32:03 How to inpaint generated image to improve quality and similarity
32:03 How to inpaint generated image to improve quality and similarity
Fit depends on VRAM fitting
Fif both fits VRAM
Fi think could be same speed
S12gb on 4070 or 24 on 3090
Fye 4070 will never catch 3090
F12gb simply not enough for SDXL
FRTX 4090 is simply faster like 26%
Ffor training,
SGot it. Just bought a 4070 ti new but may see if I can return it and find a used 3090
F3090 all the way better
Fi plan to search this for kohya gui but didnt have time yet
BThank you !
BYeah you want as much VRAM as possible.
Ffinally exporting
F
DDo we need to train face Lora for sd1. 5 to be used with animatediff? Or can we convert sdxl Lora to sd? Is it possible?
Or go for sdxl animatediff?
Or go for sdxl animatediff?
DAlso can 3080 10gb run animatediff?
S@Dr. Furkan Gözükara I saw on one of your linkedin posts that you mentioned you have a pretty decent set of "medium quality" settings for sdxl dreambooth. Have you published those settings?
TTWe really need @Dr. Furkan Gözükara to do a tuto on that
W@Dr. Furkan Gözükara i am currently going through your udemy course, and i got intel i9 13900k , rtx 4090 gaming oc, think i should be good with stable diffusion?
Ahaha absolutely, you have a combo which is a dream for many 
Wyeah, i bought it for ai/ml not for gaming, though i do sometimes, but i don't get much time, i just wanna get into stable diffusion very beginner as of now
Whttps://i.imgur.com/3ES9Pnm.png can someone help me with this error, tried generating an image with fresh sd install
Wis anybody here?
Cyeap
Cshow your cmd screen error , this ui error sir
TGitHub
When I try to use txt2image the first image is generated normally but when I try to generate the next one it shows "RuntimeError: CUDA out of memory. Tried to allocate 58.00 MiB (GPU 0; 4.00 G...
TSome solutions given there
Wit was multigpu error
Cı think you must change your settings or select your device
Wyes, that's what i did, sdxl was not able to figure out which gpu to use, i disabled my i9 integrated graphics
CIt's not a problem I've encountered before, but you can solve it like this. check out settings
Wand the error went away
Cor edit webui-user.bat
CAdd "SET CUDA_VISIBLE_DEVICES=1"
CGitHub
Is there an existing issue for this? I have searched the existing issues and checked the recent builds/commits What happened? In a dual-GPU Windows 10 system (2 x RTX-3090), only the first GPU is b...
Woh, wow thanks, that one is easier, i'll try that
Cvisit this
Bdoes anyone knows how to fix these errors?
Wthanks, man. this is exactly what i was looking for
Wnot able to install / upgrade xformers now
TWhich version of CUDNN did you install? It wont work with CUDA 12 you have to use CUDA 11.8 versions
Wi think its a read issue. ugh what to do
BThank you, I'm using Torch 2.0.1+cu118
Z@Dr. Furkan Gözükara INstalling manually on my linux machine, where do we put the models at? I don't see a models folder for pixart?
Zhow many gigs of vram does pixart use? everytime i run app.py, it says out of vram
MI just trained with a very interesting set of images. Even though I selected the best photos, I could only find the best d_coef value for a quarter of them. I tried 0.85, it didn't come out strong. With 1.02 it was too strong, and even with 0.95. Finally, with 0.9 I managed to get a near perfect Lora. On another set of images, I managed to get one that turned out perfect the first time, even though there was nothing special about the photos. So there are still some exceptions where, even if the graph is good, you still need to polish the final result a bit.
