This is a little more in-depth explanation of Input Image settings and a few other QOL tips. If you have yet to get started with Fooocus you can check out my first video here: ruclips.net/video/828wYIp2HAM/видео.html
Really great video, and I've watched several tutorials, and I know how to run a1111 sdui since a year already, but Fooocus is on another level especially with your very clear explanations, thank you very much!
You can also manually change the log to a specific date by typing it's respective numbers in the search tab, instead of existing ones. That way you don't have to search through the files every time or keep the extra window open.
Thank you so much for this really informative video. I understood everything you presented. You explained everything in simple English, in comparison to so many other FOOOCUS videos. I thank you for that. I am having so much trouble doing accurate FaceSwaps! All I want to do is put a different face onto an image (photos). I have tried every conceivable variation, and it just won't copy the face. Any help you can provide to me would be so very appreciated. I liked and subscribed. Going to check out your other videos now.
Thanks a lot, that was very helpful. Do you know what approach would be best for controlling person's eye direction? I sometimes want to have a person looking left or right, but the AI tends to generate eyes that look at camera and it's hard to fight that in prompts. Can image prompts help with that? And also there are situations when I want to change person's outfit elements to specific colors. If I do it in prompt, it mixes up stuff. For example, I say black suit with lightblue shirt and yellow tie, but it generates yellow suit. And one more thing - upscaling and inpainting details often change too much of existing details, moving brows up, changing eye size, wrinkles - that totally changes person's identity. What would be better way to improve person's photo details without losing identity?
Hi! Little bit off topic but I really need some help with this. I have several real photos taken by me, and I inpaint women on the picture. The woman looks fine, very detailed, but compared to the base picture, she look way too detailed and she really stands out from the pic ( i guess because of higher resolution than the base picture). So like is it possible to get the woman and the base picture more even and more equal to each other??? Is this what refiner would do?
No the refiner allows you to start with one model and then part way through an image generation it finishes with another model. I guess it depends on the picture, how different is the inpainted woman. is it a closeup or full body. If only using Fooocus, I would put the picture in inpaint. Mask my face, then in debug turn on invert mask (now everything outside the mask will change) and try turning the inpaint denoise way down like .15-.3 (setting in debug>inpaint) and rerun the image. Also use words like faded image, grainy image. To try and even the detail out. This will obviously change the image itself a bit, but it might be able to blend it in some. I almost never get a solid picture just from this method that doesn't need a little post processing (photoshoping/filters/etc) but this can get you closer.
Pop quiz:). I chose a pumpkin them kind of like you have shown in your videos where someone tried to take pictures of all the styles:). I am making more progress but as far as I know there all free to use and make whatever and then sell them correct? As long as it’s not copyright against any company or person it’s mine to sell correct?:). Thanks for making videos:)
Hi, I'd like to take a real image of a face and cartoonize it without loosing its essence. However, I can't get something decent. As soon as I try to add a specific style, the rendered image doesn't fit my basic photo at all. What do I have to do? Would you have a tuto that explains how to proceed?
Depending on the picture and style but say, you have a photo of a womans face and you want to Pixar it. Go to the input image>Image prompt tab, load an image into one of the 4 slots, set to PyraCanny, turn the "stop at" all the way to 1 and keep "weight" at 1, uncheck all the default styles. and type in something like "A photo of a Pixar Woman" Then from there slowly lower the PyraCanny weight. Like from 1 to 0.95 and keep lowering the weight until you get something thats enough pixar but still similar to the image. You can also add weight to Pixar in text prompt like "A photo of a (Pixar:1.4) Woman"
how work copy to clipboard? I press the key, it tells me it copied and I have to paste into the prompt area. but when I press ctrl v nothing happens. Where do I find the clipboard with the prompts.
I stable diffusion i see one add path folder and convert all photo in that folder to animation photo. Can i do that with foocus ? Please i wait your answer
You can try a cloud service with stable diffusion, many options out there but usually not free. Also several websites offer free limited generations. Leonardo.ai comes to mind since it has both images and animation. It has limited free generations that reset everyday.
Is there a setting that allows me to set the background to dark (as in your video)? Each time I launch Focus it has a white background. I'm not too savvy yet so I hope there is an option that I can select for this. Thanks.
you can add "?__theme=dark" in the address bar, like this. 127.0.0.1:7860/?__theme=dark If you understand how to edit one of the presets, you can add the line in to load every time with --theme dark Such as if you opened up the realistic.json in the preset folder. and added that line to the end, to look like this .\python_embeded\python.exe -s Fooocus\entry_with_update.py --preset realistic --theme dark
@@JumpIntoAI Thank you. I'll give it a try shortly. Just for the record, the first link you provided (http...) has an error, "This site can't be reached" at the time I'm writing this.
@@mysticgatekeeper the link was only an example. When you load fooocus (assuming this is installed and running on your machine) that will be the address in the web browser bar. simply add /?__theme=dark at the very end and you can get the dark theme. If you are using google colab version then the address will have a bunch of random letters and numbers and end in gradio.live. simply do the same thing and add /?__theme=dark
Are you using a model in this? I copied your exact steps but when i upload an image in the variation/enhancer section with vary subtle setting, my image does comes out in the clothes, pose, hair style etc but it is wayyyy more real. Your comes out exactly like your original image but with the additon of smiling etc due to your prompts. NOTE: I did also use your smiling prompt and that worked but just my own is wayy to real compared ot my original image. Why is that ? Is it a model setting perhaps or what may be the reason?
No, for this i used the default model, which is juggernaut. It's hard to say without knowing what image you are using and other settings. You don't have any extra styles selected? You are using the vary (subtle) and not vary (strong)? not using a refiner? only other suggestion would be to try to go into the debug>debug tools> and adjust the denoise in "Forced Overwrite of Denoising Strength of "Vary" setting. try putting it at around 0.5 and then lower if its changing too much or raise a little if not changing enough.
I used every thing you did, Juggernaut, kept it at subtle, no refiner, tried with and without stock lora that comes with it upon setup. I am wondering if it is because I am using a photo i uploaded from dalle3 (not one that was generated on fooocus). Also note: I used same settings as the ones i mentioned above in this msg in the image prompt section and tried with various weights/stop at and tried with pyracanny and image prompt just to see if it made a different and it all comes out looking like my girl original uploaded image but just a very real version of the girl in my image. Also, I used your method of changing the "forced overwrite of denoising strength of "vary" to 0.5 and other variation and nothing rly changes it. Still a very real version of my girl :( Anyway I could show you what i mean?@toAI
I used every thing you did, Juggernaut, kept it at subtle, no refiner, tried with and without stock lora that comes with it upon setup. I am wondering if it is because I am using a photo i uploaded from dalle3 (not one that was generated on fooocus). Also note: I used same settings as the ones i mentioned above in this msg in the image prompt section and tried with various weights/stop at and tried with pyracanny and image prompt just to see if it made a different and it all comes out looking like my girl original uploaded image but just a very real version of the girl in my image. Also, I used your method of changing the "forced overwrite of denoising strength of "vary" to 0.5 and other variation and nothing rly changes it. Still a very real version of my girl :( Anyway I could show you what i mean?@JumpIntoAI (sorry i deleted your reply to me)
@seanknowles7987 I figured it was an image not created in fooocus, however I have had luck pulling random images in and doing the same thing. My other suggestion is perhaps try masking the mouth area and using inpaint to change the expression. Just dont mask the entire face and see how this comes out. You can upload the image to something like imgur.com and post the link here if you want me to give it a shot. no promises 😉
Hey I noticed that my FOOOCUS output folder only has two days worth of photos. I have been using fooocus to make photos for almost two weeks but I only see two days store. Why is that?
Hmm, is fooocus installed on your computer or are you using the google colab version? Colab version you will need to manually save them otherwise they will dissapear. If on your computer, not entirely sure. Did you make any presets recently or any changes to the config.txt output path?
@@tristanthedarkstar Not entirely certain what would cause that. I'm sure your hard drive isnt full? are all of your images currently saving since then?
I’ve trained my Lora’s in 1024x1280 aspect ratio. Can i use this aspect somehow in fooocus? How do i set it? I dont see a custom option for aspect ratio’s
you might have seen this here: ruclips.net/video/zWLknfrhfW4/видео.html but also I didn't mention (will put in another video) is you can change this temporarily in the the debug menu>debug tools. scroll down to where it says "Forced Overwrite of Generating Width" and "Forced Overwrite of Generating Height" and put any resolution in there you want.
@@JumpIntoAInice to know as well. I’m going to test out fooocus now. I think I’ll be using auto and comfy and focus for different things. Photobashing especially in focus. I wish there was a tunnel from Photoshop into focus to be able to quick draw to inpaint denoise. Also, on a different note, I discovered Breadboard yesterday. Looks like the best Ai image viewer for metadata. But it’s sort of on hold by dev. Do you know a good image tool like that? I’m currently using Allusion, but it also is not being developed. Open source projects 😮
Yeah I liked Breadboard, wish it was continued. I been using DiffusionToolkit (it's windows only I believe though) and that's been good enough for now.
The colab version is online only. If you download this and use on your own computer then no. Once you have the basic models and all components downloaded (which wont download until the first time you use them) then it can be used completely offline.
Short answer is yes. Depending exactly what you want to do. There are lots of discussions out there on the topic. can check out these 2 reddit threads. www.reddit.com/r/StableDiffusion/comments/18msbgb/magnific_ai_but_it_is_free_a1111/ www.reddit.com/r/StableDiffusion/comments/18mqksc/after_a_full_month_experience_with_comfyui_making/
This is a little more in-depth explanation of Input Image settings and a few other QOL tips. If you have yet to get started with Fooocus you can check out my first video here: ruclips.net/video/828wYIp2HAM/видео.html
Best explanation I've seen yet for ImagePrompt, PyraCanny, CPDS and Face Swap. Thank you!
These are the best tutorials on Fooocus I've found, thanks!
Thanks. I appreciate the support. I really do!
Incredible in depth look into advanced features i never knew about, thank you lad!
No problem!
7:51 i had to stop the video to write this comment dam ur Good man this is amazign explanation shoot man thanks so much man
Really great video, and I've watched several tutorials, and I know how to run a1111 sdui since a year already, but Fooocus is on another level especially with your very clear explanations, thank you very much!
You can also manually change the log to a specific date by typing it's respective numbers in the search tab, instead of existing ones. That way you don't have to search through the files every time or keep the extra window open.
Nice tip! +1
Many thanks, great tutorial! I've just revisted Fooocus after so much fustration with Comfy UI and its joy to use and insanly powerful!
Outstanding. I've just started with Fooocus and its bit overwhelming with those settings. Thank you for this video. And i Subscribed.
Really enjoying these series, thank you.
Very well explained! Thank you!
Thank you so much for this really informative video. I understood everything you presented. You explained everything in simple English, in comparison to so many other FOOOCUS videos. I thank you for that. I am having so much trouble doing accurate FaceSwaps! All I want to do is put a different face onto an image (photos). I have tried every conceivable variation, and it just won't copy the face. Any help you can provide to me would be so very appreciated. I liked and subscribed. Going to check out your other videos now.
I learned a lot! Great video. Thank you very much!
Glad you liked it! ☺
Thanks a lot, that was very helpful.
Do you know what approach would be best for controlling person's eye direction? I sometimes want to have a person looking left or right, but the AI tends to generate eyes that look at camera and it's hard to fight that in prompts. Can image prompts help with that?
And also there are situations when I want to change person's outfit elements to specific colors. If I do it in prompt, it mixes up stuff. For example, I say black suit with lightblue shirt and yellow tie, but it generates yellow suit.
And one more thing - upscaling and inpainting details often change too much of existing details, moving brows up, changing eye size, wrinkles - that totally changes person's identity. What would be better way to improve person's photo details without losing identity?
PERFECT VIDEO!!!!!! GREAT INFORMATION!!!!!! Thank you so much for your effort.
Thank you very much!
Thank you so much for the awesome tutorial.
Great - straight to point. Thank you!
Hi! Little bit off topic but I really need some help with this. I have several real photos taken by me, and I inpaint women on the picture. The woman looks fine, very detailed, but compared to the base picture, she look way too detailed and she really stands out from the pic ( i guess because of higher resolution than the base picture). So like is it possible to get the woman and the base picture more even and more equal to each other??? Is this what refiner would do?
No the refiner allows you to start with one model and then part way through an image generation it finishes with another model.
I guess it depends on the picture, how different is the inpainted woman. is it a closeup or full body.
If only using Fooocus, I would put the picture in inpaint. Mask my face, then in debug turn on invert mask (now everything outside the mask will change) and try turning the inpaint denoise way down like .15-.3 (setting in debug>inpaint) and rerun the image. Also use words like faded image, grainy image. To try and even the detail out.
This will obviously change the image itself a bit, but it might be able to blend it in some.
I almost never get a solid picture just from this method that doesn't need a little post processing (photoshoping/filters/etc) but this can get you closer.
Perfect Tutorial
Awesome! I learned a lot! Great video
Hey, very good Video. How can I keep the face AND the body (shape) from a person but change the pose, background, facial expression etc.?
I subscribed. I just downloaded foooocus and I'm trying to learn how to do this stuff
Awesome tutorial Very informative. Keep it up 👍
Love your videos man keep it up!
Pop quiz:). I chose a pumpkin them kind of like you have shown in your videos where someone tried to take pictures of all the styles:). I am making more progress but as far as I know there all free to use and make whatever and then sell them correct? As long as it’s not copyright against any company or person it’s mine to sell correct?:). Thanks for making videos:)
As far as I understand it, yes you can do that.
I don't have the upsace/variation tab, any idea why?
Excellent explanation, thank you!
Hi, I'd like to take a real image of a face and cartoonize it without loosing its essence. However, I can't get something decent. As soon as I try to add a specific style, the rendered image doesn't fit my basic photo at all. What do I have to do? Would you have a tuto that explains how to proceed?
Depending on the picture and style but say, you have a photo of a womans face and you want to Pixar it.
Go to the input image>Image prompt tab, load an image into one of the 4 slots, set to PyraCanny, turn the "stop at" all the way to 1 and keep "weight" at 1, uncheck all the default styles. and type in something like "A photo of a Pixar Woman"
Then from there slowly lower the PyraCanny weight. Like from 1 to 0.95 and keep lowering the weight until you get something thats enough pixar but still similar to the image. You can also add weight to Pixar in text prompt like "A photo of a (Pixar:1.4) Woman"
@@JumpIntoAI Thanks a lot, I'll try it. ;^)
how work copy to clipboard? I press the key, it tells me it copied and I have to paste into the prompt area. but when I press ctrl v nothing happens. Where do I find the clipboard with the prompts.
Bonne présentation. Continue
Merci beaucoup.
I stable diffusion i see one add path folder and convert all photo in that folder to animation photo. Can i do that with foocus ? Please i wait your answer
Fooocus does not do animation/animatediff, only images, if that's what you are asking.
@JumpIntoAI my lap is poor for diffusion. Do u know. Any another app can I use it like diffusion on my lap ? Thanks
You can try a cloud service with stable diffusion, many options out there but usually not free. Also several websites offer free limited generations. Leonardo.ai comes to mind since it has both images and animation. It has limited free generations that reset everyday.
Is there a setting that allows me to set the background to dark (as in your video)? Each time I launch Focus it has a white background. I'm not too savvy yet so I hope there is an option that I can select for this. Thanks.
you can add "?__theme=dark" in the address bar, like this.
127.0.0.1:7860/?__theme=dark
If you understand how to edit one of the presets, you can add the line in to load every time with --theme dark
Such as if you opened up the realistic.json in the preset folder. and added that line to the end, to look like this
.\python_embeded\python.exe -s Fooocus\entry_with_update.py --preset realistic --theme dark
@@JumpIntoAI Thank you. I'll give it a try shortly. Just for the record, the first link you provided (http...) has an error, "This site can't be reached" at the time I'm writing this.
@@mysticgatekeeper the link was only an example. When you load fooocus (assuming this is installed and running on your machine) that will be the address in the web browser bar. simply add /?__theme=dark at the very end and you can get the dark theme.
If you are using google colab version then the address will have a bunch of random letters and numbers and end in gradio.live. simply do the same thing and add /?__theme=dark
That’s a fast computer you have. Each generation takes a few seconds. Or, did you speed up the video?
Video is sped up in those parts, just to keep things moving along.
Great content my dude. Thanks.
What is your hardware, it runs so fast for you.
Man everytime I open the run.bat file I get error ' connection errored out'. Is there any fix for this issue??
this has been linked to not having enough system swap
github.com/lllyasviel/Fooocus/blob/main/troubleshoot.md#system-swap
Are you using a model in this? I copied your exact steps but when i upload an image in the variation/enhancer section with vary subtle setting, my image does comes out in the clothes, pose, hair style etc but it is wayyyy more real. Your comes out exactly like your original image but with the additon of smiling etc due to your prompts. NOTE: I did also use your smiling prompt and that worked but just my own is wayy to real compared ot my original image. Why is that ? Is it a model setting perhaps or what may be the reason?
No, for this i used the default model, which is juggernaut. It's hard to say without knowing what image you are using and other settings. You don't have any extra styles selected? You are using the vary (subtle) and not vary (strong)? not using a refiner? only other suggestion would be to try to go into the debug>debug tools> and adjust the denoise in "Forced Overwrite of Denoising Strength of "Vary" setting. try putting it at around 0.5 and then lower if its changing too much or raise a little if not changing enough.
I used every thing you did, Juggernaut, kept it at subtle, no refiner, tried with and without stock lora that comes with it upon setup. I am wondering if it is because I am using a photo i uploaded from dalle3 (not one that was generated on fooocus). Also note: I used same settings as the ones i mentioned above in this msg in the image prompt section and tried with various weights/stop at and tried with pyracanny and image prompt just to see if it made a different and it all comes out looking like my girl original uploaded image but just a very real version of the girl in my image. Also, I used your method of changing the "forced overwrite of denoising strength of "vary" to 0.5 and other variation and nothing rly changes it. Still a very real version of my girl :( Anyway I could show you what i mean?@toAI
I used every thing you did, Juggernaut, kept it at subtle, no refiner, tried with and without stock lora that comes with it upon setup. I am wondering if it is because I am using a photo i uploaded from dalle3 (not one that was generated on fooocus). Also note: I used same settings as the ones i mentioned above in this msg in the image prompt section and tried with various weights/stop at and tried with pyracanny and image prompt just to see if it made a different and it all comes out looking like my girl original uploaded image but just a very real version of the girl in my image. Also, I used your method of changing the "forced overwrite of denoising strength of "vary" to 0.5 and other variation and nothing rly changes it. Still a very real version of my girl :( Anyway I could show you what i mean?@JumpIntoAI (sorry i deleted your reply to me)
@seanknowles7987 I figured it was an image not created in fooocus, however I have had luck pulling random images in and doing the same thing.
My other suggestion is perhaps try masking the mouth area and using inpaint to change the expression. Just dont mask the entire face and see how this comes out. You can upload the image to something like imgur.com and post the link here if you want me to give it a shot. no promises 😉
Hey I noticed that my FOOOCUS output folder only has two days worth of photos. I have been using fooocus to make photos for almost two weeks but I only see two days store. Why is that?
Hmm, is fooocus installed on your computer or are you using the google colab version? Colab version you will need to manually save them otherwise they will dissapear.
If on your computer, not entirely sure. Did you make any presets recently or any changes to the config.txt output path?
@@JumpIntoAI installed on PC-windows. made no changes to any configs.
@@tristanthedarkstar Not entirely certain what would cause that. I'm sure your hard drive isnt full? are all of your images currently saving since then?
@@JumpIntoAI it's not full.
I’ve trained my Lora’s in 1024x1280 aspect ratio.
Can i use this aspect somehow in fooocus?
How do i set it? I dont see a custom option for aspect ratio’s
you might have seen this here: ruclips.net/video/zWLknfrhfW4/видео.html
but also I didn't mention (will put in another video) is you can change this temporarily in the the debug menu>debug tools. scroll down to where it says "Forced Overwrite of Generating Width" and "Forced Overwrite of Generating Height" and put any resolution in there you want.
@@JumpIntoAInice to know as well. I’m going to test out fooocus now. I think I’ll be using auto and comfy and focus for different things. Photobashing especially in focus. I wish there was a tunnel from Photoshop into focus to be able to quick draw to inpaint denoise. Also, on a different note, I discovered Breadboard yesterday. Looks like the best Ai image viewer for metadata. But it’s sort of on hold by dev. Do you know a good image tool like that? I’m currently using Allusion, but it also is not being developed. Open source projects 😮
Yeah I liked Breadboard, wish it was continued. I been using DiffusionToolkit (it's windows only I believe though) and that's been good enough for now.
great video, thanks!
broo i need all the characters like shinchan doraemon ben10 nad such cartoons which model or lora should i install
How do we make her turn her head so they arent all from straight forward?
Does it need internet to run??
The colab version is online only.
If you download this and use on your own computer then no. Once you have the basic models and all components downloaded (which wont download until the first time you use them) then it can be used completely offline.
@@JumpIntoAI Thanks
Quite helpful. Thx
No problem!
9:21 fucking legend man ... when i do ututube videos i domt either
please make a tutorial about how we can train our own and personal models
I will have some things with Dreambooth, but that's down the road for now. :)
is there any Alternative for upscaling model that has the power of the "magnific AI Upscaler" using SDXL? @@JumpIntoAI
Short answer is yes. Depending exactly what you want to do. There are lots of discussions out there on the topic. can check out these 2 reddit threads. www.reddit.com/r/StableDiffusion/comments/18msbgb/magnific_ai_but_it_is_free_a1111/
www.reddit.com/r/StableDiffusion/comments/18mqksc/after_a_full_month_experience_with_comfyui_making/
GOD DAMN!