Realistic Face Swap with Stable Diffusion | EasyPhoto sd webui A1111
HTML-код
- Опубликовано: 19 окт 2023
- EasyPhoto is an easy way to swap faces in photos! Just install the extension for the Automatic1111 web interface (or ComfyUI), train a LoRA and then start swapping!
In the background, not only does it train a new LoRA, but you also get some tone matching, face shape matching, upscaling and other blending techniques to ensure the perfect photo style.
Uses modelscope, not roop or insightface
== Links ==
ComfyUI - github.com/THtianhao/ComfyUI-...
Automatic1111 - github.com/aigc-apps/sd-webui...
Reposer - 1 image, any style!
• Reposer = Consistent S...
How to install ComfyUI:
• How to Install ComfyUI...
== More Stable Diffusion Stuff! ==
* ControlNet Extension - github.com/Mikubill/sd-webui-...
* How do I create an animated SD avatar? - • Create your own animat...
* Installing Anaconda for MS Windows Beginners - • Anaconda - Python Inst...
* Dreambooth Playlist - • Stable Diffusion Dream...
* Textual Inversion Playlist - • Stable Diffusion Textu... 
Наука
Thank you for staying on top and for sharing! Personally, I find non-realistic faces and styles more useful/artistic, so your Reposer remains the best alchemy, so far.
I totally agree!
Is there ANYWAY you could make your reposer work with reactor? What I want to do is create a graphic novel using a drawing I have created (in midjurney). But so far I can not control the style of the art work. (color of boots, curly hair) in automatic 1111. The change is too dramatic. I am using controlnet. If I do not have full references can your comfyui achieve this?
I would like to mix your instant lora with reposer plus is there a way to do that@@NerdyRodent ?
Discontinued "Roop" or now "React" seems a very much better experience, at least in terms of hardware requirements, because it does not need training. but it is nice to have options, thanks for sharing, great content, great videos, !!!
Yup, LoRAs are great and all.. but that wait time! Reposer is instant and does more than realistic style images, but I guess at least this does face-matching.
Totally inspired, I'm going to start my own now!
Thank you for the video. Honestly, doesn't seem too much better than ROOP.
Oh, it’s certainly better than Roop for face quality. I still prefer Reposer though 😉
Roop is pretty much outdated as it only works with small images. If you want to stick with it I highly recommend ReActor. It's based on Roop but works with higher res images, SD 1.5 and SDXL and it's very simple to use.
@@Elwaves2925I've been having issues installing reactor in comfy it installs but it doesn't appear as as node 😢
@@camilovallejo5024 Not sure I can help you there. I installed it in A1111 as I only use Comfy for certain things.
Do you have Comfy Manager? That is essential and if you do, try using both update options in it, the ones for nodes and the other. Unless you've tried that.
@@Elwaves2925 They all use the same insightface model.
I always like how funny and informative your videos are
Glad you like them!
It doesn't look so much better than just using Reactor or Faceswaplab... Especially considering storage, VRam, time needed. I will pass this time :D
For sure. Reposer for me though!
Thanks so much for keeping up on top of new developments and making content for it! Personally, like others have mentioned, I don't see an advantage here over Reactor or (if you want better results) creating a face Lora Koyha.
They designed this to be better than just a plain LoRA, but personally I prefer Reposer. 1 image, no training, any style 😉
As a big fan of ReActor it would need something worthy to make me use it but the training sounds interesting. I was just typing a comment to ask if you can use the Loras outside the extension when you answered it. Cheers.
Think I'll give this a go on a backup install first.
Cz you are awesome nerdy 🙏🏿
Where does it download Chilloutremix to? Also can I use a Lora I have already trained?
So it basically does the same thing as Reactor does only with a lot more work and struggles. All Reactor needs is one face and it makes brilliant face swaps possible in just seconds if you know how to tweak the settings a bit.
This does high quality, realistic faces very well. Overall I prefer Reposer though as it does any face and style…
@@NerdyRodent you should have used some celeb as an example, so like can this be used instead of dreambooth/lora?
@@wagmi614 it’s LoRA+ loads of realistic extras, such as the face fusion before and after, etc
How disable to upload models ? I want to train at EpicRealism
A1111 is so much better for tutorials 😅
Thank you for tuttorial! But this extention doesn't appear on my page. I'm using colab, so i have just base extensions working when i'm entering it, anyway after installing EasyPhoto and restarting everything - there is no such an extension. Could you give a hint how can i fix it, please?
Install & restart worked for me, but they might have changed stuff. Best to check their GitHub page!
If you have to make a lora anyways, whats the advantage over just inpainting the face?
Accuracy and realism for this one
I'm not sure where I went wrong, tried to install, and have now completely broken my A1111 setup, it now won't even run and throws out about 100 errors!
Video is why we are all here.
Hi there man, thanks for sharing your acknowledge; I couldnt make the training to work, It stops after 100 steps, and the cmd windows says: You have disabled the safety checker for , do you know how to fix it man?
Hello, ive done step by step what ur showing on the video but im getting stuck at training the lora. Once it reaches 790/800 it says Failed to Obtain Lora after Training, Please Check the Training Process. How do i fix that?
Never had it fail myself, so tricky to tell!
Hi please cover Deforum again. since almost a year ago there are a tons of changes and improvement.
Yes, it has been a while!
I want to try that but before training it made a download around 20gb and after when I tried something generate started downloading something again. Is that normal?
Yup
Video is life. I Am alone as the multi-verse itself.
Cool! OK, back to Reposer for me... 🤓 🐁
For sure… all that LoRA training? No thanks!
This literally took all day to auto download some 7mb file that I cant find online anywhere. Why do these models download at less than a kb per second?
Great video though. Just wish I could try it out.
✌️💗🤘 🥰🐁
Peace, love & rock n roll
👍
I wish u could swap bodies so we can copy over tattoos and stuff
I still haven't found anything better than roop yet
But doesn't ROOP do that without training
Sort of. This is a multi-stage process with loads of extras such as face matching, shape matching and upscaling
after enter user id tool getting error
Maybe try another restart? Their github page should have an issues area too
up to 60GB free space 😅
It’s like a game!
@@NerdyRodent 😂
ipadater with reactor seems way more convenient than this semi working addon
It is a bit clunky for sure!
Reactor completely fails with profile views which is a bit annoying. I haven't tested if this works better though. Also how is this for commercial use, since Reactor can't be used for that, with the current model.
Ooof, tools that auto download models never go well for me, and I've learned to hard pass. Besides, Faceswap lab works great for me.
Yeah, not keen on that myself!
Thanks, 60 GB! Ouch, My Commodore 64 😊
Good job I had a bunch of the controlnet models already!
@@NerdyRodent don’t forget the CLIP diffusion and guided-diffusion models
don't forget speccy's! but yeah, c64 ftw! ;)
and in one or two years this will be an app
Just train a LoRa with 8GB VRAM and swap faces easily
I prefer Reposer myself - no training required!
no patreon or something? :D your videos are so good, marry me?
Soon 😉
@@NerdyRodent happy to hear it! but you forgot to answer if you're gonna start a patreon!
@@victorhansson3410 heh 😉
ReActor works much better on a budget for me
Without Video there is only the self.
Installed the plugin, saw the interface and was like uh... duckduckgo nerdyrodent easyphoto
Thank you. I will be skipping this one
Understandable 😉
Just maybe use reactor
Unless you want high quality, face shape matching, etc…