ฝัง
- เผยแพร่เมื่อ 5 ก.ค. 2024
- #stablediffusion #reactor #faceswap #faceswapping #a1111 #aivideo #aiimages
00:00:00 introducing ReActor in A1111 and faceswapping showcases
showcases in videos and image face swapping
00:00:18 why ReActor instead of Roop?
00:01:16 installation of Reactor
00:02:18 using ReActor in A1111
show settings and example usage
00:04:51 face swapping in batch image for videos
show how to generate frames from Davinci resolve and swap faces in a video using ReActor frame by frame in A1111
Computer Specs:
Laptop: Legion 5 Pro
Processor :AMD Ryzen 7 5800H , 3201 Mhz
System RAM: 16.0 GB
Graphics GPU: NVIDIA GeForce RTX 3070 Laptop GPU 8GB
ReActor extension for Stable Diffusion
github.com/Gourieff/sd-webui-...
faceswaplab (another extension that you maybe interested in)
github.com/glucauze/sd-webui-...
A1111 for stable diffusion.
github.com/AUTOMATIC1111/stab...
This video is about face swapping in Automatic 1111 using ReActor which is a fast 1 image face swap extension built on top of Roop that I have explained in a previous video.
We will see how it’s done in text2img, img2img, and face swapping in a video.
o Now Why ReActor instead of Roop?
o ReActor is allows uncensored NSFW images unlike the original Roop
o More features, gender detection, better scaling
o Allows to use multiple faces is source image to swap multiple face in target images
o Roop project has ceased development.
o Installing ReActor for A1111
o Using ReActor with text2image
o Using it with img2img
o Using it for face swapping in videos
o Note: After detailer can be used for face swap in images too and with LoRA may give very good results if LoRA of a person is present.
o Both Roop and Reactor uses inswapper_128.onnx model from InsightFace so their results are somewhat similar, but ReActor is more flexible and gives us more control, so it is a better option.
Face swapping is very useful to create consistent faces as well.
Notes: it is better if source and target images to have similar face structure, that will give better results.
So as a conclusion, ReActor is more flexible and gives us more control in comparison to roop and is a fast and effective face swapper extension for Automatic 1111.
Attribution:
Thanks to Pexels.com for the free videos they provide
Honye Sanges for pexels-los-muertos-crew-8447488 (720p) video
and video id ID_4723042
Thanks to Insightface developers as well who have provided the underlying technology for these amazing tools and their face detection systems, check more at:
github.com/deepinsight/insigh...
Note: ReActor can be used from Extra's tab and batch processing which can have some advantages if you dont want to go through img2img tab. - วิทยาศาสตร์และเทคโนโลยี
What a great tutorial! Thank you!
Thank you so much for this super useful tutorial!
Thanks for this I finally know how to faceswap now haha
cool, happy faceswap :)
You are so wonderful for sharing such great information!
It was very useful for me because there is not much information available online in Japan.
Thanks to you, I have been able to try various quality video generation.
I have two questions I would like to ask.
In roop, I was able to change only the face by setting the Denoising strength value to 0, but in Reactor, the hair and the background around the face were also affected a little.
Also, the images generated by the batch process are not consistent when compared before and after, and the images appear to flicker due to slight differences.
Do you know how to solve these problems?
You are welcome, You can change Restore face visibility from 1 to 0.5 for instance, that may make it smoother ....unfortunately, there are no settings that gives us option to control padding in Reactor so some hair may get affected too... also disable any other roop based extension before using this one(such as Roop or faceswaplab), possibly having larger input image size...
if it didnt give you the results you need, then considering testing github.com/glucauze/sd-webui-faceswaplab which is also roop based and good.... or github.com/facefusion/facefusion which can be run standalone for instance, and best of luck.
@@AI-HowTo
Thank you for your kind attention.
I immediately tried adjusting the Restore face visibility as you suggested.
I was able to achieve the smooth movement that I was looking for a little bit!
As for the effect on the hair, I will try to find a way to control the padding in combination with Reactor.
Thank you for the interesting video! Please forgive an extreme noob question.
But the first step here had you downloading and installing MS Visual Studio. Does this mean ReActor will only work on MS Windows systems and not say on Linux or OSX?
Thank you in advance.
You are welcome, I am not sure, but based on their webpage, it can also run on Linux and Mac, I have not seen however the exact steps there if they are different from Windows, or if MS Visual studio is a must for this process.
thanks
I am running stable diffusion in the cloud via RunDiffusion. I've gotten everything down except when i go to the settings, there is no models even though i already have the inswapper for the roop extension. :( do you know how to add models for this extension? i've tried /models/reactor/ and placing the inswapper there but it doesn't show up
you can put inswapper_128.onnx inside folder
\stable-diffusion-webui\models\insightface
hopefully that works.
in my installation the file was directly under this location, then there is another folder inside it stable-diffusion-webui\models\insightface\models\buffalo_l i find the remaining model files such as genderage.onnx
this contradics with the documentation... as the inswapp_128.onnx was not inside the reactor extension folder.
my batch job is taking alot longer, i have a 2060 with 8 gb of ram so is there particular settings that could be slowing it down? I'm even outputting at half the size
not sure, this runs on CUDA so it shoudl run with decent speed, I suggest looking up similar info using ComfyUI, hopefully there you would have better speed, A1111 has been recently buggy more than usual.
I see 'ReActor - STATUS - Running v0.4.3-b3' in terminal and showed in Extensiton list but it dosen't show up on WebUI. Do you know a reason by any chance?
sorry not sure, I suggest applying for updates and restarting hopefully that may help if you didnt try that already... I tried it on A1111 v 16.0.
Thank you for your reply!@@AI-HowTo Appearently it was because onnxruntime wasnt installed properly.(asked ChatGPT) Thanks again for your video, this is very helpful!
@@eiji24gwhat did you do then ?
after the installation, I can not see in the Stable diffusion category txt2img after the ControlNet the ReActor?
try updating, if it still doesnt work, I suggest deleting venv folder inside stable diffusion folder which will reinstall the main packages of stable diffusion, takes a while, but usually improves things especially when you have some conflicts in your installations...I didnt do that, for me, things worked without any troubles, but some users reported having similar problem to yours.
is there any way i can use a mask layer to exclude some part like if a face with cooling glass
if you are doing it for a single image, then yes you can mask the part of face that you want to swap, set denoising to 0, and enable ReActor, it works on the masked area only (Inpaint area =masked only option).... but for a Batch of images, that also works with batch processing and mask layers, but you need the right model to generate the masks with the object that you want to include/exclude, I have no info about that unfortunately.
If you already have the images saved on your harddrive then you dont need to use img->img tab. It is better to use the Extra's tab. That way you dont need to load a big SD model which steals memory for the Reactor extension. There also exists models nullmodel which when loaded uses no RAM at all.
You also can use ffmpeg from CMD/Terminal/Powershell to generate images from a movie.
create images: ffmpeg -i ".mp4" -qscale:v 2 pics/%05d.jpg
extract audio: ffmpeg -i ".mp4" -vn -acodec copy audio.aac
put it back together: ffmpeg -i pics/%05d.jpg -i audio.aac out.mp4
It is also better to use JPG (with max quality) rather then PNG, the Reactor extension has problems with PNG sometimes and crashes.
Do note that movies with decimal framerate i.e. 29.97 will be problematic to put back into a movie again (from the images), so here you first need to reencode the video into a fixed integer framerate for example 30. Handbrake is good for this.
Thanks, I didnt know ReActor can be used from Extras :) not sure if it was used from Extras back then, this indeed can be faster...thanks for the remaining tips too.
Thank you always. I succeeded in changing my face through roop, is there a way to change my outfit and hairstyle naturally?
You are welcome.
it is difficult to do it without inpainting, but it is possible using After Detailer, if you download fashion detection in huggingface.co/Bingsu/adetailer/resolve/main/deepfashion2_yolov8s-seg.pt?download=true and paste it inside >>>\stable-diffusion-webui\models\adetailer folder, restart A1111 to see the model in after detailer models drop downbox.
then in the prompt you put the description of the clothes and change inpainting options (Inpaint denoising strength to something higher than 0.5) ... this will allow automatic detection of the clothes and attempt to change them, but doesnt work accuratly all the time... you can play with the settings and try to see what works best for you....doesnt work well for videos though because some inpaints wont be successful.
Controlnet could also be utilized, but same thing, not very accurate either, you can google that possibly someone made a video about this here or there... so there is no simple direct solution to this as far as i Know.
@@AI-HowTomake a video of this topic.changes clothes only.
Or make a playlist.
ALOT OF PUBLIC SEE THIS TOPIC BUT CAN'T FIND A GOOD WORK.WE WAIT FOR YOUR VIDEO THANKS😊.
Thanks for the tip, indeed, I have put some tips here and there in some videos, but it is best to have a dedicated video for this, hopefully I will plan for this video next month as early as i can.
@@AI-HowTo changes clothes only but please not used inpainting process.implement a method that after we used this for production level app.
When i drag the folder into DR from the SD batch process, the numbers come in partially out of order (1-10) clips are out of order. all of the pictures open up in the top left master area and are not one thumb nail image as shown when you do it. when i press play , its a frozen frame and wont change unless i use the slider and drag for what seems to be eternity to get the next frame. the time is way to long for the video as well. this is my first time using DR, what am i doing wrong?
DR only accepts the images in a sequence name such as file(1), file(2)...etc.... select all files in the batch process folder, (Crtl A), make sure that the first file is in focus, then rename them all (F2), choose a name, and they should change into a the proper sequence (take a back up before that incase rename went wrong), then you can bring them in as a sequence...sorry i might have skipped that part in the video by mistake or just talked about it verbally
You can also check this short video here, maybe it helps, th-cam.com/video/3QG3l2r4yRU/w-d-xo.html
I'm having a problem with ReActor giving me the same exact image in img2img. I've generated a character, and I'm trying to swap my face to it, but no matter what I do with the settings, it simply does not change the output. I can do normal img2img outputs without ReActor and get changes. Any ideas?
if you Enabled Reactor this should work, if it works in txt2img then it should in img2img, unless you forget the checkbox of enable option... you should check the console for debug information, it often produces messages explaining what is happenings, such as finding the face, its gender...etc... otherwise
@@AI-HowTo I ended up switching to FaceSwapLab and got it working. No matter what I tried, ReActor wouldn't do anything. I'm suspecting there was some sort of corrupted file or something, but now that I've got an alternative, I'm not worried about it.
Thanks for the response.
open source can be messy... all use same underlying face detection and swapping mechanism from Insightface library, what matters is getting something working, roop, reactor, faceswaplab, they just differ in certain set of features, but the underlying part is the same, best of luck.
@@Telos954x Its because some of the neccessary files are actually missing when you follow the installation guide. You had to manually download and move certain files to get it working. Its a very recent problem related to Insightface.
Hello I am using Roop unleashed rn, its serves it purpose after some tweaking. I'd like to try this one but cannot install it, too many errors. Maybe its because I use Stability Matrix
yes, probably using other web ui, and conflics with existing onnx files and other extensions... if you are using roop unleashed, and it's working well for you, forget about ReActor, because it too built ontop of Roop and they both use the same inswapper_128 model for swapping the faces, so I doubt you will see any real improvement in quality of the output, stick with what works for you... these python libs can mess with one another easily.
@@AI-HowTo thanks for a quick reply! yeah, you might be right, for now I won't change this. Maybe someday I deceide to reinstall WebUi so then I'll try. What do you think about ComfyUi? btw. I remember once, when I put some strange files into the Buffalo models folder, I got unusual results using Roop. I wanted to replace my friend's face with Brad Pit and a black man came out of this.
ComfyUI is generally faster than A1111 (but not faster with roop), A1111 is easier to use and can easily find large number of extensions to use with A1111...didnt try it there...another image generation means large denoising strength or you are generating using text 2 image prompt, so getting a different ethnicity than your prompt happens... Roop just replaces the face, after the image is generated (After the last step) and it uses a separate engine/libraries than Comfy or SD or A1111, so Roop speed is independent of the Web UI you are using as far as i can tell.
@@maximus613 Sability Matrix works with this so that means something else is the issue. Just wanted to advise you on that.
I get this problem where Reactor wil skip any image/frame of a video that doesn't has a face in it. So when I put it back into video(output) the audio will be out of synch and video will be all choppy. is there anyway to solve this problem?
even if the image doesnt have a face, Reactor will output the image (based on my batch image tests), not sure why this is not working out for you, you should do some tests for 5 images for instance to see where the problem happens
Thanks for your detailed video. I'm trying to contact with you directly for a comissioned project. Do you have a Patreon or a way to contact you? Thanks!
Thanks, unfortunately, I cannot take any work now, nor accepting contacts outside this channel.
how do I install reactor if Im running stable diffusion from a notebook on jupiterlab?
sorry, havent tried it, so i cannt provide any useful insights about the process.
Hi, i have to ask, what workload do i need to download when installing the visual studio? it seems its not working. i still cant see reActor on my webui. and i have not installed roop.
I had visual studio installed before, however usually we download Visusual studio community edition and include:
- python developmnet
- Desktop development with C++
- Visual studio extension developemnt
these should include any dependencies
if restarting and updating didnt work, you can refer to troublshooting guide
github.com/Gourieff/sd-webui-reactor#troubleshooting
for me, i faced no problem installing it, but some did report that it was not properly installed, some reinstalled insightface library and its onnx model as mentioned in the troubleshooting guide and it worked out for them.
@@AI-HowTo thank you. i have to try this out.
Such an extension with tons of troubles in installation. Why he gotta make it so hard, eventhough I tried following every step carefully, missing module protobuf insightface keep showing up. Very frustrated
sorry to hear that, if you tried github.com/Gourieff/sd-webui-reactor#troubleshooting and didnt work still, forget about it, and try other extensions such as faceswaplab in A1111 , check github.com/glucauze/sd-webui-faceswaplab hopefully it installs without a trouble... for me Reactor installed without a trouble, however many people reported troubles with insightface models and such and had to download the models manually.
How much time is needed to install reactor on the web UI? I have been waiting hours.
depends on your internet speed, usually it takes couple of seconds or minutes to download any models needed...check the console it should report any errors or problems.
I don´t know if its roop or the inswapper_128.onnx but they don´t understand very well faces doing NSFW things... any of you know of a better model or application that does a better job with NSFW faces?
ReActor allows NSFW images too, unlike the original Roop extension.
@@AI-HowTo i used a roop without the nsfw filter and the results of faces doing stuff were not good… that’s why I asked
in general, these face swap tools are not great, if the face is not clear, or partial obstructed, the output may not always be as good as one hopes, hopefully soon, there will be better tools than these.
Hey, i dont know why but when i configure my batch it wont generate the pictures in my folder . Does it not work anymore?
I just checked it out on A1111 v1.8, it still does, there could be some conflicts in A1111 upgrades that may have caused you some troubles. I just tried it out on 1.8 with batch folder, and without and it worked still .... A1111 may require clean installation every now and then (or just delete venv folder and rerun) to remove some conflicts.
thanks for your help, reactor is running but I think I still miss :
onnx>=1.14.0
onnxruntime>=1.15.1
I really don't know how to install them
not sure why this is happening with many people, I think primarly because of a previous installation of Roop based extensions that is causing some conflicts, (if you have roop installed before, make sure it is disabled too) and follow the trouble shooting guidehopefully that helps....
github.com/Gourieff/sd-webui-reactor#troubleshooting
@@AI-HowTo I will see the guide, thx
too difficult for me, don't know where to put, how to open python command ?? : pip install onnx==1.14.1 onnxruntime==1.15.1
all commands are typed inside the command line ... the command line must be opened from within stable difffusion installation folder (for instance open that folder and type cmd in the navigation bar on top of the window, it will open) or navigate there then start applying commands... you can ask GPT for exact steps, hopefully it can guide you through step by step.... alternatively, you can try reinstalling the packages and applying updates hopefully that solves the problem all by it self... I peronally didnt get the problem you mentioned, but some did.
@@AI-HowTo thank you, this message was a great help, I wasn't injecting Insightface in the good place, I'm totally new with python console, it works now
hi, can u show me, how to uninstall roop or another extention? for clean uninstall
I usually go to Extension Tab in A1111, uncheck the extension (Roop), then go go the extensions folder inside the stable-diffusion-webui folder and delete the extension folder related, in some cases however, some left overs remain in the venv, this is why, if things didnt go as planned, i delete venv as well and let A1111 rebuild it a new with the required librararies, this takes few minutes depending on your internet speed.
If this is bugging you, use version 1.7, it will definitely work better!
Thanks for your video
But the batch process is too slow for me for about 255 images?!
Any way to make it faster?!
Beside faceswape is kinda bad it's like more a face between source and target character!!!
And even eyes color doesn't change!!
you might be using large image sizes for the swap, because 255 images should be complete in under 30 minutes i think using RTX3070 8GB laptop graphics card... if image size is smaller such as 720x1280 or smaller it becomes faster.
unfortunately, current 1 click face swap tools does a merge with help of insightface library, it is difficult to get great results with 1 image without training, thus far this is as best as it gets, not sure if there are better tools out there that can produce higher quality without Long training (some tools takes weeks of training to produce high quality face swaps)
I am using PC 3050 RTX 8 gig ,I tried to swap for 720 video and I still wonder it should take less time to do that maybe it's because some driver or sdx updates!!
But the face swap quality was so disappointing especially in comparison with your project
Can you please make a video to make a high quality deep fake with no matter how much time it take?!!
I am sorry, I have no plans to for such video, but you can check github.com/iperov/DeepFaceLab which has tutorial about this subject, but literally it takes days to train one subject to get good swap results... 1 click face swap results are usually not as good as we hope, This is why we need to select a suitable target video in which the subject would have similar face structure (for example, dont expect round face to fit with oval face so great...having matching structure improves the quality of the output... but still, quality will remain limited compared to deepfacelab which does a more accurate face swap taking input (video) to extract all face poses and output (video) so 1 image is never really enough for great faceswap) ... hopefully soon we get better tools though that allows better 1 image faceswap.
Thanks for everything man
Is there a way to swap not just the face, but hairs too?
I dont know of an exact method, Roop can only do face swap not the entire head, biggest propblem with entire head is that structure of the heads may be different.
it could be possible to use After detailer however you need to use a LoRA of a person's head... then increase the mask erosion parameter, it basically automatically inpaint hair and head, but not very effective because the erosion parameter is done in pexesl and may not work the same if used with videos in which head size changes in the video.
reactor wont show up in the ui. says its missing the insightface module
(if you have roop installed before, make sure it is disabled too) and follow the trouble shooting guide for insightface installation, hopefully that helps.... in my case, it installed normally without errors so i didnt following this guide
github.com/Gourieff/sd-webui-reactor#troubleshooting
i dont see .png or .jpeg in the format drop down menu on Davinci resolve. everything else is there but those two
I think jpg, png is avaialable from version 18.5 and after... you can do the same using Photoshop (export, render video, choose PNG or JPG as format , it gives higher contral over generation of images) ... or online tools, but the previous two options are my favorite, quick, local, and fast.
i update and see it now thanks
I have SD Running with a AMD card. Can I run ReActor also with a AMD card?
Yes.
When i put denoising to 0 i'm having issues with SDE Sampler, with other sampler works fine, why?
not sure, if another sampler works then stick with it
@@AI-HowTo denoising zero with non-sde sampler still changes some details in the picture (makes other faces look worse). I remember in the past I could go to the inpaint setting window and did not have to draw a mask but it would work (at least with faceswaplab) to change only the face it s supposed to be. now this doesnt work anymore, i have to upload a mask. strange..
I can’t get it working at all, it was working fine before I had to reinstall everything but it’s not working anymore 😞
A1111 is becoming more buggy over time unfortunately, you should check the bug info in the git hub page hopefully you find something that fixes it.
hey tanx for video
I do all the steps, but after processing it shows me the same original image without changes in the face :(
usually this happens if Enabled is not checked, double check that ... also check the console for debug messages in cases something is missing...ideally if Enabled it would displaying something like :
19:47:24 - ReActor - STATUS - Working: source face index [0], target face index [0]
19:47:24 - ReActor - STATUS - Analyzing Source Image...
19:47:27 - ReActor - STATUS - Analyzing Target Image...
19:47:28 - ReActor - STATUS - Detecting Source Face, Index = 0
19:47:28 - ReActor - STATUS - Detected: -29- y.o. Female
19:47:30 - ReActor - STATUS - Detecting Target Face, Index = 0
19:47:30 - ReActor - STATUS - Detected: -24- y.o. Female
19:47:30 - ReActor - STATUS - Swapping Source into Target
19:47:31 - ReActor - STATUS - Restoring the face with CodeFormer
Enable (Check box)
I have two extensions and neither of them works / reactor force and reactor@@AI-HowTo
sorry to hear that, for me, things worked out without any errors, several other people reported errors especially when other Roop based extensions were already installed, or errors in installation of insightface models which in my case were all installed automatically without any prorblems... these python apps and free open sources can be a real headache.
Do you have a link to resolve studio reactor?
I think you mean this, you can use the free version, it is really great, www.blackmagicdesign.com/products/davinciresolve/
What if the subject is eating a banna and the video is front view (can’t see the chin). 😂 Will it work for such scenarios too? 😄
:) nope, these 1 image face swap tools are not great at complex scenes, as in your example, or extreme camera angel... such scenes cannot be produced properly without intense training of images... hopefully soon we get better and easy to use tools that allows such training simply and such swapping more effectively...right now, nope.
I USED A VIDEO WITH SOUND, HOW CAN I RESTORE SOUND TO THE CLIP AFTER ITS COMPLETE?
in Davinci Resolve, we drag the original video also to the Time line, remove the clip and keep the audio, this may require adjusting clip speed to match audio length, it is best to look up another video about editing process in DR, it is quick, and will help you learn something new.
@@AI-HowTo i figured out my problem. the clip was segmented @ 30 fps and i didn't change the fps(which was @ 24) when reassembling the video in SD. Thanks for your reply tho
i got rtx 3060 12gb vram but my reactor face swapping takes too much time, at least 1 hour for like 300 frames or so, please HELP!
not sure, this could indicate a problem with onnxruntime installation possibly, or that you are using frames with large dimensions which results in longer processing time, check the frame dimension first, I have not faced this issue before this is why i cannot advise you more that on the matter, i didnt even had any troubles with onnxruntime installatino before... but when it is slow, then there is an issue.
@@AI-HowTo i really don't know and can't find a solution cuz i've tryed multiple command args but none of them reduced my generation time, imagine 9 seconds took 25 minutes (220 frames)
my specs are : rtx 3060 12gb / i7 3th gen / 24gb ram / normal 1TB HDD.
my command args : --xformers --no-half-vae
I think this is a good speed for applying ReActor through A1111, 25minutes for 220 frames ... sometimes the size of the image used in ReActor also may affect the output... but 25 minutes it means it is using the GPU and I dont it gets faster than this using A1111.... anything related to AI or face swap can take a while to produce a decent result and creating a long video could take hours or days depending on the quality and length of the clip.
my video card is RTX 3070GB Laptop , it can swap and generate using ReActor at speed of 15 images per minutes, the source video input image sizes are 720x1280, and i used 512x512 face input image in this test.
@@AI-HowTo thank you for your response and assitance, i think it's okay to be patient to get a 2 minutes video in arround 3 or 4 hours. better thank nothing. i was just asking myself how other famous apps give me results within few minutes even if the video is long, maybe they have multiple computers and they split the input for each computer and re-link the parts later as one video. MAYBE
What's the best way to change outfits?
inpainting, or automatic inpainting, or IP Adapter Controlnet.
it is difficult to do it without inpainting, but it is possible using After Detailer, if you download fashion detection in huggingface.co/Bingsu/adetailer/resolve/main/deepfashion2_yolov8s-seg.pt?download=true and paste it inside >>>\stable-diffusion-webui\models\adetailer folder, restart A1111 to see the model in after detailer models drop downbox.
then in the prompt you put the description of the clothes and change inpainting options (Inpaint denoising strength to something higher than 0.5) ... this will allow automatic detection of the clothes and attempt to change them, but doesnt work accuratly all the time... you can play with the settings and try to see what works best for you....doesnt work well for videos though because some inpaints wont be successful.
Controlnet such as IP Adapter could also be utilized too, but same thing, not very accurate either.
Thanks! Do you have any videos of yours related to this topic? I've been studying by watching your videos.@@AI-HowTo
@@AI-HowTo I did what you asked and the objects for the clothes were detected. However, I don't know how to change them after that. After detection, in the adetailer prompt, I write, for example, red skirt and click generate, but nothing changes. Can you tell me how to do this specifically, and can I use this to generate exactly the same garment (e.g. jeans with two holes on the left side)?
I will publish a video about this as soon as i can, dispaly how clothes could be changed automatically using a prompt, and manualy changes based on reference image... in your case, you just need to increase denoising strength in inpaint option of after detailer to high value such as 0.9, by default it is 0.4 which works great for faces, but not for clothes, clothes automatic changes requires higher denoising strength.
th-cam.com/video/shc83TaQmqA/w-d-xo.html
It doesn't show up and gives me the Error loading script: reactor_faceswap.py
I had no problems installing the package, but some didn, try following github.com/Gourieff/sd-webui-reactor#troubleshooting trouble shooting guide.... or do the installation in a different way, such as deleting stable-diffusion-webui\venv folder which will install A1111 package updates all again, this takes a longer time, but it may fix many problems with A1111 and related installed packages... because conflicts might have happened from installation of other packages that is preventing ReActor to install properly in your case.
when i set the denoising strength to 0, nothing changes @ all
this happens if Reactor is not be enabled.
I donmt get how this is so fast for you. I have same card as you and mine is WAY slower. I used reactor for a whie but I only used it for single picture uses. Now trying batch processing is way to slow. You say less then 30 minutes for 290 files, It take me well over 3 hours for the same. I dont have an error and not sure where we differntiate.
Not sure, there might be some driver issues, or ReActor installation at your end is not using the right drivers or files for some reason, some users have reported some issues with ReActor installation, my RTX 3070 is Laptop card too not PC. true, 300 images should not take that much time .... a possible reason could also be that you are swapping faces in a Large input images such, so it is taking a long time to find the faces in these imagse such as 1920+ image sizes, not sure.
@@AI-HowTo Ah thanks for the input. I guess Ill keep with facefusion for videos for now. I also want to try out how peformant reactor is with comfyui as I have better performance there. Have you tried reactor with comfyui?
nope, but in theory, ReActor face swap doesnt rely on ComfyUI or A1111 inner workflows, because it only applies the update to the image using insightface face library and its own libraries, so its performance is most likely the same on A1111, Comfy and as standalone... i think.
my reactor running but no enable box why?
there could be aproblem in installation, have not seen this before, reinstallation may help, not sure.
@@AI-HowTo reinstalation reactor?
yes, I think reinstallation of extensions is not straight forward in A1111, usually we would delete related folder in the extension folder, restart and install it again
I don't have it anymore either but I think that's normal, since it has been moved, that is to say that it is next to "ReActor" when you check, that therefore deploys it no mistake
@@AI-HowTo i will try
ReActor is NOT building atop of Roop.
Roop and ReActor use(d) the same "inswapper_128.onnx" developed by InsightFace.
InsightFace complained about it. Roop's developer's huggingface page went 404.
check github.com/Gourieff/sd-webui-reactor top left corner, the extension is forked from s0md3v/sd-webui-roop ...
also the stand alone extension at github.com/Gourieff/ReActor-UI also forked from s0md3v/roop
even the developer says that it is based on Roop, and that it is best to disable any other Roop based extensions for ReActor to work well.
so I think what I have mentioned was correct...
ofcourse ReActor has more features and improvements, so I am not trying to undermine the developer's effort, I thank him greatly for all his hard work and making this open source to the community.
They all use inswapper_128.onnx from InsightFace as mentioned, true.
if my memory services me right :) , I remember that InsightFace complained about putting inswapper_128.onnx on hugging face roop project account, as other developers were starting to take it from roop's page and thanking him, not knowing it was developed by InsightFace so, when they found out, they filed a complaint of ownership of the library (I read the complaint back then if i am not mistaken), which caused that problem....but that is another story of hugging face world... Roop git hub page is still active though, the developed mentioned he just stopped developing the extension on his page.... either ways .... the world of development gets messy some times :)
technically it doesn't face swap videos, only images, so Roop is much better in this case
I think most of movie business don't need hire fancy actors because they can replaced by AI.
I dont think is yet applicable, but soon, it might be, especially seeing how fast things are progressing this year.... AI may soon be able to replace large number of jobs unfortunately, and will be a necessity to learn for the remaining ones... if AI is not well regulated by Goverments, it may disrupt large number of business and have more negative effect than possitive unfortunately.
@@AI-HowTo Ye you r right. even most of IT celeb guys always warning to us "AI won't being in safety technology area for HUMAN". getting scare of this. Anyway thank to your nice info all the time. Thanks bro. ah and sorry for my broken English u know Eng is not my mother language but I believe you aleady understand what I mean hahaha.
@@AI-HowToThey said the exact same thing during industrialization, the digitalization and once internet came up and look were we are now. It will be the exact same thing, people complain that we'll lose our jobs but AI will create even more jobs than we had before. The only thing you need to do is adept
You have a strong point, it is best to stay optimistic, learn, adapt, and try to find opportunities in this area which is only in its begining.