to fall within the range of 6 to 8 last lastly set the denoising strength between 0.4 and 0.6 after fine-tuning the inpaint parameters it's time to configure the control net settings navigate to the control net section and expand its options make sure to enable it by checking the enable box opt for Pixel Perfect next enable upload independent control image then allow preview click on this area to open the file browser and select Angelina's head shot for the control type select IP adapter under pre-processor choose IP adapter face id+ and for the model opt for IP adapter face id+ V2 sd5 once everything is set up click on the button that resembles an explosion to preview the IP adapter effect this step is important to catch any potential issues early on now let's proceed to set up the second control net utilizing open pose to maintain the character's head pose accurately scroll up to locate control net unit one ensure that you check enable to activate this unit and select Pixel Perfect then select upload independent control image and allow preview click on this area to bring up the file browser where you'll choose the image whose face you intend to change for the control type select open pose when choosing a pre-processor go with the basic open pose option and for the model pick sd15 open pose after making these selections press the button that resembles an explosion to get a preview of the open pose effect this action allows you to visually confirm that everything is set up correctly and to identify any potential issues before proceeding you'll notice that the skeleton representation of the character's pose has been successfully generated indicating that the open pose function is ready to maintain the Integrity of the character's original posture throughout the face swap process after configuring both control Nets we're ready to bring our creation to life simply scroll up to locate the generate button and give it a click the outcome is quite impressive wouldn't you agree the faces angle and lighting blend naturally into the new context key features such as Angelina's iconic curved eyelashes captivating blue green eyes and distinctive lips are accurately captured and beautifully integrated should there be any slight discrepancies or details that need fine-tuning Photoshop becomes a valuable tool for making those adjustments

Channel: My AI Force

Views: 4,230

Keywords: aiart, stablediffusion, generativeart, stabilityai, stablediffusiontutorial, a1111, sdxl, ip-adapter, faceswapping, portraitphotography

Id: kVGdhJPTYQ4

Length: 11min 20sec (680 seconds)

Published: Sat Feb 03 2024