Stable Diffusion - Inpainting, Outpainting, Upscaling & More!
Vložit
- čas přidán 19. 04. 2024
- Ever wanted to do a bit of inpainting or outpainting with stable diffusion? Fancy playing with some new samples like on the DreamStudio website? Want to upscale or improve faces from within a web interface? Want to make stable diffusion images via your iPad while in bed or on the sofa? Then this is the video for you! :)
Update! There is now a requirements.txt file so you can "pip install -r requirements.txt" :)
Repository paths have now been fixed, as they can go in the directory for repositories!
Contents:
0:00 Overview
0:58 My Environment
3:59 Anaconda environment
6:35 Run
6:51 Run via your iPad
7:46 Web interface
8:35 Tiling
9:35 X/Y Plot
10:35 Prompt matrix
12:00 img2img and Inpainting
14:59 Loopback
16:12 Upscaling
18:29 Outpainting
Links:
Ubuntu - ubuntu.com/download/desktop
Nvidia Drivers & CUDA toolkit - developer.nvidia.com/cuda-dow...
Anaconda - www.anaconda.com/products/dis...
PyTorch - pytorch.org/get-started/locally/
GitHub - github.com/AUTOMATIC1111/stab...
GFPGAN model - github.com/TencentARC/GFPGAN/...
SD textual inversion: • Textual Inversion - Ma...
SD Dreambooth in Windows: • Train on Your Own face...
Animate images: • Create your own AI ani...
Make Stable Diffusion Videos: • Make Animations Using ...
If you can't install Stable Diffusion: • How to use Stable Diff...
Use AI to clone voices: • AI Voice Cloning - Tor... - Věda a technologie
Thanks man! Super easy tutorial to follow.
The holy passion of friendship is of so sweet and steady and loyal and enduring a nature that it will last through a whole lifetime, if not asked to lend money.”
The tiling thing is going to be absolutely insane for textures
Yup!
Check the new Blender option
it would be awesome to make custom background images for youtube videos
Thank you, AI updates are happening so fast it's difficult to stay on top of it all. Your videos are much appreciated.
Excellent work David Brent! Keep em coming!
Man, you don't know how grateful I am right now! THANK YOU SO MUCH!!!
The loopback thing is fun to use with low noise and steps but a high number of images to let you get a sequence you can turn into an animation showing the image getting gradually altered
Wow, so well done! Thanks for posting, man! Very helpful.
Hands down tNice tutorials is one of the best tutorials I've seen on any subject, ever! Clear, concise and incredibly informative and easy to follow.
excellent video friend was just what I was looking for
Thank you another excellent video. I must admit installing this on Win11 Anaconda was a right faff, but I got there in the end.
It worked perfectly!!! Thank you so much
Thanks A LOT for your clear explanations !
Glad you’re enjoying the things!
Stable Diffusion - Inpainting, Outpainting, Upscaling & More!👍👍👍👍👍👍👍👍👍
owaa amazing and helpful video with good sound and video quality with proper instruction.its working on my device.
Worked for me thank you!❤️
I was always fascinated by trap soft, been listening to trap long ti and i finally decided that i will try to make my own but i was so
I want try setting up this, I will probably fail in something, but this is amazing, I have to try. As someone have already said, this stuff are evolving so fast, it's amazing. I was writing to a friend just a couple of days ago that it would be a matter of a short time for the inpainting feature to come to Stable Diffusion locally as well, and it was actually already doable. Great.
My Nvidia 1080 will hate me in the next few days 😅
Ikr! Rip GPU! 😆
Wow. This is a very popular video. This look’s like it would be so much fun. I can’t help but think of creating a story teller to capture an audience. A little Alice meets Dorthy on an island of lost boys.
Thanks. It was nice to hear you. Ok, I’m done invading your world for a while. Keep up the brilliant creativeness my nerdy friend, it is amazing.
It is indeed fun! I did a very bad example comic on the twitters 😆
@@NerdyRodent 😂 I don’t have twitters lol. I’m watching professor of rock channel talking about Led Zeppelin and playing bit’s of the song’s. Always fun
FACTS!!!!!! Thank you so much man, you're the best!!!!!!!!!
Great video! Thank you.
Do you know if there's a way to script multiple images with different models?
man. Thank you!
Thanks. I have been searching for inpaiting for days. And now it's released.
Is this censored in some sort?
Thanks for sharing
Cool! Downloaded))
good job male, thanks!
No problem!
Thanks!! I would love to see textual inversion..
hlky has it. you need more than 12gb vram
Thank you so much!!! czcams.com/video/WsDykBTjo20/video.html
works! I love this game is my favorite
Mine too! XD
Awesome guide, could you explain more about Mask options?
Hahah, you pamper us! Update again! View bot king
it looks good
fantastic video! Thank you. I got caught on your pronunciation of Ubuntu - it's pronounced "Oo-boon-too", kind of like it rhymes with "tutu." South African over here, got to speak up for the team ;)
I love your “yah yah yah” lol. Love your videos. Btw, what’s your computer specs, like how many ram and vram?
32 & 24 😉
I cant wait, can you please upload your next video? :)
Hi, can you make a tutorial about how to use correctly the new "im2img Alternative Test" script in the "AUTOMATIC1111 - stable-diffusion-webui"?
Thx.
Thanks for the guide! The space is progressing so quickly, will this application auto-update whenever there is one or do I need to delete it and download the new one? Sorry I'm quite unfamiliar with how these things work
Just “git pull” in the directory to get all the latest updates rather than “git clone”
Amaizing
hey! I just installed the webui but it doesn't have sampling methods radio buttons, tiling checkbox, redraw option and etc
Hi Nerdy Rodent, Just a question, If my purpose is to generate street photography scenes without a main portrait subject in the foreground but I am using street photography references with subjects in the foreground (and again, I am interested as a result only in the background image), do I a) use generate via controlnet and remove the subject at the very end (assuming subject is necessary for SD to build the street scene) or b) I use reference with/out subject and inpaint the subject out? Not too sure. Been testing. What are your thoughts? thanks!
Personally I’d use a street scene in ControlNet similar to what you want, and ensure the prompt has few human related keywords
The image in my inpainting UI is very small, its not scaling to fill the frame. In your video you can zoom the area also. this option seems to be gone. Very hard to inpaint when I cant see what Im doing. Any idea? installed today 10/27, again not the size of the actual image, just the UI inpainting image.
thanks mate tNice tutorials was a big help
Are you using the Producer version, if so are all the effects and tutorial you use in the video included as standard with the Producer
This is Stable Diffusion
I'm a newbie soft soft (20.9) user, and I'm on Mac 10.14. Would you please help about how to select
Hey I'm really new to tNice tutorials, my GMS doesn't soft the sa, more like a continuous buzzing soft, does anyone know what I do? I
I'll probably not get an answer but still, when I opened my soft Roll I couldn't place the "notes" or however you call them that finely, by
Any chance you could cover the complicated process of teaching Stable Difussion new prompts?
I wanna fine tune it to understand specific characters from media better, by feeding it images.
Sure!
@@NerdyRodent ooh i really anticipate that! i just stumbled into that whole topic a few days ago when i found out about midjourney i was blown away how easy it was to enter....then i heard about stable diffusion and that you could host it on your machine and i was like...yeah if you're a computer scientist. And now i am running it on my own machine and driving my elecricity bill sky high lol
@@Nepoxification Welcome to being a nerd! 😉
I'm currently using hlky's fork of the project and from what I saw, it does quite a good job as well, even though it seems to be less stable as automatic1111's fork. Did you try it out yet and if so, what's your opinion?
I mostly prefer the CompVis release as I find the code easier to play with 😉
@@NerdyRodent I can roughly understand your opinion, but their release is hardly usable at higher resolution with 12 or less GB VRAM, so it's not really an option for the average 3080 Ti or less user.
@@BlankFX Got a 3090 for AI stuff years ago, so it’s all good with me! 😉
I was wondering the same thing. Some people were saying the Automatic fork was best... their are 2 other forks that are even more optimized to create way larger images though (even for the ultra wealthy like Nerdy Rodent with 3090's)... I might get crazy and try to combine the different forks... who knows, so much changes from day to day.
automatic1111 has VERY clean UI and is easier to install, because it doesn't hide models 3 levels deep within its hierarchy. hlky has some experimental stuff, but it is very messy overall.
In the latest release automatic1111 added ability to paste images into img2img (just Ctrl+V at img2img page), so there's not much point to stick with hlky, unless you really want *.yml for every generated image.
So, I got into the webui, but anytime I click generate, it shows me this:
RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'
Can anyone help please?
For some reason I can only move generated pics from the main tab to extra. Is send to img2img and send to inpaint not working? Either way, amazing tools my man!
Seems to work for me, but it does get updated a lot so quite possible it got broken at some point!
Thanks
Can you recommend a PC for running this? Should I buy 32vram?
The more VRAM the better at the moment. If you can afford an A6000 then go for it!
.... To quickly test tiling in gimp, use Filters-->Map-->Tile.
Ooh. Thanks for the tip!
What are the taming transformers for anyway? I've never downlaoded them used it.
Not sure, but hey - I like to get rid of warning messages 😆
by the way... what's the name of this model? I want to learn more about his potencial but I don'0t know how to call it when search for more details. Thanks a lot!!!
It’s called “Stable Diffusion 1.4”
is there any options or any way that stable diffusion use 2 VGA ?
Yup! There's a quick overview at outerbounds.com/blog/parallelizing-stable-diffusion-production-use-cases/
What video card are you using?
3090 here!
How do you get the GUI to be in dark mode? I don't see it in the settings.
Browser dark mode 😉
do you know if the web ui saves prompts anywhere?
Yup! There is an option in settings to turn that on :)
Thank you, is there a way to run stable diffusion on a 6 gigs vram intel hd, i5, 12 ram, I'm using Manjaro, I did try stable_diffusion.openvino repository, but when I run the prompt, it returns "killed".
‘Killed’ usually means the process had to be terminated as you ran out of system RAM.
It has "Turbo Mode" enabled by default, i not, in what file we activate it?
An image in 3 seconds is pretty turbo 😉
I can't afforded soft soft. what alternative can I get?
This is all free software ;)
Does anyone know about how much space this all takes up on your c drive. Python and everything
Is there a way to install anaconda on a drive other than c and have this still work?
I have it on D:
keren, sehat selalu bang, ditunggu share cara instal aplikasi macbook yang lain
Sayangnya saya tidak punya macbook :(
Inpainting didn't seem to work for me but everything else did. I can't see why but it seems that the mask isn't being passed through properly because when i selected to inpaint not masked the image returned was black making me think something isn't working with the mask. That said I can see the mask appearing when i draw in the box :/
Strange! It does get updated often, but we’ll soon be getting the official inpainting code 😀
@@NerdyRodent Thanks I eagerly await the release of the official inpainting code then 😄
On Firefox image inpainting seemed to be doing totally nothing, but works as expected on another browser (like Microsoft Edge)
Which GPU are you using if I may ask?
Nvidia 3090 here
Is it possible to import an existing image and then inpaint within it?
Yup!
once again your voice so silky smooth
Actually 11% raspy
hmm I wonder why I don't have the tiling option. Then again my UI seems to have some tabs missing e.g I don't have PNG info or settings. I must of missed something.
Old release?
@@NerdyRodent I just used the version you linked. Tbh I managed to break the thing trying to figure why real esgran model wouldn't load. The gives me a excuse to start fresh.
@@SyntheticVoices It does get updated quite a lot, so if you do a “git pull” every day or so you’ll likely get some updates!
@@NerdyRodent Installed it from scratch and everything is as it should be :) cheers
@@SyntheticVoices awesome!
u change it to "Analog app 1 TE" to get the sa soft as Nice tutorialm. Hope tNice tutorials helped!
We all learn differently, I watched, attempted and fucked up, now I’ve watched again, attempted and hopefully won’t fuck up... as much
Good luck! 😆
Will i watch your channel 24/7 just to look at that beautiful face of yours? *yes*.
please download the data command... ... I don't know well...
Hahahaha sae dudee
I really hope to see someone install it on windows, because as I was already thinking, I'm failing something while setting it. I get an error about "Couldn't launch git" while launching the webui-uer.bat, and I dont find anything about it on the troubleshooting.
EDIT: SOLVED
after 1 hour of trying to understand what was the issue.
the problem was probably caused by specifying a custom folder during git installation.
at least I think, because the solution was to uninstall Git and reinstall it without changing the folder.
Glad you got there in the end!
@@NerdyRodent thank you! Now I have to understand how to use properly the inpainting, but it work!
What the heck, from a day to another it have stopped working and tell me he can't find Python :/
You are the author of this repo GUI?
The GitHub repos I’ve done are by nerdyrodent 😉
Will this work with NVIDIA GTX 1650 4GB VRAM?
Yup, as stated in the video, 4GB is the minimum required 😉
Yes I saw that but some people say their 8 GB VRAM was still not enough for SD. Hopeing this works for me. Gonna try it out.
The Matrix has you.
I’m not even sure which Matrix is real…
please upload the pastebin link :(
Thought I was the only one "General S hettings...for the S het"
lol, it's what im currently doing!!!!
is this going to work on mac m1
There is a version of SD for M1 macs, yes!
@@NerdyRodent I have downloaded the Diffusionbee but it so restricted GUI that I can't do what you do in this video. If you know a better and advanced version of a SD for mac and tell me I will be very appreciated. thanks
@@eymengur61 Unfortunately I’ve not got a Mac to test with, so no sure of others ☹️
@@NerdyRodent no problem thanks for your help :)
just getting black images no matter the options i select. any help?
You probably have a low-end card. Try with --precision full --no-half --medvram --opt-split-attention
@@NerdyRodent luckily got it fixed. Actually by doing the -precision full part. Apparently specific cards require that for some reason (I’m on a GTX 1660). If you’re wanting to use the webui you have to put it in the arguments of the starting batch file. As for the other commands you mentioned I will have to give them a try. Thank you!
@@TooArtistic awesome stuff! Here’s to making many steampunk rodents 😉
inpainting does not work for me, im running automatic1111 v1.5 and it does nothing no matterthe settings
Try slapping the denoising up to 1 and it should fill things in amazingly well in both inpainting & outpainting!
Does it work on an AMD gpu?
There is a rocM version of Pytorch, so give it a go and let me know!
@@NerdyRodent I will also try it out tonight on my RX6800XT, the basic stable diffusion on a docker container worked great so I have no reason to think this won't.
I have tested it and it does work with my RX6800XT on arch linux, I needed to install the RocM modules on my system and the inside the conda environment that I was using install the RocM version of pytorch. Not all AMD GPUs are supported by RocM but there are some workarounds that may make it work somewhat well with an unsupported GPU
@@SergioEduP Awesome!
Do you know how to update the local installation after the main repo is updated?
Once a repo has been cloned, updating it is as easy as typing git pull inside of Miniconda when in the repo’s topmost directory downloaded by the clone command. Below you can see I used the cd command to navigate into that folder.
It's from another repo instructions, but i think it applies here too.
@@ZeroCool22 Thanks for the quick reply. I'd downloaded it as Zip file, so not sure if this will work. But I will try it. Thanks again for the video. It was really helpful.
If you downloaded as a zip file then you'll need to keep downloading as a zip file and unzipping for each and every change. "git clone" and "git pull" are a lot better as they are quicker, easier and will only update the files which changed :)
@@NerdyRodent El GIT PULL se hacia estando en (BASE) o dentro del Env. correspondiente, por ej. (LDM)?
@@ZeroCool22 git has nothing to do with anaconda so you can run it using any environment you like - it just updates the repo based on the directory you're in
Man, the "inpainting" isn't as good as Dalle-2 sadly. Still waiting for that part to get better
Value mismatch error
We need something more user friendly.
I’m with you there. What would you suggest instead of the web interface? Personally I much prefer the standard shell interface - it’s quick, easy and simple to automate!
does it work on gtx 1650?
Let's hope so, I have one as well.
But I'm always getting out of MEM errors, no matter what I do. The only way for me to use SD is by way of openvino, which uses only your CPU.
@@synthoelectro And how long does it take to generate?
Downsize the image size output
@@SyntheticVoices doesn't work, still crashes.
Gonna have to wait for a nice GUI or even a photoshop plugin
lol, it's already a nice GUI, what more you want?
Nick maybe you're just old.
It doesn't look like that fork you suggested is valid any more.
Still there for me! Last update a few hours ago.
@@NerdyRodent Could be because there are over 250 forks.
tNice tutorials tutorial was very satisfying
locally amd!!!! noy function!
is free ? sin gpu?, because colab made me hacked google.
Yup, running it locally is free 😉
@@NerdyRodent very nice!!!
i have gtx nvidia 1660 function????
@@MrLaura34 Plenty :)
again
The outpainting looks so bad with that harsh mask and without any sort of feathering lol
Yeah, it's a "poor man's" outpainting - haven't quite got it to Dalle2's quality yet. Maybe with a bit of playing... but maybe not!
@@NerdyRodent Yeah, the Outpainting "Quality" thing, is the only advantage that Dally have over SD for now...
Who's making fnf covers lmao
So that's what inpaint means. If they just called it mask, it would make more sense.
Yes, it's more like "repaint"
Can I use this to generate pornographic content?
thanks for the tutorial!
does anyone know what this means?
CodeFormer not found at path C:\Users\*****\Documents\stabledifusion\CodeFormer\inference_codeformer.py
You'll need to git clone the codeformer repo into that location