The NEW KING of Face Swappers! Reactor Tutorial. Roop is dead.
Вставка
- Опубліковано 5 жов 2024
- This content is educational for academic research purposes only.
Roop is dead. Long live Reactor. The new face swapper for changing faces. Works with CPU only!
For a1111/SDNext/Vlad github.com/Gou...
For Comfy github.com/Gou...
Prompt styles for Stable diffusion a1111 & Vlad/SD.Next: / sebs-hilis-79649068
FREE ComfyUI workflow for 1.5 models: / comfyui-1-5-86145057
FREE ComfyUI Workflow for SDXL: / comfyui-workflow-86104919
Get early access to videos and help me, support me on Patreon / sebastiankamph
Chat with me in our community discord: / discord
My Weekly AI Art Challenges • Let's AI Paint - Weekl...
My Stable diffusion workflow to Perfect Images • Revealing my Workflow ...
ControlNet tutorial and install guide • NEW ControlNet for Sta...
Famous Scenes Remade by ControlNet AI • Famous Scenes Remade b...
Get early access to videos and help me as a Patreon supporter www.patreon.com/sebastiankamph
Hi man, excellent video and congratulations on your work. I'm just starting out with Forge and would like to know if it's possible to change just parts of the face, just the eyes, mouth and nose. If possible and it's not too much to ask, make a video explaining how to change just parts of the face. Big hug from your subscriber in Brazil!
If you use sampler DPM++ 2M SDE Karras with the denoising strength at 0, it will give you an error, set denoising above 0.04 and you will be fine, just so you all know. Very informative video, like all your videos; no gimmicks or bs, you are a great tutor, thank you.
I laughed more than I should have at your little joke at the beginning which i've heard a thousand times before. Must have been your delivery. :)
Hah, glad you enjoyed it 😁🌟
Aas a non native English speaker, I really appreciate the addition of subtitles. Keep up the good work, Sebastian!
Please make a tutorial on this for ComfyUi !
FaceSwapLab is more versatile, you can 'train' faces and saving it with safetensor file for special person name , you dont have to reupload or change face any time when using it.
Well that completly broke my Automatic 1111. Thanks alot.
Thanks a lot, loving it. My 2 cents: you need at least 1 model inside insightface folder for this to work!
I've used Face Swap Lab a lot these days, with good results. It allows training faces on multiple images (I get really great results when I have video of someone showing them at various angles) and has batch processing capability. Is there any way in which Reactor is superior to FSL?
The face swapping model is the same for both (and Roop as well), so the quality of the result should be similar. I haven't tried Face Swap Lab or ReActor, so I can't say anything about the usability of either.
@@phizcI can confirm that they use the same technology and the results are pretty much the same, but I find FSL to be more configurable and I love that you can train the faces and get a little better results
Yeah I'm confused about why he is saying the resolution is greater when it's the same tech?
@@phizc
FaceSwapLab has an added advantage, video output. No more weeks of training to watch something in order to get off. 😸
Faceswap is still better, you can train from multiple photos creating a normalized face, you can play with pre and post processing, you can include in animations, upscaling, and a long etc. Yes the all based in the same technology though.
Unfortunately with all these face swappers, they are all still "bottlenecked" since they all are using inswapper128, which is 128x128 training data only. Any larger models are yet to be seen since the original project was stopped.
Inswapper128 is fine for the general facial look, but loras are still the way for very specific facial/skin features.
so how do we fix that?
@@iamxenobyteuse simswap
@@TejasviSharma-t8j simswap outputs higher quality but the result is shittier and barely looks like the input face. Experimented with that one a lot but didn't like it as inswapper_128
@@eyoo369 true, insight face inswapper is really the best. The team behind that model are some mathematics geniuses. Sadly they'll never open source their newer models, i wouldn't have either if i could make money using it, so i don't blame them.
Yup. I have been using Reactor for a while now and it worked perfectly fine with 1.0/1.5, but as soon as I try to use it with SDXL it's horrible. The faces are super ugly and no matter the settings I just can't get it right.
Amazing as usual Seb o/ ! Thank you and have great weekend everyone =)
Thank you very much, Marcelo! I hope your weekend will be excellent too 😊💫
thanks for all your great efforts, sebastian. how about a tutorial on how to use SD extensions Nextview and Reactor for face swapping video?
Great suggestion!
i did the steps but reactor does not show up. only in cmd is as in your vid (reactor status), in SD don't appear
It's a beast of a tool,and scary how good it is,newest version bring precision face masking option,so you can just tick the box faces are super clear even at highest resolutions.
With best photorealistic models for stable diffusion and correct prompting is super hard to saw a difference.
Even at crazy angles faces stay correct so you can just create same person again and again,without need for embeedings or LORA-s,it is just amazing,
this was interesting, thanks, definitely upper hand on roop. 😀
why is reactor not showing in my stable difussion? when following install?
I have this problem too, did you find any solution?
i have a similar issue on forge ui
Mona lisa was 44. Thats an awesome information to make a video dating people in old historic paints.
Yeah, pretty sweet feature if only for fun right now.
I get an error in CMD so A1111 doesn't even start up. But others seem to think that it's identical to Roop including the resolution, so I don't understand, have you even compared the two? Saying that it's now the King of Faceswappers is quite a bold statement.
I agree, it clearly uses the same version of insightface v0.7.3. So not sure why he's claiming this to be superior to Roop
@@BabylonBaller roop is no longer in development, reactor is a continuation of the sd-webui-roop, there're a lot of fixes and improvements were implemented for these months by the reactor's dev
It's working exactly like roop uncensored, how is this better?
I'm kinda suprised I'm just now seeing youtube videos about reactor, I've been using it for the last 3 or 4 months, great video as always though!
Use IP-Adapter alongside Reactor/FSL for consistent characters
I'm getting an error, it just doesn't do anything... what might be wrong?
Error running postprocess_image
I did install the c+++ tools, but only the ones required...
Im having the same problem :(
I'm getting a lot of trouble with it in ComfyUI. Have you had any luck installing it into comfy?
hi i've followed your instruction and i have that : AttributeError: 'FaceSwapScript' object has no attribute 'save_original' EDIT : found the issue : in the model folder you need to create an insightface FOLDER where you put the insightface file
Where do you get the insightface file?
which insightface file? i have same error.
@@mohitsadhu2938 you have to create it yourself
I dont know why everyone praise Reactor - I tried to use photos of many people and the result is unrecognizable. Roop is still better regarding the similarity of faces hands down
Agreed
Isn't reactor a straight fork of roop? It's the same code, based on the same 128x128 face model, just with extra features on top
The best one isn't Roop, or Reactor, it's Rope, it adds a shit ton of features over Roop like iterations strength, face angle control, background, mouth parsing...
I added the reactor but it does not appear in the interface, what should I do?
Had the same problem "*** Error loading script", solved it by installing Insightface (google "reactor extension won't appear reddit" for detailed steps).
@sebastian Kramph, I don't subscribe to many people on UA-cam, but I am always pleasantly surprised when you upload new content. Thank you for sharing!
roop still has better results when you tick both boxes so it applies face before and after in rendering.
hopefully reactor get better and that option
reactor has this option as well
roop is very buggy
I have 2 separate Roop modules installed. How would it know to apply a face before and then after?
Thanks a lot. I found your video much easier than other guys. Its more personal and not like others sounding way to professional. Today i succesfully installed Reactor. Previous day i struggle and confuse following others about how to install Roob and Reactor that i ended up installing stuffs in Visual studio, Python, C++ environment and API desktop Development = 12GB 😑 Should i uninstall if i don't require them?
roop unleashed is still developed.
wow..it worked like a charm... now I wonder if you can combine controlnet with it
Hello, can you please tell me if it is possible to replace the face, before applying lora. For example, first replace the face, and then stylize the picture as a drawing?
Yes, I'd run the resulting face swap image through an img2img or ControlNet pass if you want it in another artstyle. The face will probably not stick as well as the initial face swap though
Thank you. I have to balance between portrait likeness and creative processing. Have a good day.@@sebastiankamph
Hi, I ran into a bit of a hick-up that I hope you can help me with. The Reactor face swap downloaded correctly. I think. It shows up in the terminal as running. But it dose not appear on my SD UI. As it did in your tutorial. Is there a check box in settings that needs to be checked to make it show up on my UI. I hope you can help. I really have learnt a lot from your UA-cam cannel. Oh yes I did restart Stable Diffusion to see if that would help. Unfortunately it did no appear. Here's a couple of DAD jokes as a thank you. I hope you have not used them. 1. I used to play piano by ear, but now I use my hands. 2. How do you organize a space party? You "planet"!
I have the same issue after installing reactor. I think there is an installation bug right now. Does anyone know of a fix?
I'm glad you had the same unfortunate installation problem as I did. Being new to ai art I thought I had done something wrong. Lets hope someone can assist us. Its such a simple and effective face swap I can't wait to use it .@@MrPlasmo
@@59Marcel found the working fix until they fix the install- do this and it works (thx to mudman13):
Have you done this?
Close (stop) your SD WebUI Server if it's running
Download and put prebuilt Insightface package into the stable-diffusion-webui (or SD.Next) root folder (where you have "webui-user.bat" file)
From stable-diffusion-webui (or SD.Next) root folder run CMD and .\venv\Scripts\activate
Then update your PIP: python -m pip install -U pip
Then install Insightface: pip install insightface-0.7.3-cp310-cp310-win_amd64.whl
Enjoy!
Both Roop and Reactor deliver fairly similar results for me... Can't say one is better than the other...
They use the same base model, but reactor now is continually developed.
You're just a cool dude. I have a subscription for this extension
I use this to face swap videos. Use Adobe Media Encoder 2020 to slice the video into a jpg of each individual frame into a folder, then in the img2img tab use ReActor to batch each file from this folder into an output folder, then put the frames back together with the media encoder to make the movie with the swapped face
You can use FaceFusion to do that automatically. One limitation is that you can't use a mix of different face restorers, it's CodeFormer or nothing IIRC. I'm also not sure if you can choose what upscaler you want to use.
@@phizcany tutorials for this? Id love to do videos with faceswaps
There are tutorials for each of those steps out there that are pretty reasonable to follow.
Hi, could you please share those tutorial pretty please@@alexcamilli129
stupid inefficient method
Oh no the soda joke got me at work lol
At least it brightened your day at work, eh? 😁
@@sebastiankamph Definitely :)
im having problem. It says that Reactor is running but when I check txt2img, reactor is not there. What should I do?
*** Error loading script: reactor_swapper.py
Traceback (most recent call last):
File "D:\STABLE DIFFUSION\stable-diffusion-webui\modules\scripts.py", line 382, in load_scripts
script_module = script_loading.load_module(scriptfile.path)
File "D:\STABLE DIFFUSION\stable-diffusion-webui\modules\script_loading.py", line 10, in load_module
module_spec.loader.exec_module(module)
File "", line 883, in exec_module
File "", line 241, in _call_with_frames_removed
File "D:\STABLE DIFFUSION\stable-diffusion-webui\extensions\sd-webui-reactor\scripts
eactor_swapper.py", line 10, in
import insightface
ModuleNotFoundError: No module named 'insightface'
---
08:20:12 - ReActor - STATUS - Running v0.5.0-b1
same here! looks like we have to download the model, but I don't see where to find one, did you had any luck?
Thanks Seb for sharing this! ❤
Thanks!
Can't find the model
Reactor-Force, for video cards with 8Gb or more VRAM works faster. However, working with profiles of faces, and or anything approaching profiles does not work too well. Also, it is unclear how/where you put multiple replacement faces in the current interface. Thanks for the tutorial !
it is in the video but the git page has a better explanation. The faces are read from left to right, then top to bottom. starting as 0 and progressing.
The same goes for where the face is swapped too. If you want to swap multiple faces you input multiple numbers separated by ,'s ex 1,0,3 4,0,2. In my example the 2nd face in the picture would swap to the 5th. then the first to the first, then the 4th to the 3rd.
If I'm putting denoising strength to 0 I'm getting the following error "UnboundLocalError: local variable 'h' referenced before assignment"
I get that sometimes and unfortunately I have no idea what the hell that means.
it look like as of 10/17/23 today A1111 now can install ReActor directly from the Extensions tab
unfortunately it uses the same not entirely free and low resolution model
Who has not succeed with installing it, just try again, the installation script of the extension had been updated earlier and got some errors, now it's fixed
@3:00 information for thoose who could not get it working after "Apply and restart UI" , somehow my Disable all extensions was picked as " all " , now go there very right side of the "Apply and restart UI" button and click none
Is it possible to face swap without changing the mouth and without using inpaint?
The goal is to face swap a whole video or gif using the Batch window. But If the mouth is open, then frames with the open mouth will be ruined and I don't want to go frame by frame and manually fix the mouth with inpaint.
absolutely incredible.
This is definitely better than roop. Reactor works with all image unlike roop. I used to only get 60% detected rate with roop i got 100 with reactor. Thanks sebastian this is why im a subscriber!
Nice - it works - thank you
Clickbait!!! This software uses the same model under the hood as roop. No changes at all.
Scary Alita : third face swap :D
any ideas as to why when you swap faces with an illustrated face say, the image comes out blurry, you can still see that the face swapped perfectly but the output image on the face is very pixelated?
Face restoration isn't working would be my guess.
I can't say for sure as I haven't opened 1111 in months since discovering ComfyUI, but as far as I know all of these face swaps are at 128px, which are then processed with face restoration (gfpGAN, codeformers, etc...) to produce the result. If the face restoration fails you will end up with the unrestored raw face swap which is much lower resolution.
One work around would be to simply produce images where the face is smaller in the frame (face
why cant we simply raise the resolution ? @@ArielTavori
What sometimes works better for me is first generate an image. Then copy to im2img inpaint.
Then inpaint the face and select paint mask only.
Then below that enable faceswap, select the Gfgan and then generate.
This gives me often better results than when swapping during initial txt2img generation.
Probably obvious but this is intended for still images...not video. Can be done by feeding image series to it, but likely not the right use case for this particularly software
Yes it can be done, but is better if you use Face Swap Lab
Hey I'm just here for the soft drink joke.
1:55 dude did you just install the full version of Visual Studio Community?
That's for developers 😆
I know because I use it 🔨
Great content thanks Seb!!
Thanks!
Wow, thank you! Very kind :)
I'm only getting Errors with this extension. Sticking with Roop for now.
this tool is frightening. Great tutorial
Thank you! Use responsibly 🌟
absolutely. It's just crazy how it gets everything right down to even the overall colour of the image and even the age detection is scary close @@sebastiankamph
What if there is an object in front of the face that I would like to keep? like a hand for example.
Roop has been dead for months, I have heard of reactor once or twice but I use faceswaplab which by the looks of this, faceswaplab has more features, am I missing something?
Thanks for the vid though!
it still use 128X128 model like roop. not all the time the upscaling keep the face right.
I tried roop when it surfaced. Couldnt get it to work. Too many pain in the arse dependencies required. Tried Reactor, worked out of the box.
Great new content! Thanks, can't wait to try this. 👨↔️🧔
Hope you enjoy it! Let me know how it works out for you
Somehow the install script failed and also broke my roop configuration, too. Which sucks, because you can't reinstall roop anymore. Luckily I had a backup of my entire automatic1111 folder.
Could y ou make us a tutorial of how to make multiple faces? Because I've been trying following your general guidelines here, but it's not really working out properly, either some faces change or none do, not all as I would hope (using face 1, face2 , face3)
The first face is 0 not 1 (1 would be face 2)
Hi Sebastian, thank you very much you explain very well, I wanted to ask you why the enable section disappeared from me? Thank you
can you make a good tutorial for Lineart to Color? your channel seems like the best but I can't find anything. Just outdated stuff with old control net.
the "enable" check box isn't show up, how to fix it ?
Awesome 🙂
top !!! thank you mate !
after doing word for word it just didnt install for me. I guess insightface has to be installed too?? Thats the problem with this "new software"...its just not very user friendly. I cant wait till they develop it to where the average person can use it without having to be a PC programmer
I did everything you did and it says ModuleNotFoundError: No module named 'insightface' Any tips?
Try opening cmd and type 'pip install insightface'
@@sebastiankamph Didn't work and at this point I wonder how to uninstall the attempt because it slows down start up times 10X.
i keep getting this message "arrow::fs::FinalizeS3 was not called even though S3 was initialized. This could lead to a segmentation fault at exit"
Gives me this error: ModuleNotFoundError: No module named 'insightface'
in cmd I did run "pip install insightface" at C:\Users\USER\stable-diffusion-webui and the error msg no more. However no Reactor to be found at interface. Deleted the extension. Restart. Installed it again. Installation was much-much longer. Restart. Reactor appeared. Testrun. It automatically downloaded buffalo_1 to models/insightface/models/. And working.
@@nermal93 after type pip intall insightface i get this note " ERROR: Could not build wheels for insightface, which is required to install pyproject.toml-based projects
[notice] A new release of pip available: 22.3.1 -> 23.3
[notice] To update, run: python.exe -m pip install --upgrade pip "
how to fix it
I actually compared Roop and Reactor with the same input image. The output is IDENTICAL, so Reactor basically IS Roop!
They use the same base model, so if you use default settings I'm sure you're correct.
Both use(d) inswapper_128.onnx
Does it have trouble if your source face is straight on, but your target face is at an angle?
Guys... I do not want to bash ideas but you should just use faceswap, ipadapter plus face and control net depth. Thats it. exact high res copy of faces. I realize you guys are not ware about that.
hey do you need webUI 2.0 to do this? havent updated mine since 2022 coz all of my LORAs and models work for 1.5, but this reactor thing looks dope
Bummer. Sounds like it would only work on Windows?
Question: If I'm using a Lora in conjunction with Reactor, let's say an Oil Paint lora. The lora will make everything with an oil paint texture, buy the face swap will not have the oil texture effect, how do you tell reactor to include the lora effect? I'm guessing Img to img or control Net? Any chance you can create a tutorial? It's a question I personally get asked alot but have no solution for it.
Can u show us the comparison of reactor upscaler
Yep ReActor is awesome. The guy who made Roop shot himself in the foot with his holy morals that he took his ball and went home. RIP Roop
😂
Does this do video or is Roop still king for easy video swaps?
Great work! But how to use Reactor face swap by code? Since I am a software developer, I want to use it in my product.
unfortunately, that makes my SD not launch anymore. there're a lot of error messages in the window ("descriptors cannot be created directly"")
even after removing the reactor folder in extensions, it doesn't launch anymore. edit: even with an empty extensions folder... what the f ?!!! I don't understand...
I shouldn't install that visual studio
I'm trying to swap a face in image , but it gives me the same image why ?
it kills nothing at all, is far from exceptional and does not justify such a glowing title
Agree. I tried it and still get better results with the old roop.. people mentioning FaceSwapLab, maybe that's better.
I use it in ComfyUi together with Ip Adapter. Quick and dirty one image LoRA
Sounds great, could you please share your comfy workflow, i have tried the Olivo's tutorial and the result wasn't great for people , pretty, please.
What should I do if, when generating images, a long fringe covering the eyes results in the forehead appearing darkened?
I hope this one is better than IP adapter, that one turned out to be a disappointment
Judging by the results in the video, I wouldn't get your hopes up.
Are there updates in your patreon? im coming across alot of missing modules that i have to download through cmd. Its not playing nice with reactorRT? maybe thats the issue disable that for now>?
top top as always seabass..
Thank you kindly! 😊🚀. Rate today's joke 1-10
my reactor Running v0.7.0-b7 on Device, but no "ENABLE BOX"?
i get these errors...
ReActor preheating... Error: The 'onnxruntime-gpu' distribution was not found and is required by the application
Device: CUDA
Error: The 'insightface==0.7.3' distribution was not found and is required by the application
Error: The 'onnx>=1.14.0' distribution was not found and is required by the application
Work with videos like roop? And with colab? Thx.
Hi i love your videos!, I am having a little trouble , I have installed ReActor but its not showing on my img to img tab?! please help.
are you running pip 22.2.1 or 23.3.1