so Dreamboothing with SDXL requires 1024x1024 instance images? Or would it work well with any lower
so Dreamboothing with SDXL requires 1024x1024 instance images? Or would it work well with any lower resolutions?
Fi think 1024x1024
Ffor lower i would use sd 1.5
Fi am going to sleep hopefully see you tomorrow
Fdoing another training
F
KWow great news thank for sharing Dr.Furkan.
"Lora, hypernets, textual inversion should not work - they need to created specifically for SDXL" >> So we must gather 1024x1024 images for new training? Last mothh I tried 768x768 and Kohya gets CUDA out of memory sometimes so I'm not sure if Kohya_ss can handle this version better.
"Lora, hypernets, textual inversion should not work - they need to created specifically for SDXL" >> So we must gather 1024x1024 images for new training? Last mothh I tried 768x768 and Kohya gets CUDA out of memory sometimes so I'm not sure if Kohya_ss can handle this version better.
K@ashleyk Have you tried this tool Ashleyk? I noticed that you have installed Roop in your template, and it's a great tool for realistic models. I think this tool might also be helpful; it's almost similar to Roop but could potentially work with cartoon, 3D, and anime models too. I don't know how to install and use it so I can't try it though https://www.reddit.com/r/StableDiffusion/comments/13qculk/new_method_to_train_a_character_with_one_image/
reddit
286 votes and 91 comments so far on Reddit
XThis would be very helpful!
Ccan i make a lora/embedding of a specific object but it has to be same.
For example, I want to create a chair object and I want to assign it to a specific code or prompt. Is it possible? to be more clear, let's say I wrote "ikea2" on prompt line and will i get the same image of the chair that I have already created(lora/embedding) on the output? but it has to be EXACT image.
For example, I want to create a chair object and I want to assign it to a specific code or prompt. Is it possible? to be more clear, let's say I wrote "ikea2" on prompt line and will i get the same image of the chair that I have already created(lora/embedding) on the output? but it has to be EXACT image.
C@Dr. Furkan Gözükara
Fye new loras will be trained
Fcurrently my settings requiring 24 gb gpu even with lora
Fye message me
Fyes possible
Fespecially with sdxl
Fwait my tutorial
CI'll be waiting for you, thanks for the quick return
W@Dr. Furkan Gözükarasteps: 10%|██████▏ | 2247/23040 [1:01:54<9:32:48, 1.65s/it, loss=0.016]
When I trained Lora based on SDXL using kohya_ss, it was very slow.
I used Prodigy. How about yours?
When I trained Lora based on SDXL using kohya_ss, it was very slow.
I used Prodigy. How about yours?
Fhere my slow as yours
F
Fi am using adafactor
W
WI saw the screenshots of you on GitHub. Why are you so fast?
Wsorry,I made a mistake, It's not that fast after all.
W@Dr. Furkan Gözükara, You can take a look at this
https://github.com/bmaltais/kohya_ss/blob/master/sdxl_minimal_inference.py
https://github.com/bmaltais/kohya_ss/blob/master/sdxl_minimal_inference.py
GitHub
Contribute to bmaltais/kohya_ss development by creating an account on GitHub.
Fi use comfy ui atm for inference with kohya
Fby the way i use batch size 1 and gradient accu 1 since that is best learning for a subject
DFinally!!
Thanks for the video @Dr. Furkan Gözükara!!
You always come in at the Knick of time
Thanks for the video @Dr. Furkan Gözükara!!
You always come in at the Knick of time
Thi everyone, does anyone know how to solve the issue:
Launching launch.py...
################################################################
Using TCMalloc: libtcmalloc_minimal.so.4
Python 3.10.6 (main, May 29 2023, 11:10:38) [GCC 11.3.0]
Version: v1.4.0
Commit hash: 394ffa7b0a7fff3ec484bcd084e673a8b301ccc8
Launching Web UI with arguments: -f --port 3000 --xformers --skip-install --listen --enable-insecure-extension-access
No module 'xformers'. Proceeding without it.
*** Cannot import xformers
Traceback (most recent call last):
File "/workspace/stable-diffusion-webui/modules/sd_hijack_optimizations.py", line 138, in <module>
import xformers.ops
ModuleNotFoundError: No module named 'xformers'
Launching launch.py...
################################################################
Using TCMalloc: libtcmalloc_minimal.so.4
Python 3.10.6 (main, May 29 2023, 11:10:38) [GCC 11.3.0]
Version: v1.4.0
Commit hash: 394ffa7b0a7fff3ec484bcd084e673a8b301ccc8
Launching Web UI with arguments: -f --port 3000 --xformers --skip-install --listen --enable-insecure-extension-access
No module 'xformers'. Proceeding without it.
*** Cannot import xformers
Traceback (most recent call last):
File "/workspace/stable-diffusion-webui/modules/sd_hijack_optimizations.py", line 138, in <module>
import xformers.ops
ModuleNotFoundError: No module named 'xformers'
Ti was trying to launch Web UI in Runpod, but it shows above issue
AInstall xfomers
AObviously activate your venv before doing so.
Tthanks! let me try
Fi am working on comfy ui and sdxl lora training too
Xok just did 
VHola a todos, hello world !
Xhello
Xman i wonder what could be making dreamshaper take so long to download 
R@ashleyk hey man, are we going to see an update of your runpod template to include Sdxl for Kohya and webui? Or waiting that the shit normalizes?
FHello welcome
FFrom where do you download?
FYou can use my auto installer of kohya
FWorking with sdxl too
FI also have auto sdxl download link posted on patreon
XYouTubeSECourses
I show how to install Automatic1111 Web UI & ControlNet extension installation from scratch in this video. Moreover I show how to make amazing QR codes and inpainting and out painting of ControlNet which are very similar to Photoshop generative fill and Midjourney zoom out. Furthermore, I explain and show what are Canny, Depth, Normal, OpenPose...
Jhi guys, when sdxl 1.0 coming on a1111 ?
RYea I saw. Will test tomorrow. Traveling again with laptop 
AThere is an
sdxl branch if you want to try it out,Jnice ! gonna try it
AGitHub
Description
uses Stability-AI's repo from https://github.com/Stability-AI/generative-models
retains old repo for SD1.x models
the biggest change is that for SD XL, conditioning is now a dictio...
uses Stability-AI's repo from https://github.com/Stability-AI/generative-models
retains old repo for SD1.x models
the biggest change is that for SD XL, conditioning is now a dictio...
