if all training pics are 2048x2048, should I keep them that way, or reduce to 1024x1024?
if all training pics are 2048x2048, should I keep them that way, or reduce to 1024x1024?
οΏ½Use tier 1 when using a6000 for max quality
MAs far as I know Tier 1 only adds Train T5-XXL
Myeah it says it in the Configs Explanation Must Read File. Although I probably should have selected the 48gb config file not the 24gb
YYES
Y
Y@TheToday all these are flux lora
Y
YGuys am launching a backend kohya trainer + inference that outputs these images ! Anyone would like to test it for free ?
Y
YI now face an issue with jewelery and accesories they are not high quality do you guys know any way to optimize them ?
Y
Y
Tyes I want
Y
DTbh, jewelery works best with direction. Like img to img. Without it, the model is just lost and breaks things here and there.
DMaybe use a jewellery lora 
Adoesn't make much of a difference Tier 3 better and cheapest
οΏ½thank you, luckily I did that before I started the training. Would'nt wanna change what works, but if I did 2048, it would still work, but maybe take longer time?
οΏ½would the quality be any different though, is what I wonder
SThis is beyond impressive.
SHow did your config look, and what's your dataset looking like?
Y20 images 1024x1024, lora config
YI edited some parameters in config
Yand tweak generation in comfyUI
DHi, guys!
Quick question: I just noticed that I've been saving all my images as .jpg in Swarm.
Does saving your image as .jpg or .png make a difference in the quality of generations?
I guess, there is an option to save as .png in Swarm, but is there a quality difference?
Quick question: I just noticed that I've been saving all my images as .jpg in Swarm.
Does saving your image as .jpg or .png make a difference in the quality of generations?
I guess, there is an option to save as .png in Swarm, but is there a quality difference?
Fthis is superb quality
Fi agree but so many stuff. after swarmui flux tools i can focus on that i guess
Fyou dont do block swap on a6000
Fit fits into vram that is why :d
Falmost no diff
Fbut png losless
Fjpg has some
DI'll start saving them as .png from now on. Any loss is a loss
DI agree. @yassinemenace 's lora are perhaps the best I've seen lately. Almost dreambooth level, if not better.
YThank you guys
YI've been lately
Ytrying multi sampler too
Y0-25 euler/beta
Y25-45 ipnb/beta
Y
Y
DWhat format do you save your images?
YPNG
A
AMy Lora results
APure stuff nothing additional
ADoctors training concept is a blessing
A100x times better than anything else online atm
Aand I've tried other training methods too