Fooocus Tutorial Part 4 - Tips for Inpaint/FaceSwap/Refiner & More

Fooocus is a great Stable Diffusion tool for everyone thanks to it's easy to use interface. In this video I will go through Inpaint and Outpaint thoroughly. I will also show how to use FaceSwap in Inpaint. Plus a few extra tips about the Refiner and even a little about Magnific AI.
Realistic Vision Checkpoint: civitai.com/models/4201?model...
If you don't have Fooocus installed yet or want the overall basics explained check this video out:
• Fooocus Tutorial - Sta...
-----
00:00 Intro
00:19 Inpaint & Outpaint
04:12 Advanced Breakdown
06:56 Inpaint FaceSwap
10:07 Refiner Tip
11:35 Magnific AI Effect
-------
#stablediffusion #fooocus #focus
Music by Epidemic Sound

[@JumpIntoAI]

If you don't have it yet, this is the Fooocus Install page: github.com/lllyasviel/Fooocus

[@jyyy8247]

Bro you are actually really good at this. The pointer/arrows/visualization, the level of detail is perfect, your explanation is clear and concise. This is some real quality tutorial, very rare to find on KZbin nowadays. It's extremely helpful, thank you for the amazing content.

[@uTubeJRod]

You're making some serious quality training materials here. I used Fooocus for a month like a noob, and now I use it like a pro. Thanks man 😁😎

[@canelj]

This was the best tutorial on Fooocus's advanced features I've ever seen. Thank you

[@JumpIntoAI]

Wow I appreciate that. My editing skills are still in progress. And this took longer than it looks to put together, glad you liked it. Makes it worth the work.

[@One_-_Man-_-Clan]

But you talk too fast. slow down, zoomer. you also sound like an A.I. voice. 🤣🤣🤣@@JumpIntoAI

[@TheSporkimusPrime]

Your videos are by far the most informative I've come across! Keep up the good work. Would be cool if you talked about different models out there or LoRAs you use and the weights on them / CFG

[@JumpIntoAI]

Thanks! That's definitely an idea. If I can put together something that feels worthwhile I certainly will.

[@johntnguyen1976]

I second this idea...you're so great at explaining and breaking everything else down, Loras would be the next logical thing to wrangle down to plain English like you do. 😂

[@puja1985]

Thank you so much for this wonderful, easy-to-understand tutorial. These are really nice tips on the face-swapper. I hope you continue to educate us with a fooocus on the upcoming videos.

[@WilliamBrwn]

Thank you for the tutorials. It's really helpful if someone shows you the interface and some settings to start with.

[@peterimade003]

Previously didn't feel like this video would be worth It, but this is the best fooocus tutorial video on KZbin, so concise and precise. Please don't stop posting 👏👏👏👏

[@wolfhound2000]

I’m really enjoying your videos. No crap, just very useful content provided in a relaxed and informative way. Please keep it up 👍🏼

[@eleonoracapkova7370]

Great video. I use Fooocus and love to experiment with new features. Also today I came across Magnific and was surprised how much they charge. Thank you that you have mentioned it in video.

[@JumpIntoAI]

Yes, Magnific AI is a great product for what it does, but not at that price. I think they charge that much because in time there will be a free open source alternative. The stable diffusion community is already working on it.

[@Hey-Its-Retro]

Thanks for another truly great video! I've tried all the Stable Diffusion systems and find Fooocus to be the easiest and best one for getting effortlessly good results! I also love how you can dig in a bit and get to some really advanced features... Thank you so much for your videos - they are superb and I learn something new and incredibly useful in every one! So cheers for everything you do! You ROCK and that's OFFICIAL! Happy New Year!

[@JumpIntoAI]

Wow, thanks for such kind words. Not sure I deserve it but I'm glad you are learning something from my ramblings.

[@Hey-Its-Retro]

@@JumpIntoAI Yup... you deserve it! Keep up the good work!

[@hmm-m]

I absolutely agree, this is the best clear way informative channel on Fooocus. Would be awesome in the future if you explain also all those difficult things, e.g. related to that Realistic Vision Checkpoint, exactly what they mention on how to use it. To make clear things like: all options in Developer Debug Mode, what are all those samplers, all those schedulers, what is VAE, are they used automatically if inside the folder, what are all those other folders and what can you put there and how to use it, like clip, clip vision, diffusers, embeddings, gligen models, unet etc. etc... I'm soooo hungry for well-presented knowledge! :)

[@bennibeatnik]

Great video, you explained so much and answered so many of my questions, thanks!

[@garry3989]

Just wanted to pass on my thanks. These tutorials were fantastic and explained everything about Fooocus perfectly.

[@downunderdan]

The information you imparted in this presentation really helped this AI newb out of a rut. Many thanks, sub added

[@lrq3000]

Very clear and concise tutorial with all the practical tips required to properly operate these tools, thank you ! I especially appreciated the "Advanced breakdown" part that explained the intricacies of the settings and why the different settings worked as they worked and their respective limitations and advantages. I also appreciated the bonus bit about Magnific AI. I wasn't aware of this, and I was looking for various solutions to achieve a similar feat, so thank you for pointing that out. Combining this with Upscayl to increase resolution is I think more than enough for most purposes.

[@JumpIntoAI]

Thanks! I wasn't sure about adding that Magnific AI bit, good to see someone found it useful.

[@lrq3000]

@@JumpIntoAI Yes definitely! And I am far from a total newbie, I am an AI researcher with already some experience running stable diffusion locally, but I'm not keeping up with all the news so this definitely is helpful for me and I'm sure for a lot of other people as well!

[@XstahSie]

Thank you for this amazing Fooocus tutorial!

[@jankucera8505]

great guide from someone who understands what they are doing

[@COii3153]

As others have said, great tutorial! I spent hours trying to figure out Inpaint VS detail... Kudos to you 🙏

[@JumpIntoAI]

Glad it helped!

[@michaeldrabnov6645]

Thanks so much man! Loving your tutorials 👌🏽

[@JumpIntoAI]

You're welcome! I'm here to help. 😁

[@m0kiiy]

thanks for your tutorials. keep it up man 🙌🏻

[@dominiklebiedowicz1827]

Great work, very informative and straight to the point 👌

[@JumpIntoAI]

Thanks! I do my best to keep these snappy and without fluff.

[@ADV_UA]

Very useful! Thanks!

[@Gh0sty.14]

Thanks for all the tips! I was looking for how to faceswap with inpaint and found this video. It's working great for the face but the skin colours are so different no matter what I try that it ends up feeling kind of pointless.

[@JumpIntoAI]

Yeah the inpaint face swap is far from perfect. I usually have to try to run the image again through the vary subtle ( and i even lower the denoise in the debug) to see if I can smooth out the tones. Sometimes I can get it. Others It changes the features too much and just have to move on. In truth if you are going for 1:1 for faceswap in general it's pretty difficult.

[@berken_]

Thanks much, very informative

[@gebuildersdeveloperspvtltd3846]

great info, to the point ❤

[@Grub370]

Great video, thanks!

[@kevin7126]

You the best explain for fooocus, please do more videos please..

[@manulamalaka1368]

This was the best tutorial, give me more advanced tutorials about ai genarations

[@JumpIntoAI]

Thanks! I will! 😁

[@b.s.design]

Super! Thanks!

[@lenny_Videos]

Awesome tutorial 🙂

[@vandeth473]

I found your video, and thanks so much for sharing. I hope you drop a video about how to generate a beautiful landscape look like a real ❤

[@leonardovideomakeroficial]

🇧🇷🇧🇷🇧🇷 Very good. Muito bom, gostei muito deste tutorial, parabéns.

[@ovideotube]

amazing work! please keep up

[@JumpIntoAI]

Thanks, will do!

[@augustosalcedo5117]

Amazing 😮

[@vasja341]

Great tutorial!! Thank you very much!! I'd appreciate it if you could show a tutorial on using Fooocus outputs to create a video. What tools to use for this

[@Rey7430]

hey man, can you please make a tutorial on how to load Safetensors Model in Fooocus? that would be really helpful..great tutorial btw!

[@carl14281]

very nice video thanks!

[@JumpIntoAI]

No problem!

[@jeeadvanced4719]

Thanks man !!!

[@ImaneLhmidi]

the best tutorial

[@ksiobrga]

Thank you!

[@JumpIntoAI]

You're welcome!

[@titanmars9229]

This is really helpful. Would you be able to possibly make an api tutorial?

[@JumpIntoAI]

Possibly! But couldn't say when, have quite the list as is.

[@logudesignstudio]

thank you🙌🙌

[@videosfeoscomotucara9038]

Thanks gold boy

[@user-gx2zr7hs2j]

Great video ! Started wandering more in the "Developer Debug Mode" and in the Debug Tools section there is a method called "Refiner Swap Method" with three options - joint, separate ,vae. Could someone explain what do any of these options do. Possible idea for a video on this matter.

[@JumpIntoAI]

This is my opinion/guess on how that setting works but foooocus has it's own native refiner method that uses 1 k-sampler vs 2. this way it can give a more consistent outcome and doesn't waste time/power having to stop and start again with another k-sampler. My guess is that is the default "Joint" setting and "Separate" is the old method. unclear about "Vae" with no testing, my guess is it might be for models that don't have a vae baked in. Again, all opinion/Guessing right now.

[@fabiotgarcia2]

Excellent tutorials! Actually amazing! Thank you!! One question... is it possible to apply real products to a model, like apply a photo of a shoes to the model´s feet?

[@JumpIntoAI]

Unfortunately no, you can inpaint the feet and using the "mixing inpaint and image prompt" setting in debug, have a picture of the product in the input image>image prompt box and it will give you an image "similar" but never the exact same. The tech is getting better, but not there just yet.

[@allen_d99]

The issue when the faceswap stops working at the 15th step can be solved by setting "forced overwrite of refiner switch step" to 60 (you can find this in advanced>advanced>debug tools)

[@JumpIntoAI]

Yes! Thanks, I did figure that out and updated in other vids. Appreciate the notes though for others.

[@HiProfileAI]

👏🏾👏🏾👏🏾👏🏾👏🏾👏🏾👏🏾👏🏾👏🏾👏🏾👊🏾😊

[@aniketdhage1440]

in which folder should you paste the refiner that you download(realistic)?

[@JumpIntoAI]

they go in the same place as the base model. fooocus\models\checkpoint

[@basilica99]

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!

[@JumpIntoAI]

Sadly, right now that isnt possible. getting it to input an exact product and logo wont work. you can get something similar but it will never be the same as the original product picture.

[@basilica99]

@@JumpIntoAI thank you 🙏

[@lilillllii246]

Thanks. Can't I use image prompt(cpds,faceswap,pyracanny) and inpaint at the same time on foocus?

[@JumpIntoAI]

you Can use faceswap effectively with inpaint. And while cpds and pyracanny can be activated in inpaint, it doesn't work like you would think.

[@coder8515]

Thanks for awesome content! I'm running fooocus on Collab (payed) and I can't do a face swap mixing Inpaint with Image Prompt. I always get anderror when control models are downloaded. I checked and I'm not running out of space. Any idea on how to fix this?

[@JumpIntoAI]

I admittedly am not well versed in foocus google colab errors. but I would say you should post the error log on github.com/lllyasviel/Fooocus/issues more eyes there with better knowledge on that subject.

[@blackmambagp]

hi! which model do you use? in dreamshaper xl the face swap is not so good

[@danielchoritz1903]

for the earrings check your styles, or better un-check all styles, some have earrings as prompt

[@joacosolbes9283]

Hello mister, thanks for the videos, but, how to add an already existing image to the refiner tool? i mean, not create one, but add an already existing file and use the Realistic Vision V6.0 B1 to refine it?

[@JumpIntoAI]

Go to Input Image> Upscale or Variation> load an image and use Vary Subtle or Vary Strong. Have the refiner set correctly and then generate.

[@onlineispections]

HI. We ask if it is possible to create two characters in one Image that speak, taking the two faces of reference the same, instead of one, or more people. They are the characters of a storyboarded movie. Same faces of two or three people, how do you do it? We can Write on one prompt (character on the right, left) or have multiple image prompts below? Thank you

[@JumpIntoAI]

It's possible to get 2 specific characters in a scene with inpainting, but having them looking/interacting naturally will be the challenge. If you had a good base image of 2 people and you just wanted to replace their faces, that could easily be done with faceswap/inpaint combo. But still it will be hard to look natural as you have no control over their eyelines and such.

[@sanjuali3096]

broo i need all the characters like shinchan doraemon ben10 nad such cartoons which model or lora should i install

[@user-yf8ns8mp8i]

but I really want to understand more about the face-Swap technique. do you have a video on that topic?

[@JumpIntoAI]

I don't have a video exclusively on face swap but besides the information in this, I go through the technique in the video below. In the Image Prompt Chapter. kzbin.info/www/bejne/ZqfdpqGleL6cg6ssi=WIlwE-341HHKB9zD

[@user-yf8ns8mp8i]

@@JumpIntoAI cool, but i just want the same image with a different face. i just want to replace a face. it seems like every video i watch is just the same old "generate a new image based on X picture" . i just wish you guys had a tutorial on how to replace a face (and maybe add some details like gray eyes or dimples). that would be cool of y'all

[@JumpIntoAI]

@@user-yf8ns8mp8i Ok well the video here you are commenting on does just that, using inpaint it will put a face on an image without changing the original image. Specifically at 06:56. It takes a little more effort to accomplish but it does work. And then you can easily mask the eyes and type a different color eye and change only that.

[@pacpac973]

hi ur tuto are so greatfull , but one question how to fix fingers? :) i know the question is hard^^

[@JumpIntoAI]

It can be done with inpainting. Masking the hands/fingers and trying to re-generate those areas again.

[@pacpac973]

@@JumpIntoAI thanks but i try and try and try :)

[@seyfer110]

Is it possible to set its model etc folders from authomatic111 path? just to avoit having duplicates of about half a tb...

[@JumpIntoAI]

Yes you can! edit the Config.txt and set the paths to A1111. kzbin.info/www/bejne/bmObqIx_pZd7d68

[@seyfer110]

@@JumpIntoAI That's actually Perfect, thank you!

[@Cloud10D]

Is this a timelapse of your imagegeneration - or is it realtime ? If realtime - WHAT IS YOUR RIG ?

[@JumpIntoAI]

No I'm not running a super computer, lol. The video is sped up in those parts. But for reference I have a 4070 ti.

[@keepitshort4208]

I used the inpaint around the subject but im getting white circle around it, how to fix that ?

[@evokeaiemotion]

How do you generate images with detailed backgrounds instead of blurred backgrounds? Seems the prompts i try do nothing for it

[@JumpIntoAI]

prompts can work but SDXL seems biased towards blurry backgrounds, but you can try a couple LoRAs i've found helpful. civitai.com/models/259627/bad-quality-lora-or-sdxl civitai.com/models/135380/depth-of-field-slider-lora Also in fooocus, removing the default styles can help.

[@evokeaiemotion]

ty@@JumpIntoAI

[@jikaoru]

have you ever installed in Mac mini M1?.. does it work?..

[@CamilaJaviera-xq8uj]

hi, how i can put a refiner in the collab version?

[@hasger1941]

hello is it possible to replace with a photo of specific necklace? TIA

[@JumpIntoAI]

Sort of, you can use the "Mixing image prompt and inpaint" method and add a picture of the necklace and then inpaint the area you want it (helps to replace a similar necklace than trying to inpaint one in entirely) but it never is the same, you will end up with something "inspired" by that image.

[@hasger1941]

ok many thanks, we thought we could use it to model wearing certain accessories/clothing though it seems none of these bots can achieve this at moment

[@jeeadvanced4719]

Please tell me how can I download models like refiner model you showed while using Google colab fooocus (cloud)

[@JumpIntoAI]

I'm not an expert on the colab version but you can see here how to add models: github.com/lllyasviel/Fooocus/discussions/1228 However, I might be wrong but the refiner may be disabled in the free version on colab.

[@Adriana_blaze]

How to make a sharp background?

[@VakoCavender]

idk why the faceswap inpaint method shows errors every time i try to do it

[@Nnamdxxx]

I installed Fooocus Advanced on linux and do not see the Faceswap options. Can it be added or do I need to install this version?

[@JumpIntoAI]

You would need to install a version that has faceswap already.

[@rizwanks328]

i tried a lot to make the same person again but after every regeneration face have changed

[@himgos13]

@JumpIntoAI Sir do you recommend Juggernaunt v6 or v8 ?

[@JumpIntoAI]

I would say v8 for the increased accuracy of hands (still needs work but closer) however v6 does have it's own look that some people prefer. It's personal preference but v8 is my main now for when i use juggernaut.

[@himgos13]

@@JumpIntoAI thanks a lot :) One more question plz... I watched your second last video on various angles of character. I learned a lot from it, however I wanna know if I generate like 20 various pics with same face and create it's Lora, that would be a better approach for consistent character face? And can you make video on it if possible? :)

[@JumpIntoAI]

@himgos13 Yes, creating enough images of an exact character and turning it into a Lora is the best route. And yes a Lora Training video is definitely on the list 😉.

[@himgos13]

@@JumpIntoAI PERFECT!!! I'd love to see it asap.. Plz release as soon possible 😄 Thank you for great work! 🙏

[@deeprelaxing6825]

please fix hand problem too.

[@JumpIntoAI]

Didn't show an example but easily done with masking a bad hand and using improve detail in inpaint. 👍 As for having hands being good from the start. That will come in time as the models get more training and controlnets get fine tuned.

[@deeprelaxing6825]

@@JumpIntoAIThank you hope to see the next video.

[@useristensed]

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?

[@JumpIntoAI]

Is this with the free online colab version? If so then it is probably running out of memory. The free online version has a limited amount. If this is a local version on your machine, then it depends on the error, but if your VRam is on the low end then that might be hitting it's limit also.

[@useristensed]

@@JumpIntoAI okay then sir, if i install fooocus and then run it locally there's less chances of showing errors, is it?? And if memory limits the process, is there any way i can clear the previous memory once in a week or something like that?

[@JumpIntoAI]

@@useristensed If your machine meets the requirements yes, fooocus requires a medium to high end machine to run properly. I do not personally use the colab version so I do not know how often its resources reset.

[@useristensed]

@@JumpIntoAI okay now that I installed and ran it on my pc, it doesn't create a single image.. directly shows error.

[@379ETERNITY]

ЗДАРРРРРРРРРРРРРОВАЙИЯ;)(: НАПИШИ ПОЖАЛУЙСТА ИНСТРУКЦИЮ КАК СТАВИТЬ ФОКУС ДЛЯ ВИДЕО КАРТ ИНТЕЛ АРК И ПРОВЕРЬ РАБОТОСПОСОБНОСТЬ ПОЖАЛУЙСТА СПОСОБА

[@faramrremoved]

Can I sell a generated photo from fooocus?

[@JumpIntoAI]

As far as I understand it, yes.

[@faramrremoved]

@@JumpIntoAI thank you

[@Maltebyte2]

What i absolutely fail to understand is why im getting stunning images from my prompts but then the software fails to understand when im asking it to put cybernetic parts on the face! everything else it understands perfectly fine but REFUSES to do anything face related even using in paint! ...... pls dont hate me AI xD

[@JumpIntoAI]

might try a LoRA perhaps, not sure exactly what you are looking for but something like : civitai.com/models/280369/neon-cyberpunk-sdxl-cyberware

[@Maltebyte2]

@@JumpIntoAI Yeah i might have to! regarding the cyberware this is what im looking for what the other model just cannot do! Thanks.

[@One_-_Man-_-Clan]

how do you tell fooocus to open up in a specific browser, and not your default one? thanks.

[@JumpIntoAI]

No option for that I'm afraid and while you can add something like "start firefox.exe 127.0.0.1:7865" in the run.bat I personally don't know how to deactivate the autolaunch of the default browser in the program. Otherwise (a bit annoying) but you can copy the address "127.0.0.1:7865" after its opened up and put that in preferred browser.

[@One_-_Man-_-Clan]

Yea I use a hardened FF, but need to use chrome for Fooocus so I can drag n drop. I havent tried this, but I saved it from reddit: Start Fooocus Server without WebUI Add to bat file: --disable-in-browser @@JumpIntoAI

[@ragingpoop8658]

most of my inpaints (usually backgrounds) are blurry. How can I prevent this? I've tried negative prompts like depth of field, blurred background, narrow depth of field, etc.