Advanced Inpainting Tricks - Denoise Strength
ฝัง
- เผยแพร่เมื่อ 19 เม.ย. 2024
- In this video I am going to show you some advance inpainting tips and using the denoise strength. You can use this to help guide your inpainting for better results.🎨 #Fooocus #StableDiffusion #DigitalArt #ImageManipulation #CreativeTools
Fooocus is a great interface for using Stable Diffusion and is very easy to get started. It also will run on systems with only 4GB VRAM.
I am running on an i5 32GB RAM Nvidia 3070 8GB VRAM
If you find the video useful and would like to tip, TH-cam does have the Super Thanks option. Those are greatly appreciated so I can have my coffee!!
☕ BUY ME A CUP OF ELECTRICITY: ko-fi.com/kleebztech
Download Fooocus: github.com/lllyasviel/Fooocus
How to install Fooocus video: • Stable Diffusion - Dow...
Fooocus Beginners Guide: • Super Easy Stable Diff...
Fooocus Upscales and Variations: • Stable Diffusion - Foo...
Checkpoints and LoRAs the Basics: • Stable Diffusion - Che...
Zoom with Outpainting: • Stable Diffusion - Zoo...
Fooocus Poses and FaceSwap - Image Prompts: • Stable Diffusion - Pos...
Inpainting with Fooocus - Don't Regenerate, Fix!: • Stable Diffusion - Inp...
Background Replacement with Fooocus: • Stable Diffusion - Bac...
Fooocus Tips and Tricks (and AI Hands) • Overlooked Fooocus Tip...
Stable Diffusion - Styles in Fooocus - How do they work?: • Stable Diffusion - Sty...
FaceSwap and Consistent Character TipsPart 1: • Stable Diffusion - Fac...
FaceSwap and Consistent Character Tips - Part 2: • Stable Diffusion - Fac... - วิทยาศาสตร์และเทคโนโลยี
You are the most down-to-earth, no-bullshit, no frills, detail-oriented AI instructor I've found yet. I already subscribed last night but needed to comment on your style, so here's a massive thanks for a terrific series. I'm definitely looking forward to more.
You are welcome. And thank you for the feedback. 👍
You deserve wayyyy more subs and recognition my dude! your tutorials and guides are amazing!!!
Thank you!
You are stamping out lessons like hotcakes ))
I admire your ability to work hard and be productive! Thank you so much for your work!
Thank you for another excellent video!
Thanks for watching!
Rodney, you rawk.
Thanks for sharing ❤I really like ur videos, and I hope u drop video about landscape or generate something in the forest, like camping...❤
Will keep that in mind and add to my list although right now I can't think of a video topic to do that but maybe something will pop into my head.
As always, I already know this video is going to be amazing. Keep it up 👌
Thanks! Will do! Just hope I don't scare people off with the darker images... lol.
Hahaha even for those who might have a few nightmares because of these "darker images", I still think the content is worth it 😂
I used this technique so much I modified my Fooocus program to have an image editor tool where you can do basic painting.
I love it!
How did you do that??
@@KLEEBZTECHhey, love your videos. Can you tell me what GPU you are using?
Nvidia 3070 @@jeeadvanced4719
@@jeeadvanced4719 Fooocus is open source python code. If you know python you can do anything with the gui.
great content, as always
Thank you!
So there is still a place for photoshop! This technique can also be applied to clothes swap to help the AI work out how you want clothes to look on a body.
And I am sure there always will be a place for it. Good tip for the clothing!
Wow, this is awesome, thanks!
You're very welcome!
Thanks!
You are very welcome! And thanks for the tip! Those make my day!
Great explanation!
Thanks!
7:48 i would generate the dress, when i have something i like then i would generate the legs.
I debate with myself on that sometimes. I always did it that way but then started doing multiple areas if close to each other because worst case only one part comes out good but I might get lucky and both do.
Thanks, i did struggle sometimes with clothes, or wings etc...
This can help in many circumstances. Especially if you are better at painting than me. lol. Although it really doesn't have to be great.
5:05 lol this is awesome!
I am glad you think my painting skills are awesome! 😉
This is the BEST fooocus advice channel on YT!
I can repaint a larger pair, but today I spent many hours trying to just add a tree or two here or there in my landscape images. Using various checkpoints, even GB sized ones dedicated to trees, they'll start to appear then vanish. I too resorted to clone brushing a full tree in another photo editing software, and try inpainting again, only to get something that looks like yarn in a food blender. Now thanks to your detailed explanation to denoising and expanding the area of influence I hope to create some gorgeous landscapes! Impressed you were able to simply change the color of an apple with a colored mask, but when I try similar tricks adjusting human skin tones, they end up the color of bananas and apples rather than the ethnicity I hope to bring to my art. BTW which window is best to add text to for replacing objects?
Are you talking about when using the modify content option? Text in the main prompt will impact things and I usually leave blank. But I don't use that option much.
Thanks for the drawing trick
Welcome 😊
@@KLEEBZTECH I create NSFW art. Will this trick work with other NSFW models too?
I can't see why not.
Interesting concept. I'm wondering if this is a good use case for adding an image editor to the extensions in Automatic1111.
Sensei Kleebz.
Godlike
Good stuff! It would be good if they add a basic brush so you can do that inside of Fooocus
Someone else in the comments mentioned the created their own custom one with that.
Could you make a video showing how to modify the pose of the head or arms or some part of the body, but keeping the rest of the image, clothes, background and so on? The idea is to make a gif.
I will take the idea into consideration. 👍
Hi. Here is a solution to the problem of freezing when you using OBS. If you have two computers, you can run Fooocus on the main, more powerful computer, and share it (in fooocus settings). And on the second computer you need to open the shared Fooocus in the browser and record the screen using OBS on this computer. The neural network processing will take place on a powerful PC using all its power, and the OBS screen recording will take place on another PC.
But you need two PCs :)😅
Thanks! Technically could do one better since I do have a capture card in a system but that is currently being used for something else. I only run into issues when upscaling happens so usually not much of a problem.
thx for the tutorial mate as usual i always try it by myself and learn from ur videos 👍and do u have any idea on how can i make juggernaut x as a default model in fooocus to run it on collab or saggmaker
I have just started with playing around with it on colab and I do know you have to add additional lines to have it download other checkpoints. I am still learning that myself actually. On mine I moved the line to run it to the bottom and added a section before running to download the checkpoint I want to use.
Example imgur.com/a/EWtwhCq
@@KLEEBZTECH yup but i don't want to add additional line to add new checkpoint what i mean is when we run fooocus we get default version 8 and i want to replace it with version X without adding it as new checkpoint.. i hope you get it
Thanks for the info. Do you know if it's possible to change the Fooocus inpaint engine from v2.6 to one downloaded from CivitAI? I downloaded an inpaint model for Juggernaut XL v9 but when putting it in the Inpaint folder of Fooocus, I can't figure out how to make the switch within Fooocus. The reason I ask is I find the resolution / quality of the outpainting, when extending an image to the left, right, top or bottom, to be lower than the original image that I want to extend.
I don't think it is. Although I am not sure you would want to since Fooocus really has one of the best inpainting engines and is designed to be used with any checkpoint you have selected. No need for a special model.
@@KLEEBZTECH May be due to the fact that I use non-standard resolutions (1344 x 1344) whereas the inpainting is limited to 1024 x 1024 and has to "upscale" without improving the quality
Just saw the updated comment. When outpainting are you doing just one side at a time? It sounds like you may be but if not I have found it is best to do only one at a time for best results.
@@KLEEBZTECH Yes, I've tried one at a time and gives better results than when extending all four sides at once, but still doesn't have the same level of fidelity and detail as the original image. Oh well, thanks for responding.
Yes, this entirely possible, although a little fiddly. The Improve Detail Inpainting method never uses any of the Fooocus specific Inpainting engines. Instead it uses whatever base model and refiner (if any) that you have specified under the Model tab.
If you go to the Advanced tab, check Developer Debug Mode and then the Inpaint tab. You now have complete control of the Inpainting process. You can make the Improve Detail method work like either of the other two methods, but still you use the base model of your choice.
To emulate the Default Inpainting method, make sure "Disable initial latent in inpaint" is NOT checked, make Inpaint Denoising Strength = 1.0 and Inpaint Respective Field = 0.618.
To emulate the Modify Content Inpainting method, make sure "Disable initial latent in inpaint" IS checked, make Inpaint Denoising Strength = 1.0 and Inpaint Respective Field = 0.0.
To return to the normal Improve Detail Inpainting method, make sure "Disable initial latent in inpaint" is NOT checked, make Inpaint Denoising Strength = 0.5 and Inpaint Respective Field = 0.0.
I should also mention that all three methods always use your base model and refiner - but the Default and Modify Content methods use the special Fooocus Inpaint engine in addition to the base model and refiner.
***** UPDATE: it looks like all you have to do is go to the Inpaint Engine selector under the Inpaint tab in Developer Debug Mode and simply select "None" instead of V2.6 or the other two models. Now Fooocus will use only the base model and refiner for all three Inpainting methods. Also - that should apply to Outpainting - I just noticed that was the purpose of your question.
I'm leaving the other information there because it could be a useful reference for the differences between the three methods.
Hello, is it possible to give the face wetness, water effect, blood effect etc. without interfering with the original face?
Not really but maybe using something like Faceswap as well in combination...
@@KLEEBZTECH Well, is it possible to nude or change the body type without interfering with the original body pose? For example, I want to make myself half-naked and muscular in a photo with clothes on. While doing this, it changes the pose almost completely, but it does not make any intervention in the improve detailer part, how can I do this? Isn't there a way to remove clothes in the improve detailer section? but then it may not interfere with the original pose.
You are not going to do that using the improve detail option. Regular inpaint and masking just the area of the clothing would be the only real option. Also could do in portions if masking too much area is changing the pose. For example if long sleave shirt then just mask the arms area and ask for short sleave. Then when you get that correct mask the new image from there without masking any of the bare skin part. It should keep the pose for the most part since you will be leaving mush of the person untouched. @@DnmHsb
@@KLEEBZTECH Yes, I do this, but it is tiring. I hope that such problems will disappear in the future when artificial intelligence develops further. Thanks for your response.
Bro can I ask how can I generate disable peoples in fooocus such as Girl without hand or leg 🦵 any Checkpoint for this bro
Funny you mention that.... Did you post something similar on the Github discussions? I was already seeing what might work because of post there.
"bilateral forearm amputee" would be the type of thing to start with. That is the technical term for missing forearms. Hard part is not mentioning the terms arm and leg for example.
Fooocus is fully local right? Like no one sees the pics you make, yes? Cause, i mean, its pretty easy to make some wild s.hit.
If you download and run on your computer then yes. It does need to have a connection for some things since it needs to download the models when you first use it or if you go to use inpainting for the first time it will need to download the models it needs. But otherwise yes fully local.
I came across the fact that it’s difficult to add a chain or rope. Even if you draw it in the photo editor, Fooocus doesn’t see it, you have to draw it in small pieces, which doesn’t look realistic.
Interesting. Will see if I can come up with a good way. Have you tried with different models?
That what I was looking for, but didn't know at the time, when I was struggling just as much as you were. In stable diffusion, there is an option sketch inpaint. Essentially its the same step within the program. Have you tried same trick with hands so far?
It can work with most things including hands. It is really how the improve detail function is working. But for some things it can me tricky with the colors and finding the right denoise strength.