@Dr. Furkan Gözükara seems to be broken, when ever i click apply stage 1 before stage 2, the final i
@Dr. Furkan Gözükara seems to be broken, when ever i click apply stage 1 before stage 2, the final image is the blurry one
Rfound an a100
Rgonna test it
Fye stage 1 useless
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dwait so we only use stage 2 now?
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fwe were always
Dso stage 2 becomes stage 1
Kalso maybe a button to stop the batch upscaling in newer version?
Fye
Flet me see if possible
Rdid you some testing between jugger9 and sdxl base?
Kyep i sent you a dm
R@Dr. Furkan Gözükara i can confirm on runpod, a100 takes 70s for a 2x upscale compared to 160s for 2x on 4090
Fusing optimizations?
AHow can we use adetailer with transparent backgrounds
Roptimisation on 4090
Rnone on a100
Fthat makes diff
Ron my local 4090 though it only takes 120s
Fyou cant use yet. SD dont have transparent feature
Rand on @KG Pickles it takes 60s
Rso it doesnt make sense why runpod is so much slower than local
Aisnt forge sd?
Fyes SD
Fsd means stable diffusion
Frunpod cpu and power probably
Alayer diffusion allows transparent backgrounds in forge, so forge doesn't support adetailer?
Bdid you try out the newest a1111 version yet @Dr. Furkan Gözükara ?
U@Dr. Furkan Gözükara is it possible for you to update Stable Cascade script so that it saves metadata, then we can use it on SUPIR?
Ryou mean their secure cloud throttles on power and have inferior cpu's which makes it 50% slower than real world local machine?
R@Dr. Furkan Gözükara is it worth adding a save preset toggle down option?
Ri am down to 60s on my 4090
Rwith optimisations
VSUPIR: If I get the message in the console on Windows: Done in 18.092s, max VRAM alloc 10111.929 MB -- then the generation should be done and the file is saved ? -The UI still shows "processing" for the last 10 minutes and there is no file in the output folder.
V( I try to upscale from 2048 to 4320 (height) with 16 GB VRAM.. )
BJust FYI, I upscale a pano 2048x1024 to 4096x2048 (2x) and it was eating over 20 gigs of Vram. That's the limit my 3090 can do. My guess is you are OOM
VGenerated image 1/1 in 111.18 seconds -- that is the message when it worked. And yes.. it seems to run out of memory. I have only 50 MB left.
VBut 4K works amazing! 
-Wow.. the sharpening abilities of SUPIR are insane!
-Wow.. the sharpening abilities of SUPIR are insane!
S@Dr. Furkan Gözükara Hi, I know that you explain on a video how create subtitles in different languages can you share that tutorial and the patreon link? is this accurate?
LHey peoples. Lately (last week or so) have been having issue (perhaps I just noticed it) where Kohya creates endless folders (on Windows 11) the img folder is repeated endlessly and makes it impossible to delete via windows explorer I have to resort to change permissions and remove using rmdir /s on admin command prompt. Anyone with this issue past or present?
Ehey! quick question if anyone knows. i updated xformers to 0.0.23.post1 and torch to 2.1.2 , but my torchvision is incompatible. i cant find the right torchvision that would be compatible with both xformers and torch
Eheres my error: ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
torchvision 0.16.0 requires torch==2.1.0, but you have torch 2.1.2 which is incompatible. but in github it says torchvision 0.16.0 is compatible with torch 2.1.2 - im confusion
torchvision 0.16.0 requires torch==2.1.0, but you have torch 2.1.2 which is incompatible. but in github it says torchvision 0.16.0 is compatible with torch 2.1.2 - im confusion
Fnot yet
Fit is sd
Fyes i will add others as well
Fstable cascade and instant id too
Fthat is only a certain part
Fyes i plan to make this
Fbut dont have a tutorial yet
