Trying to create a fine tuned model of an AI person. Anyone have data augmentation tips so I can tra
Trying to create a fine tuned model of an AI person. Anyone have data augmentation tips so I can train on super realistic data? dm me
F2 stage training
F1 st crop different distances
Ftraing
Fgenerate
Fcollect
Ftrain again
Fyou can also use live portrait to make different expressions
Mshould I enable "Apply T5 Attention Mask" on the right?
M
Ait wont make much of a difference in terms of quality
Abut you will lose 1-2~it/s
Moh okay
Aif you want the best of the best go for it but i havent seen a difference
Ayou need to stand with a magnicfying glass hours looking at pixels i guess
Mhaha okay, now that you've mentioned that, I might as well go for the "best of the best" then
Mthank you
Ayeah the config is spit in Q1,Q2,Q3
Ayou decide
Fit makes minimal impact but you can apply for max quality
�Trained sdxl lora and sdxl fine tune has many artifacts like extra arms and disembodied limbs. If lora I have to use it on low weight, like 0.4, higher reduces image quality. Trained on cyberreal xl and big love for lora. Used onetrainer for fine tune, body was not learned properly. I wonder if using regularisation images caused my models body to be not learned so well. Used cicitai for lora train. You guys have any tips?
Mwhat could be the reason, I was in the middle of training a fine tuned model....
MMalso this
M...
Moh nvm I just relogged in hah
Fbase model could be reason too
Fby the way i dont have good sdxl lora config right now
Fi recommend dreambooth
Fcould be out of RAM
Fok i am glad it wasnt out of RAM 
Myes me too haha, thank you regardless
JCan someone help me diagnose this issue. Images from my recent fine tuned model are coming out with strange patterns particularly noticable on skin but also just overall in the image. Is it due to over training or could it be from particular grainy images in dataset?
Jis like specky-ness everywhere. I think it only happens after refining.
JI think it might be due to the lack of distant photos I trained with... as I used mostly high detailed mostly selfies.
MI am also curious why this is happening
�if your dataset has grainy images than it's because of that. If not then your learning rate is high. If your images are grainy try reducing the lora rank and re train
JI will try that thank you
Fcould be dataset
Mmay I ask why it says that the total amount of epochs will be 195 by the end of the training, I am sure that I selected 200
Mthis happened yesterday too when I trained another fine-tuned model, I selected 200 epochs but it did 181 I think if I remember correctly
Fbatch size number of images
Fdont worry
MOkay then
DHi @Furkan Gözükara SECourses, I cannot get either SDPA nor xformers to work with koyha_ss on my 5090. I have used both your install and also another method recommended by Bmaltais himself 2 weeks ago (https://github.com/bmaltais/kohya_ss/issues/3096#issuecomment-2730229439).
The installation of pytorch + xformers is successful. When I check in the .venv, the xformers are indeed installed.
However during training, the CrossAttention settings simply get ignored, both if I set it to xformers and to SDPA.
No mention of CrossAttention is shown in the training logs, and I notice on the VRAM usage that this is not being enabled.
I have all the requirements on my pc (python 3.10, cuda 12.8 etc).
Do you have any advice? Are the cross attention settings working in your koyha_ss install with your 5090, xformers in particular?
Thank you in advance.
The installation of pytorch + xformers is successful. When I check in the .venv, the xformers are indeed installed.
However during training, the CrossAttention settings simply get ignored, both if I set it to xformers and to SDPA.
No mention of CrossAttention is shown in the training logs, and I notice on the VRAM usage that this is not being enabled.
I have all the requirements on my pc (python 3.10, cuda 12.8 etc).
Do you have any advice? Are the cross attention settings working in your koyha_ss install with your 5090, xformers in particular?
Thank you in advance.
GitHub
Currently when I start kohya ss gui on RTX5090, I get the following error and cannot learn Lora. Maybe I should wait for an update, but is there any way to learn Lora using RTX5090? UserWarning: NV...
Fdo you have python 3.10.11 installed?
Fi am able to use both with my installer
Fmaybe not displaying on cmd?
Fdid you check your speed_? which config you used?
DThank you for your reply. Yes I have python 3.10.11, I am using the latest config "48GB_GPU_28200MB_6.3_second_it_Tier_1".
Speed is around 3.85s/it.
Perhaps it is indeed working but I find it strange there is no menion of this in the cmd windows. I remember the use of xformers was shown here on older installs (when using a 4090).
Do you see this in your koyha cmd windows when training?
Speed is around 3.85s/it.
Perhaps it is indeed working but I find it strange there is no menion of this in the cmd windows. I remember the use of xformers was shown here on older installs (when using a 4090).
Do you see this in your koyha cmd windows when training?
DFurther more, I am getting the exact same 3.85s/it speed and 27.7 GB VRAM used when I set CrossAttention to "none" in tge Koyha GUI, as when I set it to "xformers".
So that shows that there is no cross attention happening.
So that shows that there is no cross attention happening.