but its trying to pull resources i have locally 😦
but its trying to pull resources i have locally 
JDo you see where it says 501 server error? That’s not usually a local issue. If someone has another suggestion I could be mistaken. To me looks like at some point your script is trying to access something from huggingface and it’s not accessible at this time. Try again when huggingface.co is back online and it may work. Or you could copy all of the error into Chatgpt and ask what’s the error stem from.
J-but anything that’s huggingface related is going to be unworkable while the site is down
Fmy understanding is that its trying to pull VAE info from hugginface because it doesnt see it locally but i have those files i just don't know where this shit is supposed to go for it to be seen
JThen fix the code to point it to your local files or wait until huggingface is back online. Should do the trick.
FHugging Face (where we store models) is in maintenance mode
Fkik
Flol
Aid like to note something regarding supir. If you want to upscale an image that already has decent quality, do not use Stage 1! Stage one is there to smooth out extremely low res images so the upscaling process can re-add details. But if you do this on decent quality images, you rather lose details instead of gaining the best quality output. Ive tested this with a few images and the output was always better to not use stage one on already good images. You can just upload your own image instead of using stage 1.
Fwow this is super important tips
Fi am going to add to the post
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dah that makes sense, i wondered why there was an option to upload. and yes Ive also noticed the smoothing out of heavily pixaled low res images on stage 1 
nice one Death...never thought id be typing that
nice one Death...never thought id be typing thatAyep and if you do that on good images you smooth out details like jewelry or hand poses and SD likes to do some whacky stuff there
Athats where you see these errors happening
Abut you always need to do this case by case, if the image is too pixelated, it will not gain good quality without stage 1
Dyeh kinda like wiping the slate clean so sd has to work harder to recreate the details
Aexactly
Aim kinda wondering how much it will improve with SD3 in the future
Ai really have high hopes
DI just had a go at the latest Ideogram I am impressed with the text and prompt following
Ayep, and hand poses also seem better, because thats the most annoying thing on every diffusion model xD
D
Ait would be great if the hand would look like this! atleast every hand has 5 fingers
Abut now take that and remove some fngers, meld them into eachother and thats the output
DI wonder if SD3 will be on par with the text / image prompting of Ideogram, maybe their release will speed up SD3 release...
Aim not sure.. is ideogram its own diffusion model?
Fgiven the announcement page had 15 billion instances of AI safety, i'm thinking SD3 is going to be heavily crippled for anything more complex than a person standing there
Dyeh unfortunatly thats the way things seem to be going, the more this tech improves at such a rate the more people in high places are taking notice
Dideogram seems to be (at the moment) not that heavily censored
ASDXL is the same, if promp following and the architecture is good we can let it learn
Ftraining on SDXL was cripped badly
Fi get an error if i try to do stage 2 upscale in supir without doing stage 1 first
Ayou need to upload to stage 1 image
Ayour base image
Fkk
Finteresting, even with an upscale value of 1 (and default settings otherwise/no prompt) putting the image in stage 2 and running stage 2 sharpens and cleans the image up a bit
Ayea
Aand if you smooth it out first, the gun would start looking strange
Fso looks like you can use it to denoise and add detail at same res
Dyeh the metal pops , little extra detail nice one
Fbe canceling my magnific subscription soon then lol
Ai was never realy a fan of magnific. It kept adding details i didnt want haha
Fyes but sometimes thats a good thing, it can repair faces in SDXL renders
Fand for BG characters
Fso the thing you dont like is the only reason to still use it lol
Ayea, but i think if you tamper with the settings of supir, you should be able to do the same
Fye it works better in some images depends on taste
Ai hope supir gets faster in further development. Its currently really slow :/ even on a A6000
Atakes around 10 mins to upscale to 3000x3000
FAdmiral Furkan, do any of your videos deal with multi concept checkpoint training and how to tag those images and whether regularisation images are needed for concepts rather than people