looks like individual block training might have been implemented in the sd3 flux1 kohya ss branch la
looks like individual block training might have been implemented in the sd3 flux1 kohya ss branch last night
XSo fine tuning is better the Lora over all you think @Dr. Furkan Gözükara ?
D
Fi am testing already
XYou have a video on how to fine tune training?
RCan someone confirm Flux Lora Extraction in completely broken on Kohya ?
RI did a finetune of flux dev, results are amazing. I use Extract Flux LoRA tab to extract the Lora from the finetuned model. When I use the extracted Lora it makes no difference with not using Lora. @Dr. Furkan Gözükara Did you try the Extract Flux LoRA feature ?
XI’m confused on what fine tuning is, do we basically do the same training as in this video https://youtu.be/nySGu12Y05k?feature=shared but with the configuration from the new update on your patreon?
YouTubeSECourses
Ultimate Kohya GUI FLUX LoRA training tutorial. This tutorial is product of non-stop 9 days research and training. I have trained over 73 FLUX LoRA models and analyzed all to prepare this tutorial video. The research still going on and hopefully the results will be significantly improved and latest configs and findings will be shared. Please wat...
NI'd also like to know. Though I was waiting on Kohya to refine it a bit more before delving into fine tunes.
SSame here. Do you mean extracting a lora from a finetuned checkpoint?
NYeah
SFor me the extracted lora looks nothing like my subject, nothing like the checkpoint
SI guess it is broken
R@smoke yeah, thx for the confirmation
SShould probably raise an issue in sd-scripts Github since it is broken
RGitHub
2 Issues sometimes it works, but it looks nothing like the fine tuned model. Doesnt work and i get the error below During handling of the above exception, another exception occurred: Traceback (mos...
SAh yeah it definitely is broken
RBut i'm not sure it's the same issue
RBecause they say it works at 1.5 strength
RBut for me at 2.0 there's was no difference
SWell for me it is, the extracted lora does not look like my finetuned checkpoint. At 1.5 strength it starts to look more like it but it created a lot of other issues and graininess, so definitely not the best approach
SAtleast for me
RComfy ?
SI tried in Comfy yeah but I believe the same issue happened in forge
Dhas anyone got the only training specific layers yet to see a speed increase? im getting same speed trying to training only 2 layers as i do when i train all
Xwhat are layers?
Fexactly same
Fjust use new config
Fcheckpoints will be 23.8 gb
Fload config into dreambooth tab
Fi havent could be broken
Fif broken should be reported
Fi will try extract now
Fok i think i got the best single layer found 
Ftime to deepen research
Fit sucks that fused back pass not works with multi gpu
F48 gb not sufficient to do multi gpu training with flux fine tuning
MI'm not ashamed I listen this...
F@master_kawaii can you private message me
MOh am I going to jail? @Dr. Furkan Gözükara haha
Fno something else
Moh ok xD
MThere ! Master Furkan
Fi am testing lora extract right now out of flux fine tune
F
Fextracting right now
Ftesting result
Fye quality of lora reduced a lot
Fdoing another test