for example lora rank 16 = much lower quality compared to 128
for example lora rank 16 = much lower quality compared to 128
Fi compared
VI'm training again on 128
Awhat quality specifically dr?
Vdid it and now all I get are big headed photos 
D@Lecturer - Main Admin I wonder if you've ever used modal.com? How difficult is it to run kohya_ss there?
XXSomething that I see has helped in prompts that include multiple people is changing what the people are wearing their age and hair color
MI'm still waiting with Flux for several reasons. For one thing, what I see from current loras is that low weight/alpha is enough for posing, so some people use 8s. On the other hand, human anatomy is horrible, so not trained on nsfw images, so it will be a lot more work to train all body parts, and I'd rather wait for that until more sophisticated flux models are available. As I see it, it will be at least 2-3 months. The Pony is also being made for Auraflow first (which is also because of the license), so the transition to flux is a long way off for me.
WDoes fp8 vs fp16 lora quality diffrerent much ?
Fquality of training
Fa little bit
MFor sdxl training on runpod, restarting the pod or pkill -f "relauncher.py" isn't ending the gpu memory usage for me
MDoes anyone else have this problem?
Mbut if I only want to train and not use A1111 in kohya I don't have to choose Stable Diffusion template and upload the relauncher.py, right?
TIf my current image size is 16:9 and the resolution is 1024X576, is this how the size should be filled in on OneTrainer? @Dr. Furkan Gözükara
Hif I want to run Flux dev on a cloud with Forge webui, which site do you recommend? I've tried runpod but their Forge is not updated.
Lhttps://github.com/rickrender/FluxLatentUpscaler
hey i saw this, and i trained a lora model with my dataset, can i use this upscaler in the swarmui? because the results looks very good, or it's something not relatable and its his own model
hey i saw this, and i trained a lora model with my dataset, can i use this upscaler in the swarmui? because the results looks very good, or it's something not relatable and its his own model
GitHub
This Flux latent upscaler workflow creates a lower-resolution initial pass, then advances to a second pass that upscales in latent space to twice the original size. Latent space manipulations in th...
FI overwrite it and then do fuser -k 3000/tcp
FUse our kohya installer on torch 2.1 template
FI would ask this to them I don't know how exactly they handle
FMassed compute
FImage updated work out of box
FSelect our image SECourses use our coupon SECourses for 31 cents per hour on a6000
FNothing is as good as supir have you tested it?
FBut it is seperate app sadly not in swarmui
FMy strategy is select images and upscale as batch
SDr Furkan i hope everything is ok, im wondering now that you have been making test with OneTrainer who do you thinbk could be more fast to make model in local between kohya and OneTrainer?
Wdear @Dr. Furkan Gözükara what does the "segment prompt" meaning. I see in your test prompt:
photo of ohwx man a red shiny steel armor with scale patterns like a dragon scales in an epic battlefield as the leader of the army and holding a mighty gigantic diamond sword , face open looking at the camera on a bright sunny day, epic scene, cinematic shot<segment:face,0.7>photo of ohwx man
photo of ohwx man a red shiny steel armor with scale patterns like a dragon scales in an epic battlefield as the leader of the army and holding a mighty gigantic diamond sword , face open looking at the camera on a bright sunny day, epic scene, cinematic shot<segment:face,0.7>photo of ohwx man
Wis that we ask Comfy or Forge to do the face detailer without setting node or run addetailer extension
E@Dr. Furkan Gözükara I'm training a character Lora. Should I use non-default captioning for it? Would it improve the result?
EHave you figured out a way to make SUPIR faster? its only downside is the speed.
FOneTrainer first need to fix FP8 issue
Fcurrently useable only for 40 GB GPUs
Fit is automatically masking that area
Fand inpainting with your following prompt
F70% denoise
Flike adetailer but more robust
Fi found captioning only useful if you have a non consistent style dataset
Fwhich makes sense
Falso
Fnot yet :/
EIs it possible?
Fperhaps
EIt would be great if you could! 
Fwe have a work it may come
Fas you train more it makes the images much more realistic
F
Fi will try to continue training first time lets see what happens with kohya
Flora