This is great and works better for hands, I actually learned later that using an upscaled image works best when you want to inpaint. I'll upscale my images all the way up to 4k, and it's a bit slower, but totally necessary because upscaling messes up the images, so it's better to inpaint when the image is in its highest resolution.
@@JumpIntoAI Yep, it works better that way because I've noticed the quality in the skin and clothes decreases. It shows well when you zoom in on the picture and it ends up looking like an oil painting. If you inpaint after upscaling it brings back the quality of everything. I didn't discover this until about 2-3 weeks ago and now I inpaint after upscaling, it takes more time, but it's better to keep the quality.
man you are a lifesaver, new sub! There are always some images that can be pretty but the hands were horrible, with this I have been able to save some, God bless Illyasviel
Using Fooocus and a photo editor was always a way for me. As a photo editor (before Fooocus) it naturally came to my mind that for best results I can combine the two tools, depending on what I need. It often happens, that I have a number of generated variatons of an image, or fixed elements in the image, which land in Photoshop as layers, and then it's all about masking, so... multiple versions of the same image with various defects are composed into one final and perfect. :) I recommend this approach. Easier than struggling sometimes too much with only Fooocus.
Yep. People immediately shy away when mentioning photoshop as a step. And I understand since I see a lot of posts telling them to "learn to draw the hands" and that's not necessary. The better you are at using those tools the easier it is to get a good result, but as shown here, even a quick and dirty copy and paste can be blended in.
The best thing about the odd hand generation that I have found is it works out decent for alien image generation where unusual number of fingers may add to the feel of potential realism or artistic style.
I think you could have probably gotten a decent generation by just using the canny image of the image you generated as a canny and just rerunning the prompt, but thanks for showing your process.
Hello, I want you to explain in the video how to add VAE in the fooocus. Especially the "Animagine XL V3.1" model where there is a note written "This checkpoint recommends a VAE, download and place it in the VAE folder." Great videos and keep it up don't stop
Yes I think its confusing when you see the label (hands) in the improve detail setting. But really it is there to improve a hand that has the correct form (right shape and fingers overall) .
hey bro can you please make on how to completely blend subject with the background cause imrpoved details and other loras in inpainting are not able to do it 😥and also make a tutorial on Defooocus casue it has photopea with it
I really wish I had better results with that. But using the exact same settings and running a large batch of images and then comparing. Didn't really notice a big difference overall. But that's just my experience with it
Thanks for info! I would normally use all the adjustments but I wanted this to be the least bit intimidating/confusing. Even mentioning Photoshop makes people give up, so I used the most basic simple approach to not scare anyone off, and too show even an ugly amateur copy and paste can still work. lol. 😉
@@JumpIntoAI Understood! Like you, I find that Photoshop makes a great mate for this program, but then I used Photoshop first. :-) Thanks for making this video. I learned a lot.
Thanks so much. Very informative and clear. I try some architecture / angular stuff. Do you know how to upscale, or "vary subtile" which stricly respect the geometry ?
Bro can u do tutorial how to re create image of House or room or anything on fooocus that’s make our new Image generate look like or similar to pic that’s we upload ?
Hello! Could you help me please? I have an issue with running Fooocus. Its says RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver. I have amd gpu.
When I do inpaint to improve hands for example, I have the problem that the part of the mask comes out in a different color than the original skin, always lighter and the shape of the mask is noticeable.
I would try prompting for darker skin in the improve detailer prompt. Another thing is assuming you are using improve detail, you might want to try to go into debug >inpaint and raise the "Inpaint Respective Field" a little (like .1 to start and increase) This lets the AI see more of the image and blend better at the cost of detail (higher it is more it sees but less resolution for the mask)
Hi, is there a word or character count for the prompt? It seems that sometimes, depending on the prompt size, Fooocus just ignore part of the prompt. And I have the impression that ig ignore the last words... Does Fooocus prioritize the words in the prompt differently? Like, the first words have more priority and the last words less?
Sadly no character counter in fooocus. SDXL checkpoints in general do better with shorter and to the point prompts. I don't think fooocus changes the allowance length or priority. But do remember having any "styles" checked does add to your prompt length. the way the prompt is interpreted compared to something like a1111 seems the same for me. I've used exact same prompts in there with near identical results. But longer prompts will start to fall off in any case.
Might have showed us how you masked the finger towards beginning. And what masking is. What is outpaint direction? What setting were you using on that step?
Thank you a lot for all the updates and tutorials! I'm binge whatching almost every one! But I have a question, some models have a description like this: "Make sure you load this model with clip skip 2 (or -2 in some software), otherwise you will be getting low quality blobs." Where is this option in Fooocus?
Right now clip skip is not implemented. The developer tried to only have simple options to not overwhelm newbies. The Same with not being able to choose a VAE. I haven't seen any mention of adding it yet.
This can work the same way with bird feet, but the improved detail setting will almost always need "inpaint respective field" to be turned up some (amount depends on the picture) AI isn't as good at detecting bird feet, but turning that setting up will help it see more of the image and then give you a better result. Won't always work but a better chance if done this way.
![[UA] Eternal Fire проти NAVI | EPL Season 20 Malta](http://i.ytimg.com/vi/chvlsDygrZM/mqdefault.jpg)
Fooocus download page: github.com/lllyasviel/Fooocus
I can't find a better fooocus teacher atm)
Truly cool tips, thank you!
Happy to help!
This is great and works better for hands, I actually learned later that using an upscaled image works best when you want to inpaint. I'll upscale my images all the way up to 4k, and it's a bit slower, but totally necessary because upscaling messes up the images, so it's better to inpaint when the image is in its highest resolution.
Yep, it might seem a little backwards but I do most of my inpaint after the upscale.
@@JumpIntoAI Yep, it works better that way because I've noticed the quality in the skin and clothes decreases. It shows well when you zoom in on the picture and it ends up looking like an oil painting. If you inpaint after upscaling it brings back the quality of everything. I didn't discover this until about 2-3 weeks ago and now I inpaint after upscaling, it takes more time, but it's better to keep the quality.
you're very good at delivering explanations that are usually not easy for others to do, great job.
man you are a lifesaver, new sub! There are always some images that can be pretty but the hands were horrible, with this I have been able to save some, God bless Illyasviel
Using Fooocus and a photo editor was always a way for me. As a photo editor (before Fooocus) it naturally came to my mind that for best results I can combine the two tools, depending on what I need. It often happens, that I have a number of generated variatons of an image, or fixed elements in the image, which land in Photoshop as layers, and then it's all about masking, so... multiple versions of the same image with various defects are composed into one final and perfect. :) I recommend this approach. Easier than struggling sometimes too much with only Fooocus.
Yep. People immediately shy away when mentioning photoshop as a step. And I understand since I see a lot of posts telling them to "learn to draw the hands" and that's not necessary.
The better you are at using those tools the easier it is to get a good result, but as shown here, even a quick and dirty copy and paste can be blended in.
Thank you very much, i learn a lot with you.
The best thing about the odd hand generation that I have found is it works out decent for alien image generation where unusual number of fingers may add to the feel of potential realism or artistic style.
Finally, thanks so much for this extremally apricated
That's a great lesson! Thank You!
Thanks! Like always great video and inspiration!
I think you could have probably gotten a decent generation by just using the canny image of the image you generated as a canny and just rerunning the prompt, but thanks for showing your process.
Hello, I want you to explain in the video how to add VAE in the fooocus.
Especially the "Animagine XL V3.1" model where there is a note written "This checkpoint recommends a VAE, download and place it in the VAE folder."
Great videos and keep it up don't stop
Currently, there is no way to change the VAE in fooocus, hopefully this is something that the developers can add at some point.
Instant sub. Great content
Much appreciated!
omg, i never used the default inpaint XD, thx^^
Yes I think its confusing when you see the label (hands) in the improve detail setting. But really it is there to improve a hand that has the correct form (right shape and fingers overall) .
Very usefull tips, thanx, bro!
hey bro can you please make on how to completely blend subject with the background cause imrpoved details and other loras in inpainting are not able to do it 😥and also make a tutorial on Defooocus casue it has photopea with it
perfect!
Lora - Perfect hands :)
I really wish I had better results with that. But using the exact same settings and running a large batch of images and then comparing. Didn't really notice a big difference overall.
But that's just my experience with it
Curves is the best way to get the lighting the same across a picture. You move the curve, and it affects the light across all layers.
Thanks for info! I would normally use all the adjustments but I wanted this to be the least bit intimidating/confusing. Even mentioning Photoshop makes people give up, so I used the most basic simple approach to not scare anyone off, and too show even an ugly amateur copy and paste can still work. lol. 😉
@@JumpIntoAI Understood! Like you, I find that Photoshop makes a great mate for this program, but then I used Photoshop first. :-) Thanks for making this video. I learned a lot.
Thanks so much. Very informative and clear. I try some architecture / angular stuff. Do you know how to upscale, or "vary subtile" which stricly respect the geometry ?
upscale (fast 2x) will be the best bet for an upscale with no change.
Bro Can u show me how too re create any image to the similar or look like original one ? I need this so bad
Bro can u do tutorial how to re create image of House or room or anything on fooocus that’s make our new Image generate look like or similar to pic that’s we upload ?
These same methods will also work for the feet, right? ?
I don't see why not, I haven't had to repair too many feet, but it does work on bird feet, so... ;).
Hello! Could you help me please? I have an issue with running Fooocus. Its says RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver. I have amd gpu.
This Is elso my problem
Buy an nVidia card to solve that...
how do you zoom in when inpainting, thats the trick I wanna know lol
shift + mouse wheel
keep your mouse on the image
Hi is it can generate image michael jackson? Like looks a like him
When I do inpaint to improve hands for example, I have the problem that the part of the mask comes out in a different color than the original skin, always lighter and the shape of the mask is noticeable.
I would try prompting for darker skin in the improve detailer prompt. Another thing is assuming you are using improve detail, you might want to try to go into debug >inpaint and raise the "Inpaint Respective Field" a little (like .1 to start and increase)
This lets the AI see more of the image and blend better at the cost of detail (higher it is more it sees but less resolution for the mask)
@@JumpIntoAI Thanks
Hi, is there a word or character count for the prompt? It seems that sometimes, depending on the prompt size, Fooocus just ignore part of the prompt.
And I have the impression that ig ignore the last words... Does Fooocus prioritize the words in the prompt differently? Like, the first words have more priority and the last words less?
Sadly no character counter in fooocus. SDXL checkpoints in general do better with shorter and to the point prompts. I don't think fooocus changes the allowance length or priority. But do remember having any "styles" checked does add to your prompt length.
the way the prompt is interpreted compared to something like a1111 seems the same for me. I've used exact same prompts in there with near identical results.
But longer prompts will start to fall off in any case.
Hey, about inpaint, do checkpoints matter? And if yes which one do you suggest for the hands? Thank you!
Ideally, you want to use the checkpoint that created the original image.
@@JumpIntoAI Great, thanks!
🙏🙏🙏
What are u using sry im noob
In this Im using fooocus to generate images and photpea.com to edit them. both free.
Harris Margaret Gonzalez Robert Lee Dorothy
I think you didn't put enough commercials between your video.
Every 1 minute 🤦🏻♂️
What a noob
I cant drag the image from the top, when I click it it just opens full screen...
thanks a lot bro, great video and details how to fix the most difficult thing to fix. Excellent 👍👍
Thank you. It's a little different question, but how can I use the eular a sampler in fooocus?
If you go to the advanced tab>developer debug menu> debug tools tab> scroll down and you can see the sampler pull down with euler a.
Might have showed us how you masked the finger towards beginning. And what masking is. What is outpaint direction? What setting were you using on that step?
Awesome video
Thank you a lot for all the updates and tutorials! I'm binge whatching almost every one!
But I have a question, some models have a description like this:
"Make sure you load this model with clip skip 2 (or -2 in some software), otherwise you will be getting low quality blobs."
Where is this option in Fooocus?
Right now clip skip is not implemented. The developer tried to only have simple options to not overwhelm newbies. The Same with not being able to choose a VAE.
I haven't seen any mention of adding it yet.
You're a legend 🏆
how to fix feet bird
This can work the same way with bird feet, but the improved detail setting will almost always need "inpaint respective field" to be turned up some (amount depends on the picture) AI isn't as good at detecting bird feet, but turning that setting up will help it see more of the image and then give you a better result.
Won't always work but a better chance if done this way.