A Variational Autoencoder. The idea stems from an older generative algorithm from the pre GAN era, but of course Diffusion heavily derives from VAEs as well in terms of the mathematical formulations.
The fact you don't have to go off the deep end with a hundred prompts to get good results is what makes this whole process rather impressive. Now, if I only was good with art and illustration to really take generated images to the next level.
@Pop Gligor True enough. I look at AI generated art as some sort of template, proof of concept, a basis to start off with. Being able to generate ideas you have, and go about your own illustration. AI is a tool for helping you along a process, not to replace human input. I can imagine how far I could take these images and improve on them with my own personal touch.
@Pop Gligor Absolutely. We're quick to resort to max efficiency rather than time commitment, anyways. Something can do it for you? EZ - My work here is done.
@Pop Gligor the problem with AI is not the AI itself... but the fact that is using billions of copyrighted pictures to merge into the blender, so is freaking unethical, and more unnethical when you think that those artists that are being merged, are loosing their jobs because of this hiper-consumist entertainment product who needs to use their work without consent to work
@@Downsider24 and the other problem is that, just like you are using it just as a basis refference to start off with, your consumers will see it as the final product, so it doesn't matters what you do with AI... if the technology every month can approach more and more to what the client wants, you are not necessary, despites how good you can be using AI
That's crazy, I would never have thought about that and even was considering to pay Midjourney or something like that, thank you! The cursed image was literal, that's so funny
I've been looking for ways to improve the quality of my images and didn't really give this part of the software much thought, thanks for the informative vid! What do you think about people who just copy & paste a wall of text into the negative prompts textbox? I've been doing that and I'm unsure of its effectiveness.
It is quite interesting whether it is possible to synthesize very high resolution images using only the stable diffusion model itself without extraneous upscalers? That's a question no one dares to answer! But I hope for a logical and correct answer.
Thank you very much bro, I was breaking my head trying to improve my results, and you come and show me a way to not just improve my results but to get results WITHOUT prompts!
I think you should talk about AI Model because viewer might not have the same AI Model, and I think every AI model is react prompt differently, please make video about this again but way more advanced thank you
Hey guys any idea where you could find all of these Stable Diffusion Models? Are there any dedicated website that I can browse through all of these models?
Generate two characters like you normally would. They will likely look like twins. Inpaint one of the characters out and add the features you want for character #2.
no one cares. no one asked you. we can keep using if we want to. why the hell people feel like they are doing a favor by telling other people what "dos" and "donts" ? It's none of your business.
@@brunnomenezes3346 He's right though, it's just as much none of your business either, because you're not actually doing BUSINESS. Digital looting is much safer and quicker than physical, I guess.
Hello, I really need help with running webui-user.bat There's an error, I'm not sure if it's safe to paste the whole CMD here for anyone to see. There's 2 line that says: "ImportError: cannot import name 'urllib3' from 'pip._vendor'" and "RuntimeError: Couldn't install torch." What can I do to fix this?
the way he said "behold" part was the equivalent of "ok mate, this is so simple it shouldn't be a tutorial but here we are", love it, instant like
Bro, that was easily one of the best and specific stable defusion tutorial I watched on UA-cam, keep it up.
You seem to be missing the use of a VAE hence your images colours are all washed out.
sorry can you tell me what is VAE?
and what does it do
@@0SussyBaka0hehe~ ;))
A Variational Autoencoder. The idea stems from an older generative algorithm from the pre GAN era, but of course Diffusion heavily derives from VAEs as well in terms of the mathematical formulations.
Adding an SD VAE "or Post processing script as i like to call it" will give you a huge improvement on your color output xD
The fact you don't have to go off the deep end with a hundred prompts to get good results is what makes this whole process rather impressive. Now, if I only was good with art and illustration to really take generated images to the next level.
@Pop Gligor True enough. I look at AI generated art as some sort of template, proof of concept, a basis to start off with. Being able to generate ideas you have, and go about your own illustration. AI is a tool for helping you along a process, not to replace human input. I can imagine how far I could take these images and improve on them with my own personal touch.
@Pop Gligor Absolutely. We're quick to resort to max efficiency rather than time commitment, anyways. Something can do it for you? EZ - My work here is done.
@Pop Gligor the problem with AI is not the AI itself...
but the fact that is using billions of copyrighted pictures to merge into the blender,
so is freaking unethical,
and more unnethical when you think that those artists that are being merged, are loosing their jobs because of this hiper-consumist entertainment product who needs to use their work without consent to work
@@Downsider24 and the other problem is that, just like you are using it just as a basis refference to start off with,
your consumers will see it as the final product, so it doesn't matters what you do with AI...
if the technology every month can approach more and more to what the client wants,
you are not necessary, despites how good you can be using AI
Speaking as an artist myself: the fact that I do art makes the whole AI process feel boring, bland, uninspired, unnecessary
i've seen ppl used thos numbers within brackets but didn't know what that means, this helps alot, thanks!
looking forward to another tips
So the advice is to, instead of asking the AI for something, you basically tell it "Do whatever, just make it not dogshit."
Yes, it was trained extensively on dogshit and can locate it on a map. Useful for rolling in it or avoiding the pile entirely.
😂
Another tips: write hands in negative because you will get better hands. Idk why it works.
Getting good tutorials for Stable Diffusion is like generating images without prompts at all
That's crazy, I would never have thought about that and even was considering to pay Midjourney or something like that, thank you!
The cursed image was literal, that's so funny
Yes please, if you're not going to pay the artists, at least don't pay for software that took art and images without consent.
What website did you use I am confused
@@lambasted6886 It doesn't steal art, it only trains by looking at them, like normal people.
@@luv_sic artists gonna cry
1:52 bro created a horror anime character lol
Thank you for sharing this, its made such an improvement its ridiculous!
for a second I thought he will input the holy negative prompt "blurry" XD
Jesus, wish I knew about this earlier. Everyone stable diffusion guide needs this.
Thumbs up if you can actually draw like this without tracing someone else's work or using AI.
Amazing, straight to the point and extremely helpful.
Ngl I like the first 2 more because the lifeleness from those images adds charm to it
I've been looking for ways to improve the quality of my images and didn't really give this part of the software much thought, thanks for the informative vid!
What do you think about people who just copy & paste a wall of text into the negative prompts textbox? I've been doing that and I'm unsure of its effectiveness.
At this stage it's probably best to just try some controlled experiments where you use the same seed and compare a wall of text vs something smaller
@@animeaiart I thought so too, thanks. There seems to be a lot of superstition and placebo around what the right props are.
You can just set the default negative prompt in the UI config file.
You can also add some negative embeddings that can add more to it too
I liked your stuff and subscribed.
link the website please
Wow, thank you! It's helpful!
Is there a tutorial where I can understand what the inside brackets and those numbers before the : symbol?
Thank you very much for the tutorial!
Clip Skip 2 might also help in some anime models
thanks
What is the link to the website?
Wow, that was a great difference. 👍
You could also just select these words and press ctrl + arrow up or down instead of doing it manually.
can we tell the clothes whare to be (ie maybe showing a peek)
People really call himseft artist just for doing this, bruh
Use a VAE for more color like kl-8f-anime2
It is quite interesting whether it is possible to synthesize very high resolution images using only the stable diffusion model itself without extraneous upscalers?
That's a question no one dares to answer!
But I hope for a logical and correct answer.
The colours look washed out, did you use a VAE ?
Nope, this is to keep the comparison controlled, so you can see how much comes from the negative prompt instead of the VAE
Thanks, And good luck 😁
I have a question (hope somebody responds soon) but what if I already have a sketch with detailed lineart and only need to AI to paint it for me?
it was grey and low quality and stayed that way
bro, im very thanks from ur helpfully tutorial , that Awesome XD
Where can I get stable diffusion or how can I download it? I would like to try those tips, thank you
i have liked and subscribed to this incredible information
Thank you very much bro, I was breaking my head trying to improve my results, and you come and show me a way to not just improve my results but to get results WITHOUT prompts!
looks like this could use a VAE as well
dude you are awesome
easy negative also helps
the real hack to improve your AI art... is just wait for the next update of the developers ...
literally
just those? I remember putting in "bad anatomy" in negative prompt as well. Masterpiece can be weighted, too. (masterpiece:1.0)
I love the fact of those cursed images XDD
I think you should talk about AI Model because viewer might not have the same AI Model, and I think every AI model is react prompt differently, please make video about this again but way more advanced thank you
"lower than highest quality possible" can be rhetorically used though lol.
Genius!
You're a god mate, the goth girls I generated ON GOD 💀
thought I had to install some sort of program but no... i literally just had to type 2 word with quality
too big brain. im melting
Hi, I’m trying to make an anime ai out of my sketchbook. Any ideas?
Thanks G
what is this app name
Interesting it made its better thanks i Have been trying with my Ai art it’s been Terrible Especially with the anime thanks for the AI tutorial
I get why it's called negative little bit is good after 3 it's nightmare time
What is the app
1:49 i actually like this one because it looks creepy as hell
I use getimg and those negative prompt didnt work well can you help me with that?
That's not improving the art, that's changing the art style 🤨 and I personally think the first one looks way better than generic anime shading
How can I fix my upload quality on twitter can you do a vid on it it would be really really helpful
what gpu do you use in this vid ?
Can we know the configuration of your PC? The image got generated so fast!!!
What's the ai you used? Please
how do i add this is google collaboratory?
Actually cursed image holy sht
what is this website you using?
what software is this?
What website is this, please?
how can I get that software?
Hey guys any idea where you could find all of these Stable Diffusion Models? Are there any dedicated website that I can browse through all of these models?
Yep huggingface is the only one
Civitai
What's the website/app?
1:50 💀
Oo what js the website?
Which model did u used
Bro you just have to read on the top left corner and Google it
@@Warzak77 where sry I'm new to ai or stable diffusion 💀
AbyssOrangeMix2Sfw
I Tried this and got TONS OF NSFW CONTENT. Damn bro this is cursed... 💀💀💀💀💀💀💀
How'd you do it?
Tired of watching 20+ mins video, thanks for keeping it short and informative.
just add vae
wtf is this real lmao
anyway, good laugh. as a stay at home shut-in neet, I haven't laughed in years. no joking.
"Improve" "your" AI """art"""" 💀
Literally
The link to page? or you downloaded it?
Link to which page?
@@animeaiart stable diffusion maybe? looks diferent in web
@Big CREATOR the model AbyssOrangeMix does not exist
1:51 CREEPY
any tips on generating two characters?
Generate two characters like you normally would. They will likely look like twins.
Inpaint one of the characters out and add the features you want for character #2.
Agree with Bombalurina. you Can also use the New controlnet open pose. Use open pose and 2 people in img2img to place them acurately
is there an online version for controlnet? i only have 4gb vram
@@speedeespeedboi9527 not that i'm aware of sorry.
Automatic1111 Can use collab but i don't know if it is possible to use controlnet with it
WHAT its that f****** simple?
2 months learning ai art and this is all i needed to do
Looks like you need to up your game too. No high rez fix, no VAE file. No Additional-Networks or LoRAs still using DDIM. Come on brah.
Can you elaborate
I'm making the comparison fairer by keeping the other settings very plain. Maybe it's better to use other stuff, but I can't decide
HuggingFace URL?
ai?
This is old news
noob strats. i prefer putting worst quality in my positive prompts instead
I prefer putting all my positive prompts in the negative prompts
L
Look, another portrait.
or u could pay a real artist
Why pay when you can have it free
@@kangsate3459 To help support someone and their hard work?
@@kangsate3459 I mean, without those artist AI art wouldn't even exist considering it just mashes up thousands of real art
@@kangsate3459 Its like singing a song and giving zero pay and credit to the actual song writer
It’s free, unless it’s an actual masterpiece or i requested something VERY specific, i won’t be spending anything
"Your"
Or, hear me out- stop using ai cuz the art that was sourced to make the models was taken from non consenting artists (:
no one cares. no one asked you. we can keep using if we want to. why the hell people feel like they are doing a favor by telling other people what "dos" and "donts" ? It's none of your business.
@@brunnomenezes3346 He's right though, it's just as much none of your business either, because you're not actually doing BUSINESS. Digital looting is much safer and quicker than physical, I guess.
🤓☝
This is infinitely less important of an issue as you think it is
Or,hear me out- shut up
AI ART
/spit
Hello, I really need help with running webui-user.bat
There's an error, I'm not sure if it's safe to paste the whole CMD here for anyone to see.
There's 2 line that says:
"ImportError: cannot import name 'urllib3' from 'pip._vendor'"
and
"RuntimeError: Couldn't install torch."
What can I do to fix this?
you mean tips to improve stealing arts
Stealing art? What are is being stolen herec