right now i am trying to make kaggle notebook of SUPIR
right now i am trying to make kaggle notebook of SUPIR
DNice
Fok it uses too much ram
Fso crashes notebook
Pi read this yesterday! the things move fast... haha
Fi think i found a good way to improve model loading of SUPIR
Fit was using 30 gb ram
Fnope
Vsorry it was a clickbait video..
Fyep
FGitHub
Currently create_SUPIR_model inside util.py loads everything into RAM first with full precision Thus it is using more than 29 GB RAM I think model half param can be given to it as well Or perhaps s...
Pi think you can use fp8
Pits use same stable diffusion engine that a1111
Pi am busy with work jobs i cant try it yet
Pyou can send that A1111 PR to they or we can be the first to do this 
Pi think they will be very lazy to implement these optimizations
PDr Furkan, good day. Can you explain to me, in a dummy way, how can I edit the BAT file to have 2 folders with models in A1111 (original and another one) and 2 folders with LORAs (original and another one)?
Fprobably :d
Fcan you explain me more what you wan to do? i can show you
PI don't have a huge drive so I will divide my LORAs and my CHECKPOINTs in two diferent folders, one folder for LORAs and CHECKPOINTs in the drive D and other folder with other LORAs and CHECKPOINTs in the drive E. I want to have two separeted folders in two different drivers with the same installation of A1111. For this reason I want to edit the BAT file but I don't know how to.
Flet me show
Fchange your paths accordingly
Pand this way it will keep the original folder plus this one?
Fi think it only uses 1 folder
Fmultiple folders would be good i agree
PThanks
F@Dr. Furkan Gözükara Do i still need to downgrade to v21.8.4 of Kohya SS when training dreambooth on SD 1.5? Found from your patreon: https://www.patreon.com/posts/best-stable-1-5-90307694. I've been using your settings for SDXL dreambooth, but found that there did not exists any openpose for SDXL and i need to control the pictures generated.. Will try with SD 1.5
Patreon
Patreon is empowering a new generation of creators.
Support and engage with artists and creators as they live out their passions!
Support and engage with artists and creators as they live out their passions!
Fi think they fixed
Falso our latest SD 1.5 config works on latest version
Pi made a clean install of KOhya and now i dont get error when LORA extract. Git pulling oldd version didnt help. Only clean reinstall for ome reason
Fnice at least solved
FWas that the config in the Patreon post that I linked?
NModel called epiCPhotoGasm last version is pretty good
Bshould all stage1 references be remove from the GUI?
FWe may remove not useful
FPatreon
Patreon is empowering a new generation of creators.
Support and engage with artists and creators as they live out their passions!
Support and engage with artists and creators as they live out their passions!
FNice can be tested
A@Dr. Furkan Gözükara is there any way to rollback to supir v14 ?
Aits taking twice as long for me on the latest version and worse results
Rhi @Dr. Furkan Gözükara what's this gradio error?
Ralso no module called requests.
Ri installed your v19 and threw in the models
BI don't have v1.4 installed, but v1.7 quality is better than v1.9
Fthere shouldnt be this stuff
Fare you aware not it default selects Linear CFG?
Fi can turn it off
Flooks like same folder name repeat error
Finstall into different folde rname
Fcan you test turning off Linear CFG and compare?
