Face Swap (Roop Alternative) Extensions - Stable Diffusion Tutorial (Automatic1111)

so RP's development has been discontinued but don't panic because I'll be showing you two face swapping Alternatives so you can continue your workflow using these amazing tools this video will be more than an overview as I'll be going into detail on what the different functions do so you can pick up where you left off or within the automatic one user interface a shout out to the supporters and thank you for 5K subscribers and hopefully 5K likes on this video but let me give it to you bite-sized there are some functions which are common across all face swapping tools and it's worth me covering this while covering our first RP alternative called reactor so I don't have to repeat myself over and over giving the same information these common functions will usually be all you need to get face swapping to work with more advanced options being available for power users you will have an image box to drop your reference and sometimes this may be accompanied by a box for batch images or multiple image files you'll see a check boox for enabling the extension and this also works for the disabling it and stopping it from influencing your images if I enable the extension drag an image into our image box and perform a swap we will get our generated image with a swap face there will sometimes be a chat box titled something like save original which will save both the image generated without applying the face swap and then the image with the face swap applied so if I generate a face swap with this option you can see we keep both versions of the image comma separated face numbers refers to the reference number assigned to face faces within an image from left to right and each face will have a number assigned starting from zero and increasing by one until do a noral faces you can use a number to decide which face within the reference image you want to modify the gender detection is used to modify only faces of a specific gender and sometimes you will need to specify a gender while other tools can detect genders based on provided reference images so for example if I take this image and select mail only you can see see that the face swap isn't being applied as our reference is female but if we reverse it it functions normally the face restore option would allow you to select a model for restoring the face alongside selecting the strength of the restore face application while code formal weight will determine the trade-off between quality and Fidelity when using that model this is done as part of the face swap process and different models will give different results so it's worth experimenting to see which option works best for your image you will also typically see an upscale section to increase the resolution of the face swap based on the models you have installed alongside selecting the size and strength for the upscale be careful to not set this number too high as you may get artifacts and if your computer is a potato it may freeze up at high resolutions upscaler visibility is just the strength and well visibility of the upscaler on the face you will also have various other settings which you can modify although typically these will already be set out by default for the extension engine to work there's a section to select your preferred model if you have others installed and you can determine the information which goes to the console and even perform hash checks on your images which I believe is to improve performance swapping Source image is described as starting from a given base and applying the diffusion process to it which seems to give similar results to swapping in generated image but requires a loading noising strength to work so give both a try swapping generated will replace the generated image with the swap version that has the new face and this can be overridden by using save original so you can keep both versions of the image now this example was based on Reactor which is probably the best alternative to RP for its ease of use and fantastic results to install it you need to ensure you have Visual Studio 2020 installed then you can grab the extensions URL and install it from the URL option within automatic one I received an error upon first installation as the program was EXP expecting a folder called Insight face to be located within the model section of the automatic One Directory so I just created one and copied in the iner model then relaunch the automatic one so we can find the folder file and finish the installation once installed and you've restarted automatic one using this extension is as simple as dragging the reference image into the image box and selecting the enable check box then generating the image as shown on screen via text image the results are brilliant but experimentation will be needed to find which settings work the best for your image here's the face to done via image to image and thankfully outside of the settings previously covered there were no additional features to worry about except for the postprocessing order which is just to decide whether you want to do restore face or upscale first but our next rou alternative is face swap lab and this is a far more complex and feature Rich tool than reactor which I'll be spending the bulk of the video covering as there are some features not available in reactor which I'd like to demonstrate this can be installed by first installing Visual Studio 2022 then taking the GitHub URL for the extension and using the install via URL option then applying and restarting the UI now we have two sections for using this extension we have a tab along the top ribbon called face swap lab and this contains some functionalities for building and batch processing images we then have face swap lab as a drop down below the image and text image tab we have our common functionalities like the enable checkbox comma separated face numbers and face detection via gender or we can do it via size if I generate a standard image we can see that the face swap is working and providing a good result with the default settings so I won't delve into the topics we've covered before and instead I'll explain some of these Advanced features which make this a good alternative to reactor but looking at some of the more unusual options we can blend faces based on our reference image and images within the bat section which will blend the Faces by averaging their characteristics for example if I combine the faces of Alexander davio with Michael Jackson we get this combined face you also have the option for using a face checkpoint which is a checkpoint created within face swap lab for the purposes of reusing faces and we'll explore this later under the similarity section you can discard images with low similarity or no faces compared to the original face so you only keep the images that have a similarity to your reference image or calculated Blended faces both check and compute similarity enable this function but check similarity allows you to use the minimum similarity and reference similarity sliders for manual tweaking while compute is automated it's worth noting that minimum similarity refers to the faces created through blending and minimum reference similarity refers to the faces within the reference image only showing here are images being generated on a 0.9 similarity for the reference image but at one the images do not generate as the similarity is not close enough to the reference image you then have the option to do postprocessing and adjust the mask settings under the postprocessing and advanced mask options section restore face and it sliders along with the upscaler are things we've already covered but there are some additional options we have used improved segmented mask which will avoid the typical Square mask and will prevent degradation far better but comes with a downside of potential artifacts on the border of the face the sharpen option can provide better results but it can also add artifacts and this is the same for color correction so use these carefully you can adjust the mask erosion parameters using these settings and the higher the settings the more the mask is reduced which may be useful for fine-tuning the pre and post imp painting sections are the same and essentially allow you to perform in painting before or after the face SW is applied to a generated image you can adjust the D noising strength which is the amount the image can be modified from the original you can also use prompts Samplers steps and even a new model alongside selecting an in painting seed which all function the same as your automatic one user interface and is nothing new then moving on to Global postprocessing there's nothing new here except for the ability to apply Global in painting to all faces instead of having to modify each tab finally let's explore the face swap Lab at the top ribbon as this section provides you with a range of useful functions for building checkpoints comparing faces extracting faces as well as exploring models and analyzing faces within an image the build section allows you to build checkpoints which facilitate face reuse and when a checkpoint is used it takes precedence over the reference image the faces used for the creation of the checkpoint will be merged and and then stalled under the given name of the character and saved under models face swap lab in the faces folder while also being available in the dropdown menu in the main interface once refreshed compare will allow you to receive a similarity score between two faces based on how similar they are the higher the number the greater the similarity between the faces from the tools perspective extract allows you to provide a batch of images and extract the faces from those images this tool tends to ruin the quality of the faces so unless I'm missing something you're better off cropping the faces yourself using another tool explore model will let you see data of associated with the model you're using to perform the swap like in swapper 128 on NX file I'm not sure what any of this means as it's beyond the scope of this video but it's there for the computer nerd to understand nodes and attributes lastly we have analyze face which will give you the output of the Insight face analysis model on the first face found providing information like age and a bunch of numbers which I'm sure mean something to somebody the batch process tab works like the actual extension itself where you can add in Source images and then perform face swaps on all of them based on the face tabs provided saving you time then lastly we have the global postprocessing which is where all of your postprocessing settings are safe for this section of the face swap lab allowing you to modify options we've covered previously for a tailored result thanks for hitting the like button and enjoy your two re alternatives I'll be coming out with some more videos so stay tuned this is bite-size genius and I hope you enjoyed

Channel: Bitesized Genius

Views: 17,411

Keywords: face swap tutorial, face swap tutorial video, Face swap tutorial ai, stable diffusion face swap lab, stable diffusion reactor, stable diffusion reactor install, stable diffusion reactor tutorial, stable diffusion reactor video, stable diffusion reactor extension, stable diffusion faceswap lab, face swap lab stable diffusion, face swap stable diffusion, face swap ai, roop deepfake tutorial, roop stable diffusion install, roop stable diffusion not working, face swap tutorial ai

Id: NKPlOSe5u-s

Length: 10min 35sec (635 seconds)

Published: Tue Oct 31 2023