Should absolutely mix upscalers to get the best result. It takes a bit of time to dial in the mix recipe right, and you'll have to dial one in per every different model you use, but when done correctly it does yield a better result. A basic mix example that can be used if you're going for realism would be to to use Lanczos as your main upscaler, and then mix in Esrgan 4+ as your second at 0.5 visibility and tweak it up or down based on the initial result. If you're going for more a more stylized or cartoonish result, substitute Anime 6B for the base Esrgan 4+ in the mix. It's not just for anime, it is a better upscaler for any fantasy, concept art or animation style art in my opinion. A tip when choosing what to set CodeFormers at - set the visibility to 1 and the strength to whatever you have it set to in your main settings under Face Restoration. Tweak the strength up or down based on the initial upscaled result that you get. A different option to using the upscalers in Extras is to add the Ultimate SD Upscaler plugin to your scripts in Img to Img. It's more configurable and breaks your image down into grid sections and upscales each "square / rectangle" separately instead of doing 1 pass over the entire image at once. Yields very nice and often more detailed results. There's tutorials around the internet on how to use it's various options.
I started with AI Image Generation yesterday. Now I am trying to upscale a bunch of game textures... Is there some method i can try using Stability Matrix and a Propomt so I can tell the AI which material the texture is (and get better results by doing this)?
Im struggle with high res image, i see Ai images right now with pretty high res like you can see the detail of the skin. I mean how? i tried upscaling my generated images in diff settings but the quality is getting what i want, something like an ultra quality like a wallpaper. SD just giving large file but the overall quality is the almost the same. I have Rtx 3060 12gb running i windows 11. Sd 1.5 512 res
Hi can we upscale the image while processing like once the image processing is done then it should automatically upscale is there any feature like that ?
Great question, BA. I have been able to reliably use AI up scaling to increase low resolution desktop backgrounds to 4/5k scales with almost lossless quality.
Stable Diffusion doesn't come with a selection of upscalers by default -- or at least Automatic 1111 doesn't. Perhaps you can do a video on where to find them and how to install them. Thanks.
Very respectfully I am Asking !!!! I don't know How to get That software and How to Install it. Plz Help Me I am very Eager to learn and Generate AI art. Humble request 🙏🏻
all of my upscalers resulted in errors they get various errors, stops, and doesnt give me any image result these kind of errors: TypeError: argument of type 'NoneType' is not iterable RuntimeError: unexpected EOF, expected 3405 more bytes. The file might be corrupted. what i have tried: 1. Reinstall the whole nai diffusion. NOPE. same error 2. tried to google the errors above and checked like >30 reddit/github/forums but they are giving me some fix that i cant even find like add pretrained and force_download on model/configurations/tokenizers but this folder doesnt even exist in my repo 3. redownload specific models like RESRGAN. same error of stuck at 90-99% download and no result comes out
Should absolutely mix upscalers to get the best result. It takes a bit of time to dial in the mix recipe right, and you'll have to dial one in per every different model you use, but when done correctly it does yield a better result. A basic mix example that can be used if you're going for realism would be to to use Lanczos as your main upscaler, and then mix in Esrgan 4+ as your second at 0.5 visibility and tweak it up or down based on the initial result.
If you're going for more a more stylized or cartoonish result, substitute Anime 6B for the base Esrgan 4+ in the mix. It's not just for anime, it is a better upscaler for any fantasy, concept art or animation style art in my opinion.
A tip when choosing what to set CodeFormers at - set the visibility to 1 and the strength to whatever you have it set to in your main settings under Face Restoration. Tweak the strength up or down based on the initial upscaled result that you get.
A different option to using the upscalers in Extras is to add the Ultimate SD Upscaler plugin to your scripts in Img to Img. It's more configurable and breaks your image down into grid sections and upscales each "square / rectangle" separately instead of doing 1 pass over the entire image at once. Yields very nice and often more detailed results. There's tutorials around the internet on how to use it's various options.
you have no idea how much u have helped me bro , thank you !
@@darknessawaits8018 Happy to hear that. 👍
Thanks for the video. I was really struggling to find a good free upscaler for my AI images and i didn't even know that i have one in web ui)
Glad I could help! Web UI is super powerful.
if you run it locally there is a progress bar in the UI, the results image pane turns into a large loading bar, or maybe its the most recent version
very informative even today thanks
"And now I can get it printed, and that will make me much happier in my life." 😂
The dog is so cool and cute!
I started with AI Image Generation yesterday. Now I am trying to upscale a bunch of game textures... Is there some method i can try using Stability Matrix and a Propomt so I can tell the AI which material the texture is (and get better results by doing this)?
Mine always get error if try to upscale into 4k resolution, like the image wont show up
me too
Im struggle with high res image, i see Ai images right now with pretty high res like you can see the detail of the skin. I mean how? i tried upscaling my generated images in diff settings but the quality is getting what i want, something like an ultra quality like a wallpaper. SD just giving large file but the overall quality is the almost the same. I have Rtx 3060 12gb running i windows 11. Sd 1.5 512 res
try a different model
Thanks!! Nice roundup!!
I appreciate the comment Swann!
How do you make a bad quality image into a good one though? 🤔
Would you let us know where to find the SAVE button to save the upscaled image?
it gets saved on your computer automatically in automatic 1111 folder under outputs
Thanks for watching! Please share your AI creations and questions on my Discord server! discord.gg/2fcpBcpqMY
Why set the upscaler visibility to 1, while not using second upscaler?
Hi can we upscale the image while processing like once the image processing is done then it should automatically upscale is there any feature like that ?
I don't have a "Restore faces" option on my end. Then again this video is a year old, maybe its auto-included now?
So whats the realistic limit of this? Its all cool but 2K doesnt really cut it. Can you get to 8K realistically and get any usable results?
Great question, BA. I have been able to reliably use AI up scaling to increase low resolution desktop backgrounds to 4/5k scales with almost lossless quality.
I want to upscale a specific face from a video, is that possible with a tool ?
how do you get to this page?
Great Video. Thanks and Keep it up. :)
Stable Diffusion doesn't come with a selection of upscalers by default -- or at least Automatic 1111 doesn't. Perhaps you can do a video on where to find them and how to install them. Thanks.
It might depend on whose version of Auto 1111 you installed. I've been using Auto 1111 for months, and it's always had the upscalers included.
uuuh, yes it does.
Thanks yo this is awesome.
how do i change 1792 x 2304 to 2560 x 1440 ?
There is a progress bar in the command line of the server for this process.
bro, what theme you using?
is there anywhere a proper explanation about the SD Upscaler in img2img tab in Scripts? how do I properly use that one?
i just dont have the hires. fix option...
But how to add R-ESRGAN 4x+ for upscaler menu?
How To Do It In Google Colab Or In Python Code
Very respectfully I am Asking !!!!
I don't know How to get That software and How to Install it. Plz Help Me
I am very Eager to learn and Generate AI art.
Humble request 🙏🏻
thanks!
what are the aspects of your computer?
King Stix is that you?
all of my upscalers resulted in errors
they get various errors, stops, and doesnt give me any image result
these kind of errors:
TypeError: argument of type 'NoneType' is not iterable
RuntimeError: unexpected EOF, expected 3405 more bytes. The file might be corrupted.
what i have tried:
1. Reinstall the whole nai diffusion. NOPE. same error
2. tried to google the errors above and checked like >30 reddit/github/forums but they are giving me some fix that i cant even find like add pretrained and force_download on model/configurations/tokenizers but this folder doesnt even exist in my repo
3. redownload specific models like RESRGAN. same error of stuck at 90-99% download and no result comes out
goodboy
Does this work on stable diffusion online?
Absolutly!
Cool
My results with topaz have been so much better than sd reults? How can I better use sd to push it to it's limits?
Clickbait
Hi @binks_live, thank you for your video. Could you point me where I can find the colab of your video? TIA