Use Roop in Stable Diffusion for Easy Deep Fakes & Instantly Capture a Models Likeness. No LoRA!
Vložit
- čas přidán 17. 05. 2024
- Let's take a closer look at the extension ROOP in Stable Diffusion. This is much more than a deep fake tool and it can help you get the likeness of character with a single photo! No hours or training or 100's of pictures required...you just need one photo that will serve as an instant embedding / Lora file. This tool is amazing and it's part of my bag of tricks to fix my models faces, get the likeness of a someone without doing textual inversion for hours.
It's also super simple to use and works great! Oh Yeah, and you can deep fake with a few clicks as well. This tool is amazing!
Here is the S0meD3v Github Page for the ROOP Repository (All the links you need will be on this page) - github.com/s0md3v/sd-webui-roop/
Want to try the video deepfake version out? Here is the github page and a youtuber named nerdy rodent does a install tutorial for it - github.com/s0md3v/roop
The Commands you will need
Installing Insight Face
Open up Command Prompt and type:
pip install insightface==0.7.3
Other Commands I used to Troubleshoot:
To install older version of pip - python -m pip install pip==2.2.1
To Check what version of pip your running: pip --version
To Check the Paths (look for Python310/Scripts;): echo %path%
Don't forget! If you get the "none type' error then you will have to add the inswapper_128.onnx model to your Stable Diffusion WebUI/Models/Roop folder.
Here is the link to download that model - github.com/s0md3v/sd-webui-roop/
You will find the link near the bottom of the page
**** BIG NEWS! Stability AI (the makers of Stable Diffusion) release a new version of Stable Diffusion SDXL and will be rolling it out to the public (Non-Paid A1111) users shortly. This is expected to hit the free models in the middle of July! See what it has to offer - stability.ai/blog/sdxl-09-sta...
Videos you might want to check out!
-----------------------------------------------------------------
6 Extensions that you Should be using for Automatic 1111 Stable Diffusion - • Discover the Essential...
Learn more about VAE's, Custom Models, Embeddings, LoRA's here: • Create Amazing Art in ...
Don’t Have Stable Diffusion or Don’t know what A1111 is? Let’s Start from The Beginning
-----------------------------------------------------------------------------------------------------------------------
Stable Diffusion Automatic 1111 (A1111) Install Tutorial - • Updated Stable Diffusi...
Embeddings, Custom Models, and Lora Files: How to make beautiful art with ease - • Create Amazing Art in ...
Control Net v1.1: Stable Diffusions most powerful tool - • Control Net v1.1 is Am...
Want to See my Control V1.0 video? It’s older but I go into a deep dive on what each model is used for - • Master your AI Art Pos...
Got messed up hands? Lets Fix that - • Let's Fix Those Hands!...
Did you know there are sites you can use Stable diffusion for Free? Training, OutPainting, Inpainting, upscales you name it - • Three Amazing MidJourn...
Do you need to upscale something to Monstrous resolutions so you can Print them on a massive Waifu billboard? • Ultimate 8K Upscale wi...
6 Extensions that you Should be using for Automatic 1111 Stable Diffusion - • Discover the Essential...
UltraSharp 4x Licensing Info - huggingface.co/stabilityai/st...
Developed by: Robin Rombach, Patrick Esser
CreativeML Open RAIL+±M License. This license allows for non-commercial use, modification and redistribution of the model, as long as attribution is given and any changes are indicated.
creativecommons.org/licenses/...
Chapters
---------------------------------------
Times
00:00 This is what ROOP Does
00:54 Let's Install this ...but first the PreReq's
02:12 Let's Run the insightFaceCommand
02:55 Trouble Shooting pip not recognized (Skip this if you have no errors)
04:32 Trouble Shooting None Type Error - inswap model (Skip if you have no errors)
05:02 Let's Install the Roop Extension in Stable Diffusion!
05:40 Where is Roop Located
06:57 Let's take a look at Roop's settings
07:16 Selecting the Correct Face w/ Comma Separated Face numbers
07:54 Dev's Recommend 0.1 for Denoise
08:06 Restore Visibility (GFPGAN & CodeFormer) and Why you need it
10:25 Easy Face Fix or Mesh with InPainting
11:14 InPainting and Face Selection
11:47 Why You Should Use InPainting for Roop & Examples
13:41 Only Mask Bug
14:12 NSFW Not Allowed?
14:22 Other Considerations
14:28 Swap Source VS. Generated (Have no Clue...) & Testing
15:05 Img2Img Affects Entire Img
15:19 InPaint is the Way to Go!
15:47 Sample DeepFakes w/ Music!
(Song - Mansij - Passionate Choices)
#StableDiffusion #AiArt #a1111 #DeepFake #Roop - Věda a technologie
Roop is amazing! All you need is 1 image and you have instant embedding or LoRA results. You can DeepFake with it or use it to copy the likeness of yourself to finally turn yourself into a superhero without complicated training processes or textual inversion. 1 Pic is all you need. Here is a link to the S0meD3v's site and it will have mostly everything you need github.com/s0md3v/sd-webui-roop/
All links are in the description including the commands that were used to run the tool. There is a Troubleshooting section that is about 3 minutes long that you can just skip and deep fake samples are at the end of the video. Just use the video chapters to jump to any section you like.
Can we run this on runpod?🤔
Hi there, I never tried and TBH I had to look it up RunPod , my understanding is it's cloud hosting platform. I wouldn't see why it wouldn't work but take that with a grain of salt since I had to look up what Runpod even was@@Ai_Vs_Original
I installed roop on stable diffusion, it says installed also but I'm not getting the roop option
Its the really hard why no one talk about this. Its a big unsolved task.
But have you installed it?
Ohh good tip about that NSWF part.
Best explanation to install I found!!
Good guide - thanks
Is there a way to use this tool in PaperSpace?
I love how you explained its potential beyond just face swapping; it's great to see tools that can improve AI workflows.
It works great! you can actually have consistent characters by reusing a previous characters or fixing a face that just wont budge.
4:04
You have 2.2.1 in your description, instead of 22.2.1.
Hi!
At 14:30: The 2 checkboxes "source"/"generated" means applying roop before and/or after SD pipeline...
It depends on the needed resemblance you need.
So applying before denoising and not after, and a strong denoising the face can be changed (imagine you want an orc with 2 big teeth out of the mouth, but looking like Ross, you need SD to change the rooped face).
You can see the result of roop applied before in init-images folder ;)
Thank you for this video. I was having some great results using Roop, as well as some frustrating results where the faces just weren't matching close enough to be usable. Now I have more knowledge and techniques to try to resolve the issues. Much appreciated
Awesome! Glad to hear it. I kind of use it to cheat to fix faces now. If you tweak it enough you get a combo of 2 faces so it doesn't look like all my characters are scarlet Johanson or someone famous.
@@ArtificallyIntelligentAi If you copy the Roop directory and rename each copy to, say, "Roop2" and "Roop3", you can use it to change as many faces as you want. Just have the first face on the left as "0", the next as "1", and so on
@@theguru143 Oh cool trick. Wish you told me this before I made the video 😄Maybe in a future video.
First time I learnt about saving prompts as styles, thank you I wish i knew this earlier. lol
It kind of just appeared on one of the updates. The cool thing is there is an excel or csv sheet called styles in your stable diffusion folder. You can look at all your styles in there and edit them there as well.
Help...! How can I keep the detail when swapping face with "Roop" extension.
Whenever I try to swap face, it renders a clean face.
Is there a Lora or anything else that i should use.
Can this be used with Vlad Diffusion? I have installed the extension and it shows no model although it is downloaded and placed in its roop folder
Thanks dude appreciate this detailed breakdown
No probs have fun with it.
This will be so good for creating consistent characters.
That's the first thing I thought when I saw Roop. I wasn't so into the deepfake stuff but using 1 picture to get consistent characters or a particular look is pretty awesome. I think the dev's main goal is consistent characters. This is a quote from their site "It will help artists with tasks such as animating a custom character or using the character as a model for clothing etc."
@@ArtificallyIntelligentAi I haven't really been able to get it to work for really stylized images though, it seems to only try to build a realistic or semi-realistic face, regardless of the source face image. Have you found a work around for that, say with a cartoon type image face swap?
Any experiments with batch roop for animation in SD?
That is crazy! Awesome and terrifying.
With Ai, it's always like that. When i first saw Chat GPT, I was like... That's AMAZING! then I was like but it can destroy the world.
Went through the instructions step by step. But cant see 'Roop' section in a1111 ui
Great and awesome video
Great comment 😀and appreciated!
@@ArtificallyIntelligentAi 😅
I just installed roop extension from inside a1111 and installed visual studio. Roop is working fine. It downloaded inswapper onnx files and some buffalo onnx files on its own.
I never ran the pip install insightface command. Somehow its still working.
Wow cool, they might of have added to their dependency requirements for the extension so it does it automatically. There has been a couple updates since they released it, but either way have fun with the tool.
me and my friends are having so much fun playing this deepfake... its crazy how it can manipulate with one sample photo
Swapping faces with your buddies I see lol. Ya it is crazy. You guys all should swap faces and then print it out.
i installed the roop by following the video but its not showing in stable diffusion i tried all the solutions plz help me i got stuck for my work plz 😨
Thank you
un dès meilleurs tutoriaux que j'ai vu sur roop. le petit plus quand nous sommes debutant, les messages d'erreurs corrigé dans la video !!! merci
Merci pour le soutien! Je suis content que vous ayez aimé la vidéo. J’utilise google translate en passant, donc si cela semble drôle, c’est la raison.
roop can't do NSFW without editing some script. The NSFW filter sucks, even some of my SFW img got filtered too sometimes, And its development has ended.
currently using 2 extensions for face swapping, ReActor for Stable Diffusion and faceSwapLab, same as roop but faster and enhanced and importantly no nsfw filter.
Just use in-painting on the face instead of doing the whole body would work but it was intended to block nsfw yes
Roop does not appear in the list
What sampler is faster than DPM++ 2M Karras but good?
That's a good question and probably an even better video. I don't see a lot of content on that. I do visit this site often stable-diffusion-art.com/samplers/ and there is a comprehensive guide. Everyone uses DPM 2M for the final render. Better is a bit subjective but UniPC is a lot faster and sometimes produces great results but not better than 2M. Maybe sometimes it is? give it 2 weeks when the SDXL Models come out and we going to have to figure everything out all over again.
@@ArtificallyIntelligentAi I was recommended DDIM… I will test it out tomorrow.
@@ATLJB86 Ddim and the sde’s will give better quality potentially (sometimes to sharp) but I think they are much slower unless something has changed.
@@ArtificallyIntelligentAi I tried DDIM and it is faster by 1-2 seconds but not worth the drop in quality
Whoever seems to think that any output you create using this extension requires explicit attribution is probably mistaken. The ONLY mention of Output restrictions in the file linked to is this:-
_"The Output You Generate. Except as set forth herein, Licensor claims no rights in the Output You generate using the Model. You are accountable for the Output you generate and its subsequent uses. No use of the output can contravene any provision as stated in the License."_ That refers to the previous line which essentially says you aren't given permission to do anything illegal with it. The attribution and repurposing of the MODEL is what requires careful consideration. You can't just redistribute THAT, or derivitives, without attribution and adherence to their conditions. Not your images.
The issue at 13:41 is that you masked only one face, but then chose "1" in the Roop settings. You want to choose '0' in Roop when you only have one face masked regardless of how many overall faces are present. Cheers.
Thanks! Good to know. For some weird reason I was able to force it using the in paint whole picture. I found the way I did it prevents smoothing out the other face on the left, which gave me a bit of trouble trying to chase that back in forth doing it the proper way.
How can this be used with a collab folder
Hey, they actually have a few colabs out there for the video deep fake version of roop but not so many for the img2img version.
roop crashes whenever I try to use with inpaint
🤔 . Was is a none type error or did stable diffusion just close? If it’s that error check out 04:34 of the video or if u did add the i swap just double check it’s in the correct folder. It can be conflict with another inpainting extension as well. Also make sure one of swap options or both are check marked at the bottom. Swap in gen or source img.
I don't think you have to do both: If you install via the extension tab, you don't have to install using PIP.
Interesting. The instructions on their site must be outdated. I had the issue when I first installed it and I got an insightface error. Maybe they fixed it?
I'm not able to download inswapper
I too
i am your saver
drive.google.com/file/d/1eu60OrRtn4WhKrzM4mQv4F3rIuyUXqfl/view?pli=1
sr not workong url.
use that => drive.google.com/file/d/1eu60OrRtn4WhKrzM4mQv4F3rIuyUXqfl/view?usp=drive_link
@@pokezaur do you know how to uninstall extension?
@@kulwantsingh2155 hi , you can go into your stable diffusion > extensions folder when I A111 is not running and Delete the folder. Alternatively you can disable it by opening up stable diffusion and then clicking on the extensions tab and then unchecking the roop extension, then it won’t load until you check mark it again and restart.
The nsfw part is pretty funny they coded it that way. Easy workaround with inpainting. But annoying they did that
Ya I thought about that as well since the inpainting only looks at the face not the body. I didn't want to say it in the video though, but it's pretty easy to figure out.
roop uncensored version has been around for a long time
@@abzaccc u naughty boys 👿I think those versions are out side of a1111 tho, but ya like inpaint will do the job
@@ArtificallyIntelligentAi i use on a1111
It's easy to uncensor Roop :)
You lost me from this point forward 4:52
Roop not showing up 🥲
same