Open JunhaoWang opened 7 months ago
It took me some time but I managed to install everything in Linux/Debian, I was successful in creating a LoRA for dreamshaper (nice work, very simple process), not for juggernaut (SDXL), from the messages it seemed to me for the same reason as this issue. Unfortunately I didn't think about saving the logs, I'll do it again tonight, I'll save the logs and then attach them here.
Do I need to use any specific resolution to train for SDXL? 1024x1024? After seeing the error with juggernaut I had scaled all the images to 512x512 and it worked with dreamshaper_8, but I think it wasn't necessary, they should be scaled automatically right?
Just came across this repo, so I haven't yet tried it. However, SDXL is typically based on 1024x1024 images. I've seen references to 768x768, but most everything references 1024x1024. I know when I made my first SDXL checkpoint, it required 1024x1024 images
@RikkB thanks, I'm not an expert but yes I remember the same, 1024 for SDXL, but I've seen tutorials for both kohya (never used) and this node for which it is not essential to resize the images, useful for performance but not essential, they should be resized automatically during the process, however, just in case, I *resized all the images to 1024x1024, same error. Anyone having success with SDXL?
Hi, has anyone got this working on sdxl? Kindly confirm.
Will try some different size images as sources if that helps… Anyone else tried anything to get this working for SDXL?
Just verified, doesn't work with SDXL due to size mismatch