but with my LR as you do more training, you should eventually get your perfect checkpoint
but with my LR as you do more training, you should eventually get your perfect checkpoint
AYes, I eventually got a good result.
WCan you describe more out which step to do it. Where I can load the latest checkpoint and how to calculate the remain step for the latest checkpoint to continue
DI've seen Flux not really converge on realism when the LR is too low, no matter how long you run it. It changes facial features but remains in its plastic AI style with a low LR
Dare the e-06 values above with rank 128 alpha 16?
Fit wont have those
Fyou need to memorize and do yourself
Fdont change alpha we have it affects LR
DI have it at 1 but different LRs, that's why I'm asking. Alpha 16 multiplies the LR by 16 compared to 1
Frank 16 + alpha 16 you trying?
Fyou should make them equal
For your LR need to be researched again
Drank 128 alpha 1
Dthey don't have to be equal
Dalpha is there so you can change the rank without changing the LR
Fthen you need new LR
Dso whatever alpha you use, don't change it. but it doesn't matter what alpha you use.
Factually we make alpha equal to rank
Fso that you can use same LR
Dit's the other way around. if you use alpha 16, you can change the rank without changing the LR. OT (and I think kohya) automatically scale the LR by (alpha / rank)
Fi recently researched this i am 100% sure
Djust keep alpha the same
Fi even talked this with kohya
Falpha is just a multiplier that is used to balance LR for different lora ranks
Dexactly. we mean the same thing 
Ffor example when i moved to single layer, it required 10x more LR
Dkeep your alpha unchanged, keep your LR unchanged, change the rank. it'll work, because LR is automatically scaled.
Fnope opposite
D
D
Dif you have a LR that works at rank 16 alpha 16, and then change to rank 64 alpha 64, it'll fail.
Dif you change to rank 64 alpha 16 it'll still work
Dbecause it automatically lowers your LR for you. rank 64 needs a lower LR internally than rank 16
Dthis is OT. I hope kohya works the same. that would be very confusing otherwise
Fi will test this because i am pretty sure lots of inaccurate info there are
lots of inaccurate info there are
DI trained my dog model. However, I don't know what to put in the prompt to get it to inpaint or generate the photo of my dog.
Ftell me your folder name
Ftraining images
Djuny_photos_resized
Dthis is what i have tried so far. but my dog is a little black dog.
Finside it?
Fwhat is folder name
Dhome/downloads/resize/juny_photos_resized/1024x1024
Fthis wouldnt work
Fkohya has special folder structure
Flook your kohya training config
Dyou mean: downloads/Kohya GUI Flux/model/Quality_1_LittleJune(serial number).json?
Fopen that file
Fand look which training folder you used
Fenter that folder