Hello Nerdy Rodent! Did you see about the new Automatic1111 sketch inpaint method? I really can't use it, could you make a tutorial about it? It seems like an extremely interesting tool
Hey , I love your videos but ive been away from stable diffusion for quite a while and I have seen that a lot has changed. Could you make a tutorial on how to download and use this new version? like a tutorial/guide
Hello there Nerdy Rodent!! Thank you for your wonderful tutorial but the most thing that I aware is the avatar of you at the bottom right!!! How and what software did you use to generate that? And how do you sync the lip too? Please please please dm me for more information🙏🏻
this feature isn't part of automatic1111. but automatic1111 already updates itself every time you run it. when you click the webui-user.bat file, you should see lots of plus signs after some text.. thats it updating it self. if it doesn't do that you can make it do it. find webui-user.bat in the main directory, edit it with a text editor. look at the last line that says webui.bat, just above that line it should say git pull. if it doesn't, add the line git pull above it, save. then run it like normal, it should update automatically. if that fails, the easiest way to fix things is just to delete everything and reinstall it from scratch. (but if you do, maybe backup all your ckpt model files into another folder so you dont have to download them again, you can just put them back in the models folder after you reinstall)
I'm looking for a free way that can take one image like a new 2023 Gucci jacket and inpaint itself into an existing image. I can only think of one tedious way which is creating a lora based on all the existing photos of it but that is a huge hassle.
Thanks for this video!! and for the links too.. I have quick question, do you need to have Stable Defusion installed and working, or does it install a version of Stable Defusion, as I have the Automatic11111 installed, I am not sure if this will install again or do they work together?
i missed some stepts becaus i realy dont know anything about conda or the codes. so dont how to copypast them at the right spot. can you help me out. love the content
Stupid question: how to restart it once the cmd window is closed? - Thank you for the video - it worked.. (at GPU with 16gig Vram - will run only if CFG is 1 and make only one sample)
Love community. Yesterday I was looking at similar thing but I was forced to wait 6h before my picture would be processed. Locally? Will see how it work :) Thanks for share. Wait? seriously we need 48GB of VRAM ? or 48GB of RAM?
3090, 4090. 24GB of VRAM. Works perfectly fine, I generate thousands of images per day. Probably more (about 4 sec. for 1 image). 48GB of VRAM... nah. Sure it'd be better (not even sure) But 4500$ just to create images. That's a bit insane.
Hey Nredy. For some reason getting " ImportError: libavdevice.so.59: cannot open shared object file: No such file or directory". Even though I have double checked ffmpeg is installed on arch linux (I'm on Endeavour OS). Also I got automatic 1111 stable diffusion which is working fine without this error
I run automatic1111 on win10 with Python 3.10. I don't know much about python or conda. I take it simply opening python is not the same as a "conda terminal" because pasting there just brings up syntax errors.
Indeed. The anaconda terminal requires installation of the anaconda software, as linked in the description. Anaconda is basically like Steam for Windows gamers - it lets you install various “games” (Python packages) and keeps them contained for you, allowing multiple versions of the same thing to be installed at the same time.
Everyone READ please for the community!! This is for Nerdy Rodent but anyone else please read this and share in the discord and Reddit community!! Thanks! I hope someone soon makes this an extension for Automatic’s webui and it works for 12GB users. I have a request of you Nerdy Rodent, can you make spread the word about making this for Automatic’s repo because this will erase copyright issues since we can combine images so styling will vary enough to where no artist has grounds to say it’s “stealing”. We know it isn’t stealing but this will solidify this argument completely if this mixer is really pushed to the community. I’ll make a request too but I know you probably have way bigger reach
You're might've misunderstood what is the issue art community has with diffusion models. They're upset that their images were used to train the model, not that the images are too resembling of some particular style, so no amount of mix/matching will "erase copyright issues", since the model still "contains" copyrighted material.
@@xn4pl No I’m aware of the issue, but if we mix styles with this new feature my point is that if they can’t tell where it came from, therefore they can’t make the claim we used their images or “their style”. For example, if you have a rendering style that’s very popular of someone and mix it with 5 other styles, you can’t tell where it came from. The composition difference has already been solidified but if we modify rendering styles enough with this new feature, it’s completely unknown so we have to push this new feature into Automatic’s repo. This issue of copyright would/will be gone considering the fact that they couldn’t even recognize it right in front of them. Then claims can’t happen.
@@xn4pl if that's what they truly believe in the art community, aren't the famous artists supposed to be upset about new human artists using their work for training too by extension? They will inevitably replace them, c'est la vie! And even if they don't copy or reinterpret their art, even if they aren't even inspired by it, and even if there are no involuntary citations, they still improve their visual experience, aesthetic taste, and good eye for art in general because it TRAINS them. I had to watch A LOT of good photography to be a decent photographer, a lot of good editing to become an okay editor, and l am sure it's the same for painters as well. Much like a neural network relies on data about art in general to prioritise good output and avoid a lot of bad patterns. The only way to stop others from training with your art is to never show it to anyone but yourself. Because you share the data no matter what, the difference is a human turns it into knowledge and machine turns into weights. If that is the issue, but it's still about art, and not publicity, earnings or fame, not sharing art wouldn't be a problem at all, go try that. And also... I wonder what Greg Rutkowski would say about mentioning him in the negative prompt. That would be an interesting one.
@@HAJJ101 what you're saying is utter nonsence. Stable diffusion already mixes styles as you wish it in prompts, and adds a bit of its own "touch". This mixer does literally exactly the same, just from images. No one can claim that their style is stolen anyway, because diffusion models can't replicate styles exactly, and styles are not protected by a copyright anyway
There are no copyright issues and the are no grounds for "stealing" with the base models. Stable diffusion already does precisely what you want it to, so you should edit your comment as to not contribute to spreading misinformation. Have a good one!
@@NerdyRodent yes it appears I only have 8 gigs of that particular type of ram. Spent hours trying to understand and getting around the issue but my tech wizardry level is much too pathetic. It is as if I am flailing my wet noodly appendages at a steel door hoping the hinges are rusted through : P I should probably instead figure out how to delete all the many gigs of stuff anaconda installed here and there
@@NerdyRodent sorry don't take me wrong I love all your videos. I just meant that for many of us it's impossible to upgrade a video card, because then we need also to upgrade a power supply, and maybe even the motherboard..... So it's not easy for us old schooler artists that are learning this new AI, do have to suffer the consequences off not having a large ram of video card, hell I cannot even do any training at all locally because I need at least 8vram for dreambooth.
@@liquidmind Hence the renting option being quite good. A hour for less than a pint of beer sounds like pretty good value entertainment 😉 There’s colab free as well, and this should run in there if you just do 1 image at a time! Colab pro for not much a month too, which is probably the best value for GPU time. But yes, the very first thing I saved up for when I got into AI was a capable GPU because I wanted to play with all the toys! Helped that I’m a gamer too…
plz plz plz mr we need really a full detail beginner tutorial that how you make that face talking with ur custom face expression right below on corner with very long duration pz???
@@NerdyRodent no no no mi squirrel hybrid type human i know that link but its useless bcz its incomplete there you don't teach us that how you make 5 minutes long custom face feature animation video with Thin plate ??? there ithink 1 mintue restriction??????
@@AaliDGr8 You’ll need to run it locally as shown if the free demo is limited to just 1 minute. Even then you could just do it 5 times for a 5 minute driving video 😉
@@NerdyRodent tnku! mi deer human squirrel u r great but but but here is the twist T__T i am fully newbie totally 0 what is locally run this thing how which which thing need to dwnld etc............ if anyone have detail guide on it not just a 10 minute tutorial :((((( plz help
@@AaliDGr8 The word “local” is defining a region. In this case, rather than running it on a remote computer you could instead run it on your own computer. So another way to say “running locally on your own pc” would be “running on the computer you own in front of you”. It is the opposite of running a program remotely, such as via google colab or a huggingface space website in the internet. I hope that clears up the possible options for running the program! 😀
Awesome, thanks for sharing! Stablediffusion is so good, I barely don’t use negative prompts anymore
@@BlankFX , of course! But, we have so many great checkpoints with Stablediffusion. Have fun!
For anyone curious about local, the app takes about 20 GB of VRAM to run (not sure if he mentioned the local vram usage).
Yup, I do indeed explain that… twice, in fact! 😉
we will Just wait for a automatic1111 implementation 😆
What if you only have 16 GB 🥺
@@louie8362 basically just wait for a webui extension haha
why so much?
why cant this just be an extension to automatic 1111 :(
Unfortunately the demo on huggingface isnt working :(
Is there a Google Colab of this?
the filter on the cam is on point for this topic!
that 24GB limit kills me every time lol :D ... but thanks for sharing
can you use your own model with this like 1.5 or have to use their mixer model?
The people want to know.
no, but you could use images made with your custom models as inits
So many projects going on. I wish someone looked to optimize this for lower VRam local requirements
So many things to try!
thanks for explainig how to install !
Amazing tool, thank you for bringing this to my attention.
Thanks. How do you create your talking avatar?
How do I do that animated face thing? - th-cam.com/video/Z7TLukqckR0/w-d-xo.html 😉
@@NerdyRodent great, thank you. Do you know if there is also a way in the meantime that runs in automatic1111 directly without all the anaconda stuff?
Hello Nerdy Rodent! Did you see about the new Automatic1111 sketch inpaint method? I really can't use it, could you make a tutorial about it? It seems like an extremely interesting tool
Sketching things in is fun 😉
Hey , I love your videos but ive been away from stable diffusion for quite a while and I have seen that a lot has changed. Could you make a tutorial on how to download and use this new version? like a tutorial/guide
Do you mean tutorial of installing stable diffusion or of a webui like Automatic1111
Hello there Nerdy Rodent!! Thank you for your wonderful tutorial but the most thing that I aware is the avatar of you at the bottom right!!! How and what software did you use to generate that? And how do you sync the lip too? Please please please dm me for more information🙏🏻
I use tpsmmfia - links in the description!
great job!
keep up the great work!
Thanks!
I always wanted something like this (which is similar to Artbreeder) but the requirements killed my excitement instantly (8 GB card XD)
Can be this used with poses? the character that I want and the poses in other image?
It can be use like posing with txt2img or img2img, but not like poses with inpainting
HiNerdy Rodent. I really want to try ,but why I cannt open the Image Mixer's link?
Awesome!😁
What's the simplest idiot proof way to upgrade our local install if we already have the a1111
this feature isn't part of automatic1111.
but automatic1111 already updates itself every time you run it.
when you click the webui-user.bat file, you should see lots of plus signs after some text.. thats it updating it self. if it doesn't do that you can make it do it.
find webui-user.bat in the main directory, edit it with a text editor. look at the last line that says webui.bat, just above that line it should say git pull. if it doesn't, add the line
git pull
above it, save. then run it like normal, it should update automatically.
if that fails, the easiest way to fix things is just to delete everything and reinstall it from scratch. (but if you do, maybe backup all your ckpt model files into another folder so you dont have to download them again, you can just put them back in the models folder after you reinstall)
I am loving your content so much. So jealous I don’t have a GPU yet but I am definitely now convinced to get one.
Excellent tool, thanks!
Yeeeeeeees.... 'interesting cursed horrors' forever! 🤓🐁🐭...🖱🪤🤟
You should generate a Nerdy Rodent human like avatar through Midjourney or Stable Diffusion and use that as your avatar.
how you made the animation of your avatar, is that a green screen?
Links in the description!
@@NerdyRodent thankyou! Gonna binge watching all of your video, love your work
Hello my nerdy friend! 😊
Hello there! 👋
it can be run locally on 11 GB vram ? 1080
I'd say probably unlikely
Thank you , was looking for a video about this! Can I launch the application through the windows cmd? Like with stable diffusion?
Can't wait until someone ports it into Automatic1111. :)
Hey boss, is your Thin Plate Spline Motion Model avatar done in real time V-tuber style or pre-recorded then converted like in your previous video?
Exactly as shown in the previous videos 😉
"Cursed Horrors" for $200, Alex.
How do you make the images look less deformed?
Great Vid, thx!
Glad you liked it!
I'm looking for a free way that can take one image like a new 2023 Gucci jacket and inpaint itself into an existing image. I can only think of one tedious way which is creating a lora based on all the existing photos of it but that is a huge hassle.
the local install is not working for some reason, it was ok before. Does anyone has a similar situation?
Awesome!!
How do you make that webcam substitution?
I use the thin plate spline motion model for image animation linked in the description 😉
Thanks for this video!! and for the links too.. I have quick question, do you need to have Stable Defusion installed and working, or does it install a version of Stable Defusion, as I have the Automatic11111 installed, I am not sure if this will install again or do they work together?
This is a totally different app
i missed some stepts becaus i realy dont know anything about conda or the codes. so dont how to copypast them at the right spot. can you help me out. love the content
No codes here, luckily!
@@NerdyRodent You godly rodent. Can you lend me your hand and figure out how to run on anything other than port 7860.
Stupid question: how to restart it once the cmd window is closed? - Thank you for the video - it worked.. (at GPU with 16gig Vram - will run only if CFG is 1 and make only one sample)
Just re-activate your environment and run the app again!
Is it really slow&janky for you guys too? I have a high-end setup.
Love community. Yesterday I was looking at similar thing but I was forced to wait 6h before my picture would be processed. Locally? Will see how it work :) Thanks for share.
Wait? seriously we need 48GB of VRAM ? or 48GB of RAM?
idk what that 48GB relates to but I have it running OK locally on a 3090 w/ 16GB RAM. Takes 10 seconds to make 2 samples
@@billMcLatentspace I have 8GB and got the error "RuntimeError: CUDA out of memory."😐
You can run this locally on a 3090/4090. It takes about 20 GB of VRAM to use (at least from what I tested).
3090, 4090. 24GB of VRAM. Works perfectly fine, I generate thousands of images per day. Probably more (about 4 sec. for 1 image).
48GB of VRAM... nah. Sure it'd be better (not even sure) But 4500$ just to create images. That's a bit insane.
As mentioned, you’ll need a 24 GB VRAM card to run exactly as shown. Less if you only generate 1 sample at a time.
Hey Nredy. For some reason getting " ImportError: libavdevice.so.59: cannot open shared object file: No such file or directory". Even though I have double checked ffmpeg is installed on arch linux (I'm on Endeavour OS). Also I got automatic 1111 stable diffusion which is working fine without this error
If you’re sure it’s ffmpeg then you could conda install ffmpeg instead
awsome
I run automatic1111 on win10 with Python 3.10. I don't know much about python or conda. I take it simply opening python is not the same as a "conda terminal" because pasting there just brings up syntax errors.
Indeed. The anaconda terminal requires installation of the anaconda software, as linked in the description. Anaconda is basically like Steam for Windows gamers - it lets you install various “games” (Python packages) and keeps them contained for you, allowing multiple versions of the same thing to be installed at the same time.
@@NerdyRodent Thank you for taking the time to explain.
"A person with a beard"
That's exactly me but I don't look like that
Cool, hope i am capable of making friends with an 'anaconda terminal' as I have a 3090 dying to try this!
Go for it! Python yourself up 😀
Everyone READ please for the community!!
This is for Nerdy Rodent but anyone else please read this and share in the discord and Reddit community!!
Thanks! I hope someone soon makes this an extension for Automatic’s webui and it works for 12GB users. I have a request of you Nerdy Rodent, can you make spread the word about making this for Automatic’s repo because this will erase copyright issues since we can combine images so styling will vary enough to where no artist has grounds to say it’s “stealing”. We know it isn’t stealing but this will solidify this argument completely if this mixer is really pushed to the community. I’ll make a request too but I know you probably have way bigger reach
You're might've misunderstood what is the issue art community has with diffusion models. They're upset that their images were used to train the model, not that the images are too resembling of some particular style, so no amount of mix/matching will "erase copyright issues", since the model still "contains" copyrighted material.
@@xn4pl No I’m aware of the issue, but if we mix styles with this new feature my point is that if they can’t tell where it came from, therefore they can’t make the claim we used their images or “their style”. For example, if you have a rendering style that’s very popular of someone and mix it with 5 other styles, you can’t tell where it came from. The composition difference has already been solidified but if we modify rendering styles enough with this new feature, it’s completely unknown so we have to push this new feature into Automatic’s repo. This issue of copyright would/will be gone considering the fact that they couldn’t even recognize it right in front of them. Then claims can’t happen.
@@xn4pl if that's what they truly believe in the art community, aren't the famous artists supposed to be upset about new human artists using their work for training too by extension? They will inevitably replace them, c'est la vie! And even if they don't copy or reinterpret their art, even if they aren't even inspired by it, and even if there are no involuntary citations, they still improve their visual experience, aesthetic taste, and good eye for art in general because it TRAINS them. I had to watch A LOT of good photography to be a decent photographer, a lot of good editing to become an okay editor, and l am sure it's the same for painters as well. Much like a neural network relies on data about art in general to prioritise good output and avoid a lot of bad patterns.
The only way to stop others from training with your art is to never show it to anyone but yourself. Because you share the data no matter what, the difference is a human turns it into knowledge and machine turns into weights. If that is the issue, but it's still about art, and not publicity, earnings or fame, not sharing art wouldn't be a problem at all, go try that.
And also... I wonder what Greg Rutkowski would say about mentioning him in the negative prompt. That would be an interesting one.
@@HAJJ101 what you're saying is utter nonsence. Stable diffusion already mixes styles as you wish it in prompts, and adds a bit of its own "touch". This mixer does literally exactly the same, just from images.
No one can claim that their style is stolen anyway, because diffusion models can't replicate styles exactly, and styles are not protected by a copyright anyway
There are no copyright issues and the are no grounds for "stealing" with the base models. Stable diffusion already does precisely what you want it to, so you should edit your comment as to not contribute to spreading misinformation.
Have a good one!
now just wait for automatic plugin to use it without horrors of conda instalation ;D
That's not the horror here, it's the amount of VRAM needed!😉
@@ZeroCool22 i give it 3 weeks tops and they reduce it with smart coding ;)
You could do it on Python instead, but honestly Conda isn't that bad!
Conda and venv are awesome! Anyone who uses Python loves virtual environments 😍
with 10GB of VRAM cuda out of memory!!
Yes someone who shows how SD is used, the others don't do that
😂I just realize I can't run it on 8GB VRAM.
how am I supposedly running out of memory when I have 32gigs
Could be you are running out of video memory
@@NerdyRodent yes it appears I only have 8 gigs of that particular type of ram. Spent hours trying to understand and getting around the issue but my tech wizardry level is much too pathetic. It is as if I am flailing my wet noodly appendages at a steel door hoping the hinges are rusted through : P
I should probably instead figure out how to delete all the many gigs of stuff anaconda installed here and there
But does it do anime?
Sure does!
I am new at this. Anyone can guide me on a step by step to install it on stable diffusion?
sir why are you a deepfake
Isn’t everyone? 😏
Whatever you're using for your avatar is just ever so slightly creepy
😉
24 GB VRAM? LMAO.... ok
Some people have AI as a hobby, and others… don’t! 😉
@@NerdyRodent yes i do have AI as a hobby, i meant that in order to RUN this script locally, i need 24 VRAM and i only have 6vram, LOL.
@@liquidmind Luckily they have the free space or you can rent hardware too! So much choice 😀
@@NerdyRodent sorry don't take me wrong I love all your videos. I just meant that for many of us it's impossible to upgrade a video card, because then we need also to upgrade a power supply, and maybe even the motherboard..... So it's not easy for us old schooler artists that are learning this new AI, do have to suffer the consequences off not having a large ram of video card, hell I cannot even do any training at all locally because I need at least 8vram for dreambooth.
@@liquidmind Hence the renting option being quite good. A hour for less than a pint of beer sounds like pretty good value entertainment 😉 There’s colab free as well, and this should run in there if you just do 1 image at a time! Colab pro for not much a month too, which is probably the best value for GPU time. But yes, the very first thing I saved up for when I got into AI was a capable GPU because I wanted to play with all the toys! Helped that I’m a gamer too…
Pretty terrible results compared to blend in MidJourney
plz plz plz mr we need really a full detail beginner tutorial that how you make that face talking with ur custom face expression right below on corner with very long duration pz???
Links are in the description! 😉
@@NerdyRodent no no no mi squirrel hybrid type human i know that link but its useless bcz its incomplete there you don't teach us that how you make 5 minutes long custom face feature animation video with Thin plate ??? there ithink 1 mintue restriction??????
@@AaliDGr8 You’ll need to run it locally as shown if the free demo is limited to just 1 minute. Even then you could just do it 5 times for a 5 minute driving video 😉
@@NerdyRodent tnku! mi deer human squirrel u r great but but but here is the twist T__T i am fully newbie totally 0 what is locally run this thing how which which thing need to dwnld etc............ if anyone have detail guide on it not just a 10 minute tutorial :((((( plz help
@@AaliDGr8 The word “local” is defining a region. In this case, rather than running it on a remote computer you could instead run it on your own computer. So another way to say “running locally on your own pc” would be “running on the computer you own in front of you”. It is the opposite of running a program remotely, such as via google colab or a huggingface space website in the internet. I hope that clears up the possible options for running the program! 😀
My stable diffusion looks like shit and only outputs shit. I must not have installed it correctly.