Fooocus Face Swap With Ease!
Vložit
- čas přidán 16. 05. 2024
- Today, we're diving into how face swapping is done in Fooocus. It's incredibly simple to do, and I've got some pre-generated images to show you just how easy it can be. With a click of a button, you'll be able to swap faces very easily. Get ready to be amazed and don't forget to subscribe!
Fooocus is a free stable diffusion image generator that can be run on any PC with at least 4GB of VRAM (GPU) and it's the easiest to run SDXL locally.
Download Fooocus here github.com/lllyasviel/Fooocus
7Zip to extract the files www.7-zip.org/
⏲Time Stamps
0:00 How to use face swap
2:10 Using controlnet canny (edge)
3:16 Does it work with real photos?
4:43 It may not work for some of you
*Disclaimer Affiliate Links Below*
📸 Gear I use
Sony A7C CAN amzn.to/3spWX8C
Sony 35mm 1.8 CAN amzn.to/36Apekr
OBS (FREE) obsproject.com/
Editor Davinci Resolve www.blackmagicdesign.com/ca/p...
Audio: Rode Podmic amzn.to/35sSnxv
🎵 Epidemic Sound
🔦 Find us on:
Discord: / discord
Instagram: / monzonmedia
Facebook: / monzonmedia
Is there anything in Foocus you want me to cover? Maybe a more in-depth look at inpainting?
How to combine 2 photos in one like a girl's pic and a boy's pic make it look like real ... that's a great topic you can cover like selfie and portrait
Excellent video!!!, maybe you can cover the *errored out* problem that a lot of people are experiencing, not limited, but mostly using 4gb VRAM. I can’t figure out what’s the problem, everything runs fine until you hit start and a couple of seconds after that the model just pause and show the errored out.
Sd1. 5 level realistic images in sdxl?
Is it really possible?
I have tried and it looks like too photo retouched airbrushed look.
No realism? How do you get it?
How do I retain the same face but change expressions? What if I want the image to be of someone winking or sticking their tongue out?
@danrazART I cover it here! 👍Works really well! czcams.com/video/1x_WTbRU55M/video.htmlsi=8bVlrkJwCJ1LwK5d
Please continue making these little tutorials, they help immensely. I'd like to see ways to get prompts more detailed to direct the generation better.
My friend, I've had quite the backlog of videos to edit, especially for Fooocus and I'll be catching up on them. I have a video planned for your specific question too! In fact I have one today I'm working on for photorealism and using SD1.5 models. 👍 Should be published later today!
@@MonzonMedia ill be on the lookout for it! Thank you!
wow, a great quick paced video with tons of info, it seems most open source youtubers now will drag on for 10 minutes ahem but I apreaciate you got straight to the point!, please make a tutorial on how to make your own SXDL models to have full control of the model and eliminate "bias"
I appreciate the encouragement! I like to get right to it so, you can expect more of it from me 👍. Fooocus can't train your own models, it's not that robust and something like that requires at least 24GB of VRAM on the GPU. All fine tuned models have a bias but if you want something less bias than it's best to go with raw SDXL or even SD1.5. It just takes a bit more creative prompting though.
am good on GPU front, I've used koyha for SD 1.5 loras but since rate of updates in the SD scene so fast I haven't seen a good sdxl tut unless they have a ridiculous patreon paywall or poorly are explain and dont allow users to deviate from a specific look or style
Thank you, very lucky you uploaded this on the day I'm working on a school project, thank you so much good sir!
Glad to help! I hope it's not due tomorrow! hahaha! 😆 curious to know what you're working on?
Thank you so much for this. It's mind blowing. Quick question for you...
I'm trying to generate an image of me holding my company's product, which is a tea box with the brand and logo on it.
How do I generate an image such an image?
I know I can use the face swap for my face but for the object, I really have no clue how to do it. I know I can use the image prompt for the product but the images generated won't be have my exact product.
Appreciate your help and thank you so much again!
I noticed that the refiner messes up the face swap. When I had it enabled, it generated the correct face at 30/60 steps, but at step 31 the face got swapped for a random one. That was with the „realistic“ model.
If I could request a video: More tips on having character consistency with focus. Example: I was trying to generate a Christmas elf doing all sorts of things, but the elf looked different all the time.
Bonus challenge: Inpaint a consistent character into a pre-existing foto (i .e. you have taken a picture of a room or a landscape and you try to insert a premade character into that photo)
Thanks for you helpful videos!
Yes because you have to set the refiner higher to at least 0.7-0.8 to retain the face however it's a bit of a trade off in realism. I'll do a video on consistency but generally speaking it's very difficult to do without training lora's but there are work arounds to get it to a point where you can do some post production.
another vote for this video idea!! @@MonzonMedia
another another vote for this video idea!! @MonzonMedia
@@MonzonMediathanks for that tip dude, your videos are great! from all of us here thank you!
Yes please this is a good suggestion.
Nice one bro, ang galing galing 😁😁😁
I'd like to know what the options for the AI using reference images are, what do they do, and how to best take advantage of them (ImagePrompt, PyraCanny, CPDS, Faceswap). All I've ever seen anyone mention is the PyraCanny to use as reference for pose/posture. And FaceSwap for... well, faceswapping.
Also, what do the parameters mean/do? "Stop at" I see it's a percentage of where the resemblance engine should be halted in favor of a cleaner more realistic result, but "weight"?
It would kind of seem to me that you can put in 4 reference images for faceswapping and give different "weight" loads to them so that the ai picks from multiple, but playing around with fooocus, I didn't get the expected result so I have no clue if this has any truth or meaning.
Could you do a follow-on video addressing how differences in the reference image effect the Faceswap? For example, does using a tight portrait give a better result? In your example you used a 3/4 view and got just ok results.
Maybe character stability with fooocus may be great !
Thanks for your work
I'm actually working on something to see if it can work in Fooocus. Will share my progress soon! 👍
Awesome, very helpful! Thank you. Do you have experience with the other algorithm CPDS? What is it used for?
It decolorizes the image to make it sort of like the "depth" controlnet. I'll be doing a video on it soon.
Fooocus simply good, which has everyting for avarage user or beginner. Everything you need included, no need costum control models, upscaler, extentions. if you could do inpainting in fooocus would be great. I believe soon the developer will include SDXL Turbo. LCM already hidden in fast generations section, which is works fine
Agree 100% it's really a joy to use, although limited in features, it's more than enough for most. You know you could run the SDXL turbo fine tuned models, just put them in your checkpoints and choose extreme speed. Although I will say it's yet to be optimized, artistic styles work well, photorealism can use some work but that's expected. I'm getting roughly 7 seconds for a 1024x1024 image.
Maybe it's a recent add, but you can do in and out painting in Fooocus and there's an upscaler.
@@MonzonMedia but due size of file is impossible test with virustotal about virus
After in painting you may do another video about styles , That is the treasure chest try your prompts first with the default styles after that disable default styles and try to mix 2 -3 styles and mix them with default styles, even without any prompt just create with mixed styles you can spent hours no other webui this much rich styles
Yes very true, honestly the styles are more than enough and work very well.
I want to know if i have the right to use the generated images from it for commercial use, or not ?
Thanks
Hey sir, I'm getting errors while using the tool. If i faceswap, then go to upscale or unpainted the pic, it shows error. Meaning, i couldn't able to multiple task back to back. To do the next work, i have to again close the page and then again run the process. Can you tell me how to encounter it?
Hello! I use this mostly but hands and eyes keeps ending up weird, I'm just using the basic juggernaut model and a text prompt with a face swap.
I tried downloading the mre fooocus mentioned in your other video because I heard 'controlnet' helps with hands/feet/etc. But I cant seem to face swap prompt there, so I'm a bit confused on:
How to face swap using one of the two fooocus setups (with text prompt as here), but that also has control net or some way to have more reliably realistic feet/hands/eyes?
Have to tried inpainintg? czcams.com/video/Ky6B8oRStMU/video.htmlsi=oHheqgLznfovucGy
@@MonzonMedia It takes 20 tries a lot of the time to remove all defects, if feet and hands are near each other? Absolutely impossible
Great video, thanks!
You're welcome!
very instructive tutorial! do you know the best way to create different emotions for the same face?
Thank you! Just write it in your prompt, smiling, happy, angry etc. to get consistent faces you 1-2 names, if you can't think of any just use a random name generator. There are a few online.👍
what is your settings? i do the same and i never have accurate face swap.
I am one of those who are not recognized in the face exchange.
The funny thing is that if I use an online AI generation service with the ability to use a custom face, this works perfectly and I have my face generated on an AI model.
Hey really nice explanations, what version are you using? every time i use fooocus and use the input image i get a error which is really sad i really want to play around with all those features, if you know some of the issues and know how to fix them then a troubleshoot video would be nice
Probably you have too little VRAM. I have had the same problem.
ow really, yea that could properly be the case then, ill look into that thanks@@eugenmalatov5470
i just checked and i have 12288 MB Dedicated video memory i would assume that would be good enough 🙂@@eugenmalatov5470
Not fooocus related but I been trying to use faceswap in discord midjourney but the results are non consistant, Any advise would be cool. Thank you!
Buenas tardes, tengo un RX 6700XT + Ryzen 5 5600 , como hago para que me funcione Foooocus. O la unica manera es con NVIDIA? gracias
Hello, great content, I have a question, If I have a GPU Nvidia card can I use the software offline?
Generally yes but some of the lower end 1000 series cards may not. Which gpu?
@@MonzonMedia well I didn't buy it yet....i am planing to buy one
@@BrainwavesBinauralBeats Awesome! Well anything to do with stable diffusion an Nvidia GPU is the way to go as it utilizes cuda cores which is only found in Nvidia GPU's. Get the best card you can afford with as much VRAM as possible. Although 8GB is generally sufficient, the way things are going 12GB is a minimum these days. Stay away from AMD unless you want headaches and poor performance.
@@MonzonMedia Thank you, I will)
wow that's an amazing channel
Appreciate it!
how to faceswap multiple people, if i have 2 or 3 different faces I want to be swapped for different people in an image
can I import a char pic I did in another AI tool, and then faceswap its face unto a second char pic I did in the other AI tool ?
Fooocus is taking face swap to the next level! 😮 Easy, fun, and a bit scary how good the technology has become. What do you think about the ethical side of this tech? Could we be entering a new era of digital identity? Share your thoughts! 🎭📱
Try ReActor in Stable Diffusion, and then see what FaceSwap means, in Fooocus it sucks
Hello I have en error when I do the step faceswap in image prompt. I write the prompt, do « generate » and I have an error. I tried several times
Hi! I see that you can change the model (realisticstockphoto) , but I only have one on my environment, Please, how do you install all the other models? Thanks
Yes, please see this video czcams.com/video/w5EA358IPCs/video.htmlsi=jdBmU2s-OoN-3O20&t=144 You just have to place your models in the checkpoints folder
thanks
Welcome!
How to combine 2 photos in one like a girl's pic and a boy's pic make it look like real ... that's a great topic you can cover like selfie and portrait
I haven't done this in Fooocus yet but I have some ideas. Unfortunately it doesn't have a canvas but I think there are some work arounds.
@@MonzonMedia that would be really great if you will make video on that, because i don't think there's any video regarding this .. like focusing on how to combine 2 people or 2 photos in one using AI
Are you talking about using real photos?
@@MonzonMedia yes,, real photos sir
Yes please, make the tutorial for this, Sir.
Hey I am a beginner. I am wondering how to do such face swap which includes also hair in sd webui. Roop and reactor seem to only swap face without other head attributes. Do anybody know how to achieve the same effect there?
Hence why it called face swap ☺ Well the tendency is for the face swap to copy the angle and direction of the face so you can use an external editor like photoshop or photopea to mask or clone the hair.
anyway to bypass the google collab gpu usage error?
Hi, great video. Is it possible to change the face to be unique?
In what way? Can you explain?
@@MonzonMedia Cause I feel that anyone can have my own model and have repeated characters in the web. Like the nose, the size of the eyes...
Hi Sir, i am unable to download this tool can you please provide me the link and process of installation as well.
I was doing the same, but when I'm doing generate then they show me an error. Can you tell me what problem this is?
Can this be done on an already-made image? basically for it to do said face swap though it fully recreates the original image 1 to 1.
not 100%, the nature of stable diffusion is to be random, there are some work arounds abut it would require some post production. I'll be covering consistent characters soon.
I had the same question. this would be a helpful tutorial
@@MonzonMedia
Hey Monzon, could you explain how can I install models inside Fooocus if im using it through Google Colab? Everytime i add it inside models folder in Google Colab I can't run RealisticVisionV6 in Styles cause it gives me an error ''Connection Timed out'' :(
Sorry my man, I don't use colab unfortunately. I cover quite a few SD platforms so I chose not to. You can try the github discussion board, where the link for Fooocus takes you. (in the description)
@@MonzonMedia Thank you for the reply anyway, I appreciate your help! :)
No advanced settings for face swap?
Im a total noob still, but I think Iget how to get a consistent face, but how do you get a consistent body for the character, the different in hight for exempel is so easy to see as a viewer
Hi I keep getting this ERRORS after I try to FaceSwap, Any Idea where I can search for problems? Thanks
salamat sa info, pinoy ka pala!
Syempre pare 😬😊 Appreciate the support my man!
@@MonzonMedia started explorint stable diffusion last year and was using A1111 then ComfyUI.
pero tama, itong Fooocus mas maganda kasi you can focus on creating! saw your very informative videos, salamat
Question: Always when I try the fist thing you show, it makes an error message (Error conneciton lost). I have a macbook pro late 2013 ( I know its ten years old) but still seaart and other generations in fooocus work but the image input does not work. Can you help me
It's likely because of your Mac, I suspect it's intel based? I don't know for sure but it may not be supported github.com/lllyasviel/Fooocus#minimal-requirement
Wow, thank you for your prompt answer. Yes, it's indeed intel based. I will check the requirements and compare. I run it (fooocus) via the public link in github, so I havent intstalled it locally. @@MonzonMedia
is there a Linear model for Fooocus ? canny is not very efficient
Unfortunately not at the moment but you can sort of work with what's there with canny and depth and get decent results. Hopefully there will be more controlnet added soon.
I think it just uses Clip Vision to read the image instead of inswapper to properly change it like Roop or Reac do
It's actually a controlnet IP-Adapter
Can you please help, when I am clicking on the link, it is showing 504 gATEWAY ERROR
❤❤❤❤
Does it help to put in multiple variations of the same person?
Yeah sure, 3 is usually pretty good to use and different angles. Although not absolutely necessary.👍
@@MonzonMedia how do you adjust the weight and "stop at" options when you are using 3 images?
Since I started tinkering with comfyui, I like it much better than webui but I come back to fooocus a lot because it's so simple
Same, Fooocus is just so nice and easy to use although I love the versatility of ComfyUI even though it takes some work to do. It's also a great way for me to learn the various pipelines of SD.
Need help, I'm having "connection error"
Is there a way to fix that ?
I have a 1660ti tho
It may be your GPU as the 1660Ti is not guaranteed to work with Fooocus github.com/lllyasviel/Fooocus#minimal-requirement
@@MonzonMedia you maybe right but minimum is 4 but mine is 6 so can there be another issue ?, mine starts up with edge so is there a way to change the browser for it as in a command ?
why does that not work for me :(.. its every time a diferent face
Hello there, have you tried to install on a Mac M1?.
Unfortunately not, there are instructions on the install page github.com/lllyasviel/Fooocus and you can ask questions on the discussion board.
why cant i see image promt box for face swap ..not there or on foocus 2.078.5 mre
Did you check on advance? 0:35 I don't believe the MRE version supports it. The developer stopped updating it back in October.
i have fooocus and it generates some amazing quality pics but the only problem i face is that it takes few mins to make 1 picture and also my laptop starts to hang for around 20 secs when it starts to generate pic, how can i fix that, i might have to install something for fooocus that will generate faster and without being hang?
All stable diffusion platforms are heavily reliant on the graphics card. Do you know what you have? Fooocus is already optimized out of the box so typically it runs very well on it's own but of course depends on ones syste, specs.
@@MonzonMedia i also have stable diffusion but that runs very smooth, 1 pic in few seconds with 512×512, i have rtx 3060
@@donschannel9310 Interesting, I have a 3060Ti 8GB card and it works great. Are you running it on an SSD hard drive or mechanical one? It can make a difference.
@@MonzonMedia i really don't know that lol i watched a video on yt on how to download it and that's it, I'm sure there must be some settings to make it run smoothly.
after downloading the zip file and extracting how does one execute those files? pls help
Watch this video, the .bat files are what you click on to start Fooocus whenever you want to use it. Make sure to leave the command window open when using the program. czcams.com/video/w5EA358IPCs/video.htmlsi=ajFIJ2XlraIRuMDt&t=77
Hello, thank you for video. You say it is feasible to work on 4 gb ram gpu on PC and it works ok. But I cant find any tutorial how to install Fooocus on Mac. I have a 8 gb Pro Mac and it cant even launch any generation. Will you have a video how to install it on Mac properly maybe? I really want to try it but dont think i want to buy pc for that. Can you please comment and thank you in advance
4GB VRAM for Nvidia users. It will work on M1/M2 mac but I don't have one. Just look on the github page for instructions for Mac.
github.com/lllyasviel/Fooocus
@@MonzonMedia I did follow the instructions (i am on M1) but it just keeps freezing and crushing. Looks like it doesnt work for Ma c os as it should have to (((
why sometimes dont run gpu????
ok.. so. If I give a name in the prompt, it will generate the same "person". Then Face Swap will work, still naming the "person" at other occasions? Then you will have sufficient photos to train a Lora
if you are using the face swap no need to name it anymore in the prompt although it doesn't hurt to leave the name in there. Just be aware that if you are changing models or using Loras the face can change so it's always good to use the same models/loras when using the same character in other scenes.
why its always generate brown eyes
Please show me in your next video how to do face replacement in AOOOCUS. I really need it.
uuuuummmm I'm confused? This is a video about face swap?
Most tutorials, like yours also, unfortunately 'Fooocus' (pun intended) on the low hanging fruits of moving a face to a completely new AI generated target image. To my opinion a much more common use case would be moving a face with the correct style to an already existing target image where ONLY the face is being replaced, keeping every thing else as it was. Meaning a combination of face swapping and inpainting. Can you offer as, solution for this also?
You can do this with Face Fusion czcams.com/video/WqpcAxoPhL0/video.htmlsi=wkqqt7tb8zt7D4Tp
Whenever I try to use face-swap i get disfigured eyes and sometimes faces. I cant even regenerate them, it still messes them up :S Have tried to up the weight, leaving Style with only V2, Photograph and negative and tried to tweak my guidance scale and image sharpness. Not really sure why it keeps disfiguring my faces. Tried multiple different faces as well :/ Anyone got any ideas?
well it depends on the perspective, if you are creating images that are full body or in some cases half body, it's not unusual for that to happen but you can fix it with inpainting. czcams.com/video/Ky6B8oRStMU/video.htmlsi=hVI_IZVcpaW-Zpt1
i am getting an unsharpened background every time so it looks like person is "photoshopped" into scenery... how to fix, that it looks more realistic?
Are you talking about the background blur? Not sure what you mean exactly.
@@MonzonMedia yes. Exact
@@derhitpula Put in the negative prompt blurry background, bokeh, shallow depth of field, and also in the main prompt it’s often good to reference something in the background like mountains as an example. Wide or deep depth of field can work too. As an example "portrait photo of (subject or person), nature, mountain background, deep depth of field. Negative prompt "blurry background, bokeh, shallow depth of field".
How i can upscale for image in fooocus on colab 😢
I don’t use collab, try the discussion board on GitHub github.com/lllyasviel/Fooocus/discussions
@MonzonMedia ur tricks so nice I hope to ues them on fooocus
can we do more than one face swap in the same video?
Do you mean for multiple people in the same image?
about video faceswap?
Can't do it with Fooocus but you can check out face fusion for both images and video czcams.com/video/WqpcAxoPhL0/video.htmlsi=hYnGawIXekmylSzO
Can we do face seap in easy diffusion?
Nah unfortunately not and also it's no longer receiving any new updates, just maintained that's why I no longer cover it.
How can I use this on a mac?
Assuming this is your first time installing anything related to stable diffusion, I'd suggest using the Pinokio installer pinokio.computer/ However it only works on an M1/M2/M3 mac. It's very easy to install. I have a video here but it's for PC czcams.com/video/FWrP8scitBI/video.html&ab_channel=MonzonMedia. Once that's installed, search for Fooocus and install it and you're done. This is the easiest way. Manually can be quite challenging for most.
why always lost connection
But why you swap a face here that is almost identical and hardly see a difference ?
5:16 yes you can make buutiful models. bcoz trained. who train you xD. and allways can swap face, but why is not do it LOL
yet again same. is it stable fusion or forge. and why need. same works on
Bro when i click on link I got error 504 gateway timed out
The links in the description? I just checked and they are fine?
Have you tuned in to VideoGPT yet? It's the under-the-radar gem transforming video storytelling.
How can i install Fooocus ?
Here's an install video for PC and if you have an NVidia GPU, otherwise there are directions on the github page. Link in the description of the vid. 👍 czcams.com/video/w5EA358IPCs/video.htmlsi=AQrHIVnEnPjPe21u
FaceSwap doesn't work like in ReActor, so it's not much use in Fooocus yet.
no it's not but I still find it useful for my needs.
But what that "weight" means? What does it do?
With weighting you can emphasize a word in the prompt a bit more so let's say you had a "woman with blue and purple striped hair" if you wanted the purple to be more prominent you can add "woman with blue and purple:1.5 striped hair" . I've been thinking of doing a prompt video in general, maybe I can show this in fooocus.
Please anyone say, is that free?
Yes if your system meets the requirements and NVidia GPU's are highly recommended github.com/lllyasviel/Fooocus#minimal-requirement
This seems kinda dangerous right? Whats stopping from using this technology on an ex girlfriend or friends to add their face to an explicit photo?
It’s no different than photoshop and of course the users have to use this responsibly. That’s not on me
i would like to see open pose in fooocus
Yeah would be nice to have more ControlNet wouldn’t it?
Is it free
Yes as long as your system meets the requirements. czcams.com/video/w5EA358IPCs/video.htmlsi=ElWrHubw-TRmyWVF
When I do it, with an image, and then a face swap of my face, it ignores the original image and just renders me with a bunch of F'n cameras. Some kind of default setting. Why is it always outputing old school film cameras in my output? I don't have any prompt at all, and yet it is splattering my outputs with random physical representation of cameras everywhere... like I'm some kind of camera store sales girl or something. There are literally no cameras in my source images, no camera word in my prompt string, nothing.... Fooocus is the most infuriating program ever. I've generated like 30 pictures now and in every shot there is at least one film camera, or worse, I'm surrounded by them...
Must be in the styles. I've never had that issue.
is he work on video??
Unfortunately it's only text to image, not video.
@MonzonMedia no no haha u not understand nnsnse i mean to say its could be achive like deep fake?? For videos not t2v
Is it only me? I find the image of the guy in the purple suit inside the cathedral not photorealistic at all. Just the opposite, the face looks cartoonish in a dall-e way.
when you prompt simply it can generate a wide variety of styles, besides it was just for demoing the face swap.
As long as you keep the name, of course it's the same face ... this should be with removing the name
Works just as well without the name. I leave it on there because fooocus face swap isn’t as good as I’d like it to be.
@@MonzonMedia can you define new characters inside fooocus? Like creating a new ID
@Vypeerz not sure what you mean by creating an id?
@@MonzonMedia when you call out a celebrity/brand name the library has this indexed and will generate images based on how it knows that person looks like the one you used.
Is there any way to give the algorithm 1 or more pictures and say “this is xyz” and from then, when you prompt xyz it will generate img around that ID/entity
@@Vypeerz not exactly, there are a few ways to do something similar. You can use a lora to identify a particular person, either one that exists and is already available or you can train your own but that requires a decent GPU, still doable with 8GB VRAM Nvidia card. The other way in Fooocus is to use wild cards or presets. But this method will only take a dedicated prompt for that particular look. It won't help with consistency when it comes to the same attire but should at least give you the same face. I have a video to edit on this topic soon.
NOT EASY NO CLUE WHERE TO START ANNOYING
This isn’t the a video for if your new to focus. I go over some basics here. czcams.com/video/w5EA358IPCs/video.htmlsi=U17sD-2JhWLXm8Pt and have other related videos. Just have to ask
I am facing problem in downloading this, can you help me? Whenever I try to download this, authentication fails after downloading till 50% above.
What are your system specs?
Susbcribe , like ,, share